diff --git a/doc/source/_static/bib/references.bib b/doc/source/_static/bib/references.bib
index 12f378d7..913ca687 100644
--- a/doc/source/_static/bib/references.bib
+++ b/doc/source/_static/bib/references.bib
@@ -1295,5 +1295,34 @@ @book{monnier2021coursevariational
   url = {https://www.math.univ-toulouse.fr/~jmonnie/Enseignement/CourseVDA.pdf}
 }
 
+@article{Astagneau_2022,
+author = {Paul C. Astagneau and François Bourgin and Vazken Andréassian and Charles Perrin},
+title = {Catchment response to intense rainfall: Evaluating modelling hypotheses},
+journal = {Hydrological Processes},
+volume = {26},
+number = {e14676},
+year  = {2022},
+doi = {10.1080/02626667.2021.1923720},
+
+URL = { 
+        https://doi.org/10.1002/hyp.14676
+    
+},
+eprint = { 
+        https://doi.org/10.1002/hyp.14676
+}
+}
+
+@Article{hess-22-5317-2018,
+AUTHOR = {Douinot, A. and Roux, H. and Garambois, P.-A. and Dartus, D.},
+TITLE = {Using a multi-hypothesis framework to improve the understanding of flow dynamics during flash floods},
+JOURNAL = {Hydrology and Earth System Sciences},
+VOLUME = {22},
+YEAR = {2018},
+NUMBER = {10},
+PAGES = {5317--5340},
+URL = {https://hess.copernicus.org/articles/22/5317/2018/},
+DOI = {10.5194/hess-22-5317-2018}
+}
 
 @Comment{jabref-meta: databaseType:bibtex;}
diff --git a/doc/source/math_num_documentation/forward_structure.rst b/doc/source/math_num_documentation/forward_structure.rst
index 7249ade9..42144591 100644
--- a/doc/source/math_num_documentation/forward_structure.rst
+++ b/doc/source/math_num_documentation/forward_structure.rst
@@ -351,8 +351,120 @@ Hydrological processes can be described at pixel scale in `smash` with one of th
     Same as ``gr4`` transfer, see :ref:`GR4 Transfer <math_num_documentation.forward_structure.hydrological_module.gr4>`
 
 
-.. _math_num_documentation.forward_structure.hydrological_module.gr6:
+.. _math_num_documentation.forward_structure.hydrological_module.gr5_ri:
+
+.. dropdown:: gr5_ri (Génie Rural 5 with rainfall intensity terms)
+    :animate: fade-in-slide-down
+
+    This hydrological module is derived from the model introduced in :cite:p:`Astagneau_2022`.
+
+        
+    **Production**
+    
+
+    In the classical gr production reservoir formulation, the instantaneous production rate is the ratio between the state and the capacity of the reservoir,
+    :math:`\eta = \left( \frac{h_p}{c_p} \right)^2`. The infiltration flux :math:p_s is obtained by temporal integration as follows:
+
+    .. math::
+        :nowrap:
+
+        \begin{eqnarray}
+
+            &p_s = \int_{t-\Delta t}^{t} (1 - \eta) dt \\
+        
+        \end{eqnarray}
+        
+    Assuming the neutralized rainfall :math:p_n constant over the current time step and thanks to analytically integrable function, the infiltration flux into the production reservoir is obtained:
+
+    .. math::
+        :nowrap:
+        
+        \begin{eqnarray}
+
+            &p_s = & c_p \tanh\left(\frac{p_n}{c_p}\right) \frac{1 - \left( \frac{h_p}{c_p} \right)^2}{1 + \frac{h_p}{c_p} \tanh\left( \frac{p_n}{c_p} \right)} \\
+            
+        \end{eqnarray}
+
+    To improve runoff production by a gr reservoir, 
+    even with low production level in dry condition, 
+    in the case of high rainfall intensity, :cite:p:`Astagneau_2022` suggests a modification 
+    of the infiltration rate :math:p_s depending on rainfall intensity :math:p_n:
+    :math:`\eta = \left( 1 - \gamma \right) \left( \frac{h_p}{c_p} \right)^2 + \gamma` with :math:`\gamma = 1 - \exp(-p_n \times \alpha_1)`
+    and :math:`\alpha_1` in :math:`mm` per time unit.
+
+    .. math::
+        :nowrap:
+
+        \begin{eqnarray}
+
+        &p_s& &=& &\int_{t-\Delta t}^{t} (1 - \eta) dt\\
+
+        && &=& &\int_{t-\Delta t}^{t} \left(1 - (1-\gamma) \left(\frac{h_p}{c_p} \right)^2 \right) dt - \int_{t-\Delta t}^{t} \gamma dt\\
+        
+        && &=& &\left[ \frac{ c_p }{ \sqrt{1-\gamma} } \tanh \left( \frac{\sqrt{1-\gamma} \  h_p}{c_p} \right) \right]_{t-\Delta t}^t - \gamma \Delta t
+        
+        \end{eqnarray}
+
+
+    We denote :math:`\lambda := \sqrt{1 - \gamma}`, then
 
+    .. math::
+        :nowrap:
+        
+        \begin{eqnarray}
+
+        \tanh \left( \lambda \frac{h_p + p_n}{c_p} \right) - \tanh\left( \lambda \frac{h_p}{c_p} \right) &=& 
+        \tanh \left( \lambda \frac{p_n}{c_p} \right) \left(1 - \tanh \left( \lambda \frac{h_p + p_n}{c_p} \right) \tanh \left( \lambda \frac{h_p}{c_p} \right) \right) \\
+        &=& \tanh \left( \lambda \frac{p_n}{c_p} \right) \left(1 - \frac{ \tanh \left( \lambda \frac{h_p}{c_p} \right) + \tanh \left( \lambda \frac{p_n}{c_p} \right) } { 1 + \tanh \left( \lambda \frac{h_p}{c_p} \right) \tanh \left( \lambda \frac{p_n}{c_p} \right) } \tanh \left( \lambda \frac{h_p}{c_p} \right) \right) \\
+        &\sim& \tanh \left( \lambda \frac{p_n}{c_p} \right) \left(1 - \frac{ \lambda \frac{h_p}{c_p} + \tanh \left( \lambda \frac{p_n}{c_p} \right) } { 1 + \lambda \frac{h_p}{c_p} \tanh \left( \lambda \frac{p_n}{c_p} \right) }  \lambda \frac{h_p}{c_p} \right) \\
+        &=& \tanh \left( \lambda \frac{p_n}{c_p} \right) \frac{1 - \left( \lambda \frac{h_p}{c_p} \right)^2}{1 + \lambda \frac{h_p}{c_p} \tanh \left( \lambda \frac{p_n}{c_p} \right)}
+        \end{eqnarray}
+        
+    Thus
+
+    .. math::
+        :nowrap:
+        
+        \begin{eqnarray}
+
+        p_s &=& \frac{c_p}{\lambda} \tanh \left( \lambda \frac{p_n}{c_p} \right) \frac{1 - \left( \lambda \frac{h_p}{c_p} \right)^2}{1 + \lambda \frac{h_p}{c_p} \tanh \left( \lambda \frac{p_n}{c_p} \right)} - \gamma \Delta t
+        \end{eqnarray}
+
+
+    .. note::
+
+        Note that if :math:`\alpha_1 = 0`, we return to the general writting of the instantaneous production rate.
+        
+        
+    **Transfer**
+    
+    In context of high rainfall intensities triggering flash flood responses, it is crucial to account for fast dynamics related to surface/hypodermic runoff 
+    and slower responses due to delayed/deeper flows (e.g. :cite:p:`hess-22-5317-2018`). 
+    Following :cite:p:`Astagneau_2022` for a lumped GR model, we introduce at pixel scale in `smash` a function to modify the partitioning between fast 
+    and slower transfert branches depending on rainfall intensity of the current time step only (small pixel size):
+    
+    .. math::
+        :nowrap:
+
+        \begin{eqnarray}
+
+            &p_{rr}& =& (1 - Q_9)(p_r + p_{erc}) + l_{exc}\\
+            &p_{rd}& =& Q_9(p_r + p_{erc}) \\
+            &Q_9& =& 0.9 \tanh(\alpha_2 p_n)^2 + 0.1
+            
+        \end{eqnarray}
+
+    with :math:`\alpha_2` in :math:`mm` per time unit.
+
+
+    .. note::
+
+        If :math:`\alpha_2 = 0`, we return to the ``gr-4/gr-5`` writting of the transfer.
+        If :math:`\alpha_2 = \alpha_1 = 0`, it is equivalent to ``gr-5`` structure.
+        
+
+
+.. _math_num_documentation.forward_structure.hydrological_module.gr6:
 
 .. dropdown:: gr6 (Génie Rural 6)
     :animate: fade-in-slide-down
diff --git a/smash/_constant.py b/smash/_constant.py
index a1c3d4dc..4d791cd0 100644
--- a/smash/_constant.py
+++ b/smash/_constant.py
@@ -64,10 +64,12 @@ def get_neurons_from_hydrological_module(hydrological_module: str, hidden_neuron
 
 HYDROLOGICAL_MODULE = [
     "gr4",
+    "gr4_ri",
     "gr4_mlp",
     "gr4_ode",
     "gr4_ode_mlp",
     "gr5",
+    "gr5_ri",
     "gr6",
     "grc",
     "grd",
@@ -97,10 +99,12 @@ def get_neurons_from_hydrological_module(hydrological_module: str, hidden_neuron
         HYDROLOGICAL_MODULE,
         [
             ["ci", "cp", "ct", "kexc"],  # % gr4
+            ["ci", "cp", "ct", "alpha1", "alpha2", "kexc"],  # % gr4_ri
             ["ci", "cp", "ct", "kexc"],  # % gr4_mlp
             ["ci", "cp", "ct", "kexc"],  # % gr4_ode
             ["ci", "cp", "ct", "kexc"],  # % gr4_ode_mlp
             ["ci", "cp", "ct", "kexc", "aexc"],  # % gr5
+            ["ci", "cp", "ct", "alpha1", "alpha2", "kexc", "aexc"],  # % gr5_ri
             ["ci", "cp", "ct", "be", "kexc", "aexc"],  # % gr6
             ["ci", "cp", "ct", "cl", "kexc"],  # % grc
             ["cp", "ct"],  # % grd
@@ -139,10 +143,12 @@ def get_neurons_from_hydrological_module(hydrological_module: str, hidden_neuron
         HYDROLOGICAL_MODULE,
         [
             ["hi", "hp", "ht"],  # % gr4
+            ["hi", "hp", "ht"],  # % gr4_ri
             ["hi", "hp", "ht"],  # % gr4_mlp
             ["hi", "hp", "ht"],  # % gr4_ode
             ["hi", "hp", "ht"],  # % gr4_ode_mlp
             ["hi", "hp", "ht"],  # % gr5
+            ["hi", "hp", "ht"],  # % gr5_ri
             ["hi", "hp", "ht", "he"],  # % gr6
             ["hi", "hp", "ht", "hl"],  # % grc
             ["hp", "ht"],  # % grd
@@ -183,10 +189,12 @@ def get_neurons_from_hydrological_module(hydrological_module: str, hidden_neuron
         HYDROLOGICAL_MODULE,
         [
             ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "qr", "qd", "qt"],  # % gr4
+            ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "qr", "qd", "qt"],  # % gr4-ri
             ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "qr", "qd", "qt"],  # % gr4_mlp
             ["pn", "en", "lexc", "qt"],  # % gr4_ode
             ["pn", "en", "lexc", "qt"],  # % gr4_ode_mlp
             ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "qr", "qd", "qt"],  # % gr5
+            ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "qr", "qd", "qt"],  # % gr5-ri
             ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "pre", "qr", "qd", "qe", "qt"],  # % gr6
             ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "prl", "qr", "qd", "ql", "qt"],  # % grc
             ["ei", "pn", "en", "pr", "perc", "prr", "qr", "qt"],  # % grd
@@ -250,13 +258,15 @@ def get_neurons_from_hydrological_module(hydrological_module: str, hidden_neuron
 
 RR_PARAMETERS = [
     "kmlt",  # % ssn
-    "ci",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc)
-    "cp",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc, grd)
-    "ct",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc, grd)
+    "ci",  # % (gr4, gr4_ri, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr5_ri, grc, gr6)
+    "cp",  # % (gr4, gr4_ri, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr5_ri, grc, gr6, grd)
+    "ct",  # % (gr4, gr4_ri, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr5_ri, grc, gr6, grd)
+    "alpha1",  # % (gr4_ri, gr5_ri)
+    "alpha2",  # % (gr4_ri, gr5_ri)
     "cl",  # % grc
-    "be",  # % gr6
-    "kexc",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc)
-    "aexc",  # % (gr5, gr6)
+    "be",  # % (gr6)
+    "kexc",  # % (gr4, gr4_ri, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr5_ri, grc, gr6)
+    "aexc",  # % (gr5, gr5_ri, gr6)
     "ca",  # % loieau
     "cc",  # % loieau
     "kb",  # % loieau
@@ -276,9 +286,9 @@ def get_neurons_from_hydrological_module(hydrological_module: str, hidden_neuron
 
 RR_STATES = [
     "hs",  # % ssn
-    "hi",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc)
-    "hp",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc, grd)
-    "ht",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc, grd)
+    "hi",  # % (gr4, gr4_ri, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr5_ri, grc, gr6)
+    "hp",  # % (gr4, gr4_ri, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr5_ri, grc, gr6, grd)
+    "ht",  # % (gr4, gr4_ri, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr5_ri, grc, gr6, grd)
     "hl",  # % grc
     "he",  # % gr6
     "ha",  # % loieau
@@ -302,6 +312,8 @@ def get_neurons_from_hydrological_module(hydrological_module: str, hidden_neuron
             (0, np.inf),  # % ci
             (0, np.inf),  # % cp
             (0, np.inf),  # % ct
+            (0, np.inf),  # % alpha1
+            (0, np.inf),  # % alpha2
             (0, np.inf),  # % cl
             (0, np.inf),  # % be
             (-np.inf, np.inf),  # % kexc
@@ -361,6 +373,8 @@ def get_neurons_from_hydrological_module(hydrological_module: str, hidden_neuron
             1e-6,  # % ci
             200,  # % cp
             500,  # % ct
+            3e-4,  # % alpha1
+            1e-3,  # % alpha2
             500,  # % cl
             10,  # % be
             0,  # % kexc
@@ -418,6 +432,8 @@ def get_neurons_from_hydrological_module(hydrological_module: str, hidden_neuron
             (1e-6, 1e2),  # % ci
             (1e-6, 1e3),  # % cp
             (1e-6, 1e3),  # % ct
+            (1e-6, 5e-4),  # % alpha1
+            (1e-5, 1.0),  # % alpha2
             (1e-6, 1e3),  # % cl
             (1e-3, 20),  # % be
             (-50, 50),  # % kexc
diff --git a/smash/fcore/forward/forward_db.f90 b/smash/fcore/forward/forward_db.f90
index 9f8182a4..e99d16cb 100644
--- a/smash/fcore/forward/forward_db.f90
+++ b/smash/fcore/forward/forward_db.f90
@@ -13470,6 +13470,260 @@ SUBROUTINE GR_PRODUCTION(fq_ps, fq_es, pn, en, cp, beta, hp, pr, perc)
     hp = hp_imd - perc*inv_cp
   END SUBROUTINE GR_PRODUCTION
 
+!  Differentiation of gr_ri_production in forward (tangent) mode (with options fixinterface noISIZE context):
+!   variations   of useful results: hp perc pr
+!   with respect to varying inputs: alpha1 hp en cp pn
+  SUBROUTINE GR_RI_PRODUCTION_D(pn, pn_d, en, en_d, cp, cp_d, beta, &
+&   alpha1, alpha1_d, hp, hp_d, pr, pr_d, perc, perc_d, dt)
+    IMPLICIT NONE
+    REAL(sp), INTENT(IN) :: pn, en, cp, beta, alpha1
+    REAL(sp), INTENT(IN) :: pn_d, en_d, cp_d, alpha1_d
+    REAL(sp), INTENT(IN) :: dt
+    REAL(sp), INTENT(INOUT) :: hp
+    REAL(sp), INTENT(INOUT) :: hp_d
+    REAL(sp), INTENT(OUT) :: pr, perc
+    REAL(sp), INTENT(OUT) :: pr_d, perc_d
+    REAL(sp) :: inv_cp, ps, es, hp_imd
+    REAL(sp) :: inv_cp_d, ps_d, es_d, hp_imd_d
+    REAL(sp) :: lambda, gam, inv_lambda
+    REAL(sp) :: lambda_d, gam_d, inv_lambda_d
+    INTRINSIC EXP
+    INTRINSIC SQRT
+    INTRINSIC TANH
+    REAL(sp) :: arg1
+    REAL(sp) :: arg1_d
+    REAL(sp) :: arg2
+    REAL(sp) :: arg2_d
+    REAL(sp) :: pwx1
+    REAL(sp) :: pwx1_d
+    REAL(sp) :: pwr1
+    REAL(sp) :: pwr1_d
+    REAL(sp) :: temp
+    REAL(sp) :: temp0
+    REAL(sp) :: temp1
+    REAL(sp) :: temp2
+    REAL(sp) :: temp3
+    REAL(sp) :: temp4
+    inv_cp_d = -(cp_d/cp**2)
+    inv_cp = 1._sp/cp
+    pr = 0._sp
+    gam_d = EXP(-(pn*alpha1))*(alpha1*pn_d+pn*alpha1_d)
+    gam = 1._sp - EXP(-(pn*alpha1))
+    temp = SQRT(-gam + 1._sp)
+    IF (1._sp - gam .EQ. 0.0) THEN
+      lambda_d = 0.0_4
+    ELSE
+      lambda_d = -(gam_d/(2.0*temp))
+    END IF
+    lambda = temp
+    inv_lambda_d = -(lambda_d/lambda**2)
+    inv_lambda = 1._sp/lambda
+    arg1_d = inv_cp*(pn*lambda_d+lambda*pn_d) + lambda*pn*inv_cp_d
+    arg1 = lambda*pn*inv_cp
+    arg2_d = inv_cp*(pn*lambda_d+lambda*pn_d) + lambda*pn*inv_cp_d
+    arg2 = lambda*pn*inv_cp
+    temp = TANH(arg2)
+    temp0 = lambda*hp*temp + 1._sp
+    temp1 = -(lambda*hp*(lambda*hp)) + 1._sp
+    temp2 = cp*inv_lambda*temp1
+    temp3 = TANH(arg1)
+    temp4 = temp3*temp2/temp0
+    ps_d = (temp2*(1.0-TANH(arg1)**2)*arg1_d+temp3*(temp1*(inv_lambda*&
+&     cp_d+cp*inv_lambda_d)-cp*inv_lambda*2*lambda*hp*(hp*lambda_d+&
+&     lambda*hp_d))-temp4*(temp*(hp*lambda_d+lambda*hp_d)+lambda*hp*(1.0&
+&     -TANH(arg2)**2)*arg2_d))/temp0 - dt*gam_d
+    ps = temp4 - dt*gam
+    temp4 = TANH(en*inv_cp)
+    temp3 = TANH(en*inv_cp)
+    temp2 = hp*cp*(-hp+2._sp)
+    temp1 = temp2*temp3/((-hp+1._sp)*temp4+1._sp)
+    es_d = (temp3*((2._sp-hp)*(cp*hp_d+hp*cp_d)-hp*cp*hp_d)+temp2*(1.0-&
+&     TANH(en*inv_cp)**2)*(inv_cp*en_d+en*inv_cp_d)-temp1*((1._sp-hp)*(&
+&     1.0-TANH(en*inv_cp)**2)*(inv_cp*en_d+en*inv_cp_d)-temp4*hp_d))/((&
+&     1._sp-hp)*temp4+1._sp)
+    es = temp1
+    hp_imd_d = hp_d + inv_cp*(ps_d-es_d) + (ps-es)*inv_cp_d
+    hp_imd = hp + (ps-es)*inv_cp
+    IF (pn .GT. 0) THEN
+      pr_d = pn_d - cp*(hp_imd_d-hp_d) - (hp_imd-hp)*cp_d
+      pr = pn - (hp_imd-hp)*cp
+    ELSE
+      pr_d = 0.0_4
+    END IF
+    pwx1_d = 4*hp_imd**3*hp_imd_d/beta**4
+    pwx1 = 1._sp + (hp_imd/beta)**4
+    pwr1_d = -(0.25_sp*pwx1**(-1.25)*pwx1_d)
+    pwr1 = pwx1**(-0.25_sp)
+    perc_d = (1._sp-pwr1)*(cp*hp_imd_d+hp_imd*cp_d) - hp_imd*cp*pwr1_d
+    perc = hp_imd*cp*(1._sp-pwr1)
+    hp_d = hp_imd_d - inv_cp*perc_d - perc*inv_cp_d
+    hp = hp_imd - perc*inv_cp
+  END SUBROUTINE GR_RI_PRODUCTION_D
+
+!  Differentiation of gr_ri_production in reverse (adjoint) mode (with options fixinterface noISIZE context):
+!   gradient     of useful results: alpha1 hp cp pn perc pr
+!   with respect to varying inputs: alpha1 hp en cp pn
+  SUBROUTINE GR_RI_PRODUCTION_B(pn, pn_b, en, en_b, cp, cp_b, beta, &
+&   alpha1, alpha1_b, hp, hp_b, pr, pr_b, perc, perc_b, dt)
+    IMPLICIT NONE
+    REAL(sp), INTENT(IN) :: pn, en, cp, beta, alpha1
+    REAL(sp) :: pn_b, en_b, cp_b, alpha1_b
+    REAL(sp), INTENT(IN) :: dt
+    REAL(sp), INTENT(INOUT) :: hp
+    REAL(sp), INTENT(INOUT) :: hp_b
+    REAL(sp) :: pr, perc
+    REAL(sp) :: pr_b, perc_b
+    REAL(sp) :: inv_cp, ps, es, hp_imd
+    REAL(sp) :: inv_cp_b, ps_b, es_b, hp_imd_b
+    REAL(sp) :: lambda, gam, inv_lambda
+    REAL(sp) :: lambda_b, gam_b, inv_lambda_b
+    INTRINSIC EXP
+    INTRINSIC SQRT
+    INTRINSIC TANH
+    REAL(sp) :: arg1
+    REAL(sp) :: arg1_b
+    REAL(sp) :: arg2
+    REAL(sp) :: arg2_b
+    REAL(sp) :: pwx1
+    REAL(sp) :: pwx1_b
+    REAL(sp) :: pwr1
+    REAL(sp) :: pwr1_b
+    REAL(sp) :: temp
+    REAL(sp) :: temp_b
+    REAL(sp) :: temp0
+    REAL(sp) :: temp_b0
+    REAL(sp) :: temp1
+    REAL(sp) :: temp2
+    REAL(sp) :: temp3
+    REAL(sp) :: temp_b1
+    REAL(sp) :: temp4
+    REAL(sp) :: temp_b2
+    REAL(sp) :: temp_b3
+    REAL(sp) :: temp_b4
+    REAL(sp) :: temp_b5
+    INTEGER :: branch
+    inv_cp = 1._sp/cp
+    gam = 1._sp - EXP(-(pn*alpha1))
+    lambda = SQRT(1._sp - gam)
+    inv_lambda = 1._sp/lambda
+    arg1 = lambda*pn*inv_cp
+    arg2 = lambda*pn*inv_cp
+    ps = cp*inv_lambda*TANH(arg1)*(1._sp-(lambda*hp)**2)/(1._sp+lambda*&
+&     hp*TANH(arg2)) - gam*dt
+    es = hp*cp*(2._sp-hp)*TANH(en*inv_cp)/(1._sp+(1._sp-hp)*TANH(en*&
+&     inv_cp))
+    hp_imd = hp + (ps-es)*inv_cp
+    IF (pn .GT. 0) THEN
+      CALL PUSHCONTROL1B(0)
+    ELSE
+      CALL PUSHCONTROL1B(1)
+    END IF
+    pwx1 = 1._sp + (hp_imd/beta)**4
+    pwr1 = pwx1**(-0.25_sp)
+    CALL PUSHREAL4(perc)
+    perc = hp_imd*cp*(1._sp-pwr1)
+    pwx1 = 1._sp + (hp_imd/beta)**4
+    pwr1 = pwx1**(-0.25_sp)
+    inv_cp = 1._sp/cp
+    perc_b = perc_b - inv_cp*hp_b
+    inv_cp_b = -(perc*hp_b)
+    CALL POPREAL4(perc)
+    cp_b = cp_b + hp_imd*(1._sp-pwr1)*perc_b
+    pwr1_b = -(hp_imd*cp*perc_b)
+    pwx1_b = -(0.25_sp*pwx1**(-1.25)*pwr1_b)
+    hp_imd_b = hp_b + cp*(1._sp-pwr1)*perc_b + 4*hp_imd**3*pwx1_b/beta**&
+&     4
+    CALL POPCONTROL1B(branch)
+    IF (branch .EQ. 0) THEN
+      pn_b = pn_b + pr_b
+      hp_imd_b = hp_imd_b - cp*pr_b
+      hp_b = cp*pr_b
+      cp_b = cp_b - (hp_imd-hp)*pr_b
+    ELSE
+      hp_b = 0.0_4
+    END IF
+    es_b = -(inv_cp*hp_imd_b)
+    temp4 = TANH(en*inv_cp)
+    temp3 = (-hp+1._sp)*temp4 + 1._sp
+    temp1 = TANH(en*inv_cp)
+    temp0 = hp*cp*(-hp+2._sp)
+    temp_b1 = es_b/temp3
+    temp_b0 = (2._sp-hp)*temp1*temp_b1
+    temp_b4 = -(temp0*temp1*temp_b1/temp3)
+    hp_b = hp_b + hp_imd_b + cp*temp_b0 - hp*cp*temp1*temp_b1 - temp4*&
+&     temp_b4
+    ps_b = inv_cp*hp_imd_b
+    temp_b = (1.0-TANH(en*inv_cp)**2)*temp0*temp_b1
+    temp_b5 = (1.0-TANH(en*inv_cp)**2)*(1._sp-hp)*temp_b4
+    en_b = inv_cp*temp_b5 + inv_cp*temp_b
+    cp_b = cp_b + hp*temp_b0
+    arg1 = lambda*pn*inv_cp
+    arg2 = lambda*pn*inv_cp
+    inv_lambda = 1._sp/lambda
+    temp = TANH(arg2)
+    temp0 = lambda*hp*temp + 1._sp
+    temp1 = -(lambda*hp*(lambda*hp)) + 1._sp
+    temp2 = cp*inv_lambda*temp1
+    temp3 = TANH(arg1)
+    temp_b0 = ps_b/temp0
+    arg1_b = (1.0-TANH(arg1)**2)*temp2*temp_b0
+    temp_b1 = temp3*temp_b0
+    temp_b3 = -(temp3*temp2*temp_b0/temp0)
+    arg2_b = (1.0-TANH(arg2)**2)*lambda*hp*temp_b3
+    inv_cp_b = inv_cp_b + (ps-es)*hp_imd_b + en*temp_b5 + en*temp_b + &
+&     lambda*pn*arg2_b + lambda*pn*arg1_b
+    cp_b = cp_b + inv_lambda*temp1*temp_b1 - inv_cp_b/cp**2
+    inv_lambda_b = cp*temp1*temp_b1
+    temp_b2 = -(2*lambda*hp*cp*inv_lambda*temp_b1)
+    lambda_b = hp*temp*temp_b3 + hp*temp_b2 + pn*inv_cp*arg2_b + pn*&
+&     inv_cp*arg1_b - inv_lambda_b/lambda**2
+    IF (1._sp - gam .EQ. 0.0) THEN
+      gam_b = -(dt*ps_b)
+    ELSE
+      gam_b = -(dt*ps_b) - lambda_b/(2.0*SQRT(1._sp-gam))
+    END IF
+    hp_b = hp_b + lambda*temp*temp_b3 + lambda*temp_b2
+    temp_b = -(EXP(-(pn*alpha1))*gam_b)
+    pn_b = pn_b + lambda*inv_cp*arg2_b + lambda*inv_cp*arg1_b - alpha1*&
+&     temp_b
+    alpha1_b = alpha1_b - pn*temp_b
+  END SUBROUTINE GR_RI_PRODUCTION_B
+
+  SUBROUTINE GR_RI_PRODUCTION(pn, en, cp, beta, alpha1, hp, pr, perc, dt&
+& )
+    IMPLICIT NONE
+    REAL(sp), INTENT(IN) :: pn, en, cp, beta, alpha1
+    REAL(sp), INTENT(IN) :: dt
+    REAL(sp), INTENT(INOUT) :: hp
+    REAL(sp), INTENT(OUT) :: pr, perc
+    REAL(sp) :: inv_cp, ps, es, hp_imd
+    REAL(sp) :: lambda, gam, inv_lambda
+    INTRINSIC EXP
+    INTRINSIC SQRT
+    INTRINSIC TANH
+    REAL(sp) :: arg1
+    REAL(sp) :: arg2
+    REAL(sp) :: pwx1
+    REAL(sp) :: pwr1
+    inv_cp = 1._sp/cp
+    pr = 0._sp
+    gam = 1._sp - EXP(-(pn*alpha1))
+    lambda = SQRT(1._sp - gam)
+    inv_lambda = 1._sp/lambda
+    arg1 = lambda*pn*inv_cp
+    arg2 = lambda*pn*inv_cp
+    ps = cp*inv_lambda*TANH(arg1)*(1._sp-(lambda*hp)**2)/(1._sp+lambda*&
+&     hp*TANH(arg2)) - gam*dt
+    es = hp*cp*(2._sp-hp)*TANH(en*inv_cp)/(1._sp+(1._sp-hp)*TANH(en*&
+&     inv_cp))
+    hp_imd = hp + (ps-es)*inv_cp
+    IF (pn .GT. 0) pr = pn - (hp_imd-hp)*cp
+    pwx1 = 1._sp + (hp_imd/beta)**4
+    pwr1 = pwx1**(-0.25_sp)
+    perc = hp_imd*cp*(1._sp-pwr1)
+    hp = hp_imd - perc*inv_cp
+  END SUBROUTINE GR_RI_PRODUCTION
+
 !  Differentiation of gr_exchange in forward (tangent) mode (with options fixinterface noISIZE context):
 !   variations   of useful results: l
 !   with respect to varying inputs: kexc fq_l ht
@@ -14803,17 +15057,15 @@ SUBROUTINE GR4_TIME_STEP(setup, mesh, input_data, options, returns, &
     END DO
   END SUBROUTINE GR4_TIME_STEP
 
-!  Differentiation of gr4_mlp_time_step in forward (tangent) mode (with options fixinterface noISIZE context):
+!  Differentiation of gr4_ri_time_step in forward (tangent) mode (with options fixinterface noISIZE context):
 !   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-  SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
-&   returns, time_step, weight_1, weight_1_d, bias_1, bias_1_d, weight_2&
-&   , weight_2_d, bias_2, bias_2_d, weight_3, weight_3_d, bias_3, &
-&   bias_3_d, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
-&   ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d, ac_ht, &
-&   ac_ht_d, ac_qt, ac_qt_d)
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt
+  SUBROUTINE GR4_RI_TIME_STEP_D(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d&
+&   , ac_ct, ac_ct_d, ac_alpha1, ac_alpha1_d, ac_alpha2, ac_alpha2_d, &
+&   ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d, ac_ht, ac_ht_d, &
+&   ac_qt, ac_qt_d)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -14821,46 +15073,27 @@ SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1_d
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1_d
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2_d
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2_d
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3_d
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
 &   ac_kexc
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci_d, ac_cp_d, &
 &   ac_ct_d, ac_kexc_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1_d, &
+&   ac_alpha2_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_d, ac_hp_d, &
 &   ac_ht_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_d
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_d
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer_d
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d, pn_d, en_d
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
-    REAL(sp) :: pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    REAL(sp) :: pn_d, en_d, pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d&
+&   , split_d
+    INTRINSIC TANH
     INTRINSIC MAX
     REAL(sp) :: temp
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
@@ -14871,9 +15104,7 @@ SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
     ac_prcp = ac_prcp + ac_mlt
 ! Beta percolation parameter is time step dependent
     beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
-    en_d = 0.0_4
     pn_d = 0.0_4
-! Interception with OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -14882,54 +15113,13 @@ SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
           IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
             CALL GR_INTERCEPTION_D(ac_prcp(k), ac_prcp_d(k), ac_pet(k), &
 &                            ac_ci(k), ac_ci_d(k), ac_hi(k), ac_hi_d(k)&
-&                            , pn(k), pn_d(k), en(k), en_d(k))
-          ELSE
-            pn_d(k) = 0.0_4
-            pn(k) = 0._sp
-            en_d(k) = 0.0_4
-            en(k) = 0._sp
-          END IF
-        END IF
-      END DO
-    END DO
-    output_layer_d = 0.0_4
-! Forward MLP without OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            input_layer_d(:) = (/ac_hp_d(k), ac_ht_d(k), pn_d(k), en_d(k&
-&             )/)
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP_D(weight_1, weight_1_d, bias_1, bias_1_d, &
-&                        weight_2, weight_2_d, bias_2, bias_2_d, &
-&                        weight_3, weight_3_d, bias_3, bias_3_d, &
-&                        input_layer, input_layer_d, output_layer(:, k)&
-&                        , output_layer_d(:, k))
-          ELSE
-            output_layer_d(:, k) = 0.0_4
-            output_layer(:, k) = 0._sp
-          END IF
-        END IF
-      END DO
-    END DO
-! Production and transfer with OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            CALL GR_PRODUCTION_D(output_layer(1, k), output_layer_d(1, k&
-&                          ), output_layer(2, k), output_layer_d(2, k), &
-&                          pn(k), pn_d(k), en(k), en_d(k), ac_cp(k), &
-&                          ac_cp_d(k), beta, ac_hp(k), ac_hp_d(k), pr, &
-&                          pr_d, perc, perc_d)
-            CALL GR_EXCHANGE_D(output_layer(4, k), output_layer_d(4, k)&
-&                        , ac_kexc(k), ac_kexc_d(k), ac_ht(k), ac_ht_d(k&
-&                        ), l, l_d)
+&                            , pn, pn_d, en, en_d)
+            CALL GR_RI_PRODUCTION_D(pn, pn_d, en, en_d, ac_cp(k), &
+&                             ac_cp_d(k), beta, ac_alpha1(k), &
+&                             ac_alpha1_d(k), ac_hp(k), ac_hp_d(k), pr, &
+&                             pr_d, perc, perc_d, setup%dt)
+            CALL GR_EXCHANGE_D(0._sp, 0.0_4, ac_kexc(k), ac_kexc_d(k), &
+&                        ac_ht(k), ac_ht_d(k), l, l_d)
           ELSE
             pr = 0._sp
             perc = 0._sp
@@ -14938,14 +15128,13 @@ SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
             perc_d = 0.0_4
             pr_d = 0.0_4
           END IF
-          temp = -(output_layer(3, k)*output_layer(3, k)) + 1._sp
-          prr_d = 0.9_sp*(temp*(pr_d+perc_d)-(pr+perc)*2*output_layer(3&
-&           , k)*output_layer_d(3, k)) + l_d
-          prr = 0.9_sp*(temp*(pr+perc)) + l
-          temp = 0.9_sp*(output_layer(3, k)*output_layer(3, k)) + 0.1_sp
-          prd_d = (pr+perc)*0.9_sp*2*output_layer(3, k)*output_layer_d(3&
-&           , k) + temp*(pr_d+perc_d)
-          prd = temp*(pr+perc)
+          split_d = 0.9_sp*2*TANH(ac_alpha2(k)*pn)*(1.0-TANH(ac_alpha2(k&
+&           )*pn)**2)*(pn*ac_alpha2_d(k)+ac_alpha2(k)*pn_d)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
+          prr_d = (1._sp-split)*(pr_d+perc_d) - (pr+perc)*split_d + l_d
+          prr = (1._sp-split)*(pr+perc) + l
+          prd_d = (pr+perc)*split_d + split*(pr_d+perc_d)
+          prd = split*(pr+perc)
           CALL GR_TRANSFER_D(5._sp, ac_prcp(k), prr, prr_d, ac_ct(k), &
 &                      ac_ct_d(k), ac_ht(k), ac_ht_d(k), qr, qr_d)
           IF (0._sp .LT. prd + l) THEN
@@ -14964,21 +15153,18 @@ SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-  END SUBROUTINE GR4_MLP_TIME_STEP_D
+  END SUBROUTINE GR4_RI_TIME_STEP_D
 
-!  Differentiation of gr4_mlp_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context):
-!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-  SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
-&   returns, time_step, weight_1, weight_1_b, bias_1, bias_1_b, weight_2&
-&   , weight_2_b, bias_2, bias_2_b, weight_3, weight_3_b, bias_3, &
-&   bias_3_b, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
-&   ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b, ac_ht, &
-&   ac_ht_b, ac_qt, ac_qt_b)
+!  Differentiation of gr4_ri_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt
+  SUBROUTINE GR4_RI_TIME_STEP_B(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b&
+&   , ac_ct, ac_ct_b, ac_alpha1, ac_alpha1_b, ac_alpha2, ac_alpha2_b, &
+&   ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b, ac_ht, ac_ht_b, &
+&   ac_qt, ac_qt_b)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -14986,47 +15172,28 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)) :: &
-&   weight_1_b
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
-    REAL(sp), DIMENSION(setup%neurons(2)) :: bias_1_b
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)) :: &
-&   weight_2_b
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
-    REAL(sp), DIMENSION(setup%neurons(3)) :: bias_2_b
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)) :: &
-&   weight_3_b
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
-    REAL(sp), DIMENSION(setup%neurons(4)) :: bias_3_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac) :: ac_mlt_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
 &   ac_kexc
     REAL(sp), DIMENSION(mesh%nac) :: ac_ci_b, ac_cp_b, ac_ct_b, &
 &   ac_kexc_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
+    REAL(sp), DIMENSION(mesh%nac) :: ac_alpha1_b, ac_alpha2_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_b, ac_hp_b, &
 &   ac_ht_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_b
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_b
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer_b
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b, pn_b, en_b
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
-    REAL(sp) :: pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    REAL(sp) :: pn_b, en_b, pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b&
+&   , split_b
+    INTRINSIC TANH
     INTRINSIC MAX
+    REAL(sp) :: dummydiff_b
     REAL(sp) :: temp_b
     INTEGER :: branch
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
@@ -15036,67 +15203,25 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
     ac_prcp = ac_prcp + ac_mlt
 ! Beta percolation parameter is time step dependent
     beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
-! Interception with OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
 &           local_active_cell(row, col) .EQ. 0) THEN
-          CALL PUSHCONTROL2B(0)
+          CALL PUSHCONTROL1B(0)
         ELSE
           k = mesh%rowcol_to_ind_ac(row, col)
           IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4(en)
+            CALL PUSHREAL4(pn)
             CALL PUSHREAL4(ac_hi(k))
             CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
-&                          k), pn(k), en(k))
-            CALL PUSHCONTROL2B(2)
-          ELSE
-            pn(k) = 0._sp
-            en(k) = 0._sp
-            CALL PUSHCONTROL2B(1)
-          END IF
-        END IF
-      END DO
-    END DO
-! Forward MLP without OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0) THEN
-          CALL PUSHCONTROL2B(0)
-        ELSE
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            CALL PUSHREAL4ARRAY(input_layer, setup%neurons(1))
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
-&                      weight_3, bias_3, input_layer, output_layer(:, k)&
-&                     )
-            CALL PUSHCONTROL2B(2)
-          ELSE
-            output_layer(:, k) = 0._sp
-            CALL PUSHCONTROL2B(1)
-          END IF
-        END IF
-      END DO
-    END DO
-! Production and transfer with OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0) THEN
-          CALL PUSHCONTROL1B(0)
-        ELSE
-          CALL PUSHINTEGER4(k)
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+&                          k), pn, en)
             CALL PUSHREAL4(perc)
             CALL PUSHREAL4(pr)
             CALL PUSHREAL4(ac_hp(k))
-            CALL GR_PRODUCTION(output_layer(1, k), output_layer(2, k), &
-&                        pn(k), en(k), ac_cp(k), beta, ac_hp(k), pr, &
-&                        perc)
-            CALL GR_EXCHANGE(output_layer(4, k), ac_kexc(k), ac_ht(k), l&
-&                     )
+            CALL GR_RI_PRODUCTION(pn, en, ac_cp(k), beta, ac_alpha1(k), &
+&                           ac_hp(k), pr, perc, setup%dt)
+            CALL GR_EXCHANGE(0._sp, ac_kexc(k), ac_ht(k), l)
             CALL PUSHCONTROL1B(1)
           ELSE
             CALL PUSHREAL4(pr)
@@ -15106,9 +15231,11 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
             l = 0._sp
             CALL PUSHCONTROL1B(0)
           END IF
+          CALL PUSHREAL4(split)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
           CALL PUSHREAL4(prr)
-          prr = 0.9_sp*(1._sp-output_layer(3, k)**2)*(pr+perc) + l
-          prd = (0.1_sp+0.9_sp*output_layer(3, k)**2)*(pr+perc)
+          prr = (1._sp-split)*(pr+perc) + l
+          prd = split*(pr+perc)
           CALL PUSHREAL4(ac_ht(k))
           CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
 &                    qr)
@@ -15121,13 +15248,13 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-    output_layer_b = 0.0_4
-    en_b = 0.0_4
+    ac_prcp_b = 0.0_4
     pn_b = 0.0_4
     DO col=mesh%ncol,1,-1
       DO row=mesh%nrow,1,-1
         CALL POPCONTROL1B(branch)
         IF (branch .NE. 0) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
           ac_qt_b(k) = mesh%dx(row, col)*1e-3_sp*mesh%dy(row, col)*&
 &           ac_qt_b(k)/setup%dt
           qr_b = ac_qt_b(k)
@@ -15144,89 +15271,47 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
           CALL POPREAL4(ac_ht(k))
           CALL GR_TRANSFER_B(5._sp, ac_prcp(k), prr, prr_b, ac_ct(k), &
 &                      ac_ct_b(k), ac_ht(k), ac_ht_b(k), qr, qr_b)
-          output_layer_b(3, k) = output_layer_b(3, k) + 2*output_layer(3&
-&           , k)*0.9_sp*(pr+perc)*prd_b - 2*output_layer(3, k)*(pr+perc)&
-&           *0.9_sp*prr_b
-          temp_b = (0.9_sp*output_layer(3, k)**2+0.1_sp)*prd_b
-          pr_b = temp_b
-          perc_b = temp_b
+          split_b = (pr+perc)*prd_b - (pr+perc)*prr_b
+          pr_b = split*prd_b + (1._sp-split)*prr_b
+          perc_b = split*prd_b + (1._sp-split)*prr_b
           CALL POPREAL4(prr)
-          temp_b = (1._sp-output_layer(3, k)**2)*0.9_sp*prr_b
           l_b = l_b + prr_b
-          pr_b = pr_b + temp_b
-          perc_b = perc_b + temp_b
+          CALL POPREAL4(split)
+          temp_b = (1.0-TANH(ac_alpha2(k)*pn)**2)*2*TANH(ac_alpha2(k)*pn&
+&           )*0.9_sp*split_b
+          ac_alpha2_b(k) = ac_alpha2_b(k) + pn*temp_b
+          pn_b = pn_b + ac_alpha2(k)*temp_b
           CALL POPCONTROL1B(branch)
           IF (branch .EQ. 0) THEN
             CALL POPREAL4(perc)
             CALL POPREAL4(pr)
           ELSE
-            CALL GR_EXCHANGE_B(output_layer(4, k), output_layer_b(4, k)&
-&                        , ac_kexc(k), ac_kexc_b(k), ac_ht(k), ac_ht_b(k&
-&                        ), l, l_b)
+            CALL GR_EXCHANGE_B(0._sp, dummydiff_b, ac_kexc(k), ac_kexc_b&
+&                        (k), ac_ht(k), ac_ht_b(k), l, l_b)
             CALL POPREAL4(ac_hp(k))
             CALL POPREAL4(pr)
             CALL POPREAL4(perc)
-            CALL GR_PRODUCTION_B(output_layer(1, k), output_layer_b(1, k&
-&                          ), output_layer(2, k), output_layer_b(2, k), &
-&                          pn(k), pn_b(k), en(k), en_b(k), ac_cp(k), &
-&                          ac_cp_b(k), beta, ac_hp(k), ac_hp_b(k), pr, &
-&                          pr_b, perc, perc_b)
-          END IF
-          CALL POPINTEGER4(k)
-        END IF
-      END DO
-    END DO
-    DO col=mesh%ncol,1,-1
-      DO row=mesh%nrow,1,-1
-        CALL POPCONTROL2B(branch)
-        IF (branch .NE. 0) THEN
-          IF (branch .EQ. 1) THEN
-            k = mesh%rowcol_to_ind_ac(row, col)
-            output_layer_b(:, k) = 0.0_4
-          ELSE
-            k = mesh%rowcol_to_ind_ac(row, col)
-            CALL FORWARD_MLP_B(weight_1, weight_1_b, bias_1, bias_1_b, &
-&                        weight_2, weight_2_b, bias_2, bias_2_b, &
-&                        weight_3, weight_3_b, bias_3, bias_3_b, &
-&                        input_layer, input_layer_b, output_layer(:, k)&
-&                        , output_layer_b(:, k))
-            output_layer_b(:, k) = 0.0_4
-            CALL POPREAL4ARRAY(input_layer, setup%neurons(1))
-            ac_hp_b(k) = ac_hp_b(k) + input_layer_b(1)
-            ac_ht_b(k) = ac_ht_b(k) + input_layer_b(2)
-            pn_b(k) = pn_b(k) + input_layer_b(3)
-            en_b(k) = en_b(k) + input_layer_b(4)
-          END IF
-        END IF
-      END DO
-    END DO
-    ac_prcp_b = 0.0_4
-    DO col=mesh%ncol,1,-1
-      DO row=mesh%nrow,1,-1
-        CALL POPCONTROL2B(branch)
-        IF (branch .NE. 0) THEN
-          IF (branch .EQ. 1) THEN
-            k = mesh%rowcol_to_ind_ac(row, col)
-            en_b(k) = 0.0_4
-            pn_b(k) = 0.0_4
-          ELSE
-            k = mesh%rowcol_to_ind_ac(row, col)
+            CALL GR_RI_PRODUCTION_B(pn, pn_b, en, en_b, ac_cp(k), &
+&                             ac_cp_b(k), beta, ac_alpha1(k), &
+&                             ac_alpha1_b(k), ac_hp(k), ac_hp_b(k), pr, &
+&                             pr_b, perc, perc_b, setup%dt)
             CALL POPREAL4(ac_hi(k))
+            CALL POPREAL4(pn)
+            CALL POPREAL4(en)
             CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
 &                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
-&                            , pn(k), pn_b(k), en(k), en_b(k))
-            pn_b(k) = 0.0_4
-            en_b(k) = 0.0_4
+&                            , pn, pn_b, en, en_b)
+            pn_b = 0.0_4
           END IF
         END IF
       END DO
     END DO
     ac_mlt_b = ac_mlt_b + ac_prcp_b
-  END SUBROUTINE GR4_MLP_TIME_STEP_B
+  END SUBROUTINE GR4_RI_TIME_STEP_B
 
-  SUBROUTINE GR4_MLP_TIME_STEP(setup, mesh, input_data, options, returns&
-&   , time_step, weight_1, bias_1, weight_2, bias_2, weight_3, bias_3, &
-&   ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, ac_ht, ac_qt)
+  SUBROUTINE GR4_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&   , time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_alpha1, ac_alpha2, &
+&   ac_kexc, ac_hi, ac_hp, ac_ht, ac_qt)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15234,26 +15319,16 @@ SUBROUTINE GR4_MLP_TIME_STEP(setup, mesh, input_data, options, returns&
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
 &   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    INTRINSIC TANH
     INTRINSIC MAX
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
@@ -15262,7 +15337,6 @@ SUBROUTINE GR4_MLP_TIME_STEP(setup, mesh, input_data, options, returns&
     ac_prcp = ac_prcp + ac_mlt
 ! Beta percolation parameter is time step dependent
     beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
-! Interception with OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -15270,50 +15344,18 @@ SUBROUTINE GR4_MLP_TIME_STEP(setup, mesh, input_data, options, returns&
           k = mesh%rowcol_to_ind_ac(row, col)
           IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
             CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
-&                          k), pn(k), en(k))
-          ELSE
-            pn(k) = 0._sp
-            en(k) = 0._sp
-          END IF
-        END IF
-      END DO
-    END DO
-! Forward MLP without OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
-&                      weight_3, bias_3, input_layer, output_layer(:, k)&
-&                     )
-          ELSE
-            output_layer(:, k) = 0._sp
-          END IF
-        END IF
-      END DO
-    END DO
-! Production and transfer with OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            CALL GR_PRODUCTION(output_layer(1, k), output_layer(2, k), &
-&                        pn(k), en(k), ac_cp(k), beta, ac_hp(k), pr, &
-&                        perc)
-            CALL GR_EXCHANGE(output_layer(4, k), ac_kexc(k), ac_ht(k), l&
-&                     )
+&                          k), pn, en)
+            CALL GR_RI_PRODUCTION(pn, en, ac_cp(k), beta, ac_alpha1(k), &
+&                           ac_hp(k), pr, perc, setup%dt)
+            CALL GR_EXCHANGE(0._sp, ac_kexc(k), ac_ht(k), l)
           ELSE
             pr = 0._sp
             perc = 0._sp
             l = 0._sp
           END IF
-          prr = 0.9_sp*(1._sp-output_layer(3, k)**2)*(pr+perc) + l
-          prd = (0.1_sp+0.9_sp*output_layer(3, k)**2)*(pr+perc)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
+          prr = (1._sp-split)*(pr+perc) + l
+          prd = split*(pr+perc)
           CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
 &                    qr)
           IF (0._sp .LT. prd + l) THEN
@@ -15328,16 +15370,19 @@ SUBROUTINE GR4_MLP_TIME_STEP(setup, mesh, input_data, options, returns&
         END IF
       END DO
     END DO
-  END SUBROUTINE GR4_MLP_TIME_STEP
+  END SUBROUTINE GR4_RI_TIME_STEP
 
-!  Differentiation of gr4_ode_time_step in forward (tangent) mode (with options fixinterface noISIZE context):
+!  Differentiation of gr4_mlp_time_step in forward (tangent) mode (with options fixinterface noISIZE context):
 !   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt
-  SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
-&   returns, time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d&
-&   , ac_ct, ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d&
-&   , ac_ht, ac_ht_d, ac_qt, ac_qt_d)
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+  SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
+&   returns, time_step, weight_1, weight_1_d, bias_1, bias_1_d, weight_2&
+&   , weight_2_d, bias_2, bias_2_d, weight_3, weight_3_d, bias_3, &
+&   bias_3_d, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
+&   ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d, ac_ht, &
+&   ac_ht_d, ac_qt, ac_qt_d)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15345,21 +15390,47 @@ SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
-    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt_d
-    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
-&   ac_kexc
-    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci_d, ac_cp_d, &
-&   ac_ct_d, ac_kexc_d
-    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
-    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_d, ac_hp_d, &
-&   ac_ht_d
-    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
-    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_d
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1_d
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1_d
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2_d
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2_d
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3_d
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci_d, ac_cp_d, &
+&   ac_ct_d, ac_kexc_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_d, ac_hp_d, &
+&   ac_ht_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_d
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_d
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer_d
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d, pn_d, en_d
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: l
+    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d
+    INTRINSIC MAX
     REAL(sp) :: temp
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
@@ -15367,6 +15438,8 @@ SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
 &                              , 'pet', ac_pet)
     ac_prcp_d = ac_mlt_d
     ac_prcp = ac_prcp + ac_mlt
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
     en_d = 0.0_4
     pn_d = 0.0_4
 ! Interception with OPENMP
@@ -15388,18 +15461,71 @@ SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-! Production and transfer without OPENMP
+    output_layer_d = 0.0_4
+! Forward MLP without OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
 &           local_active_cell(row, col) .EQ. 0)) THEN
           k = mesh%rowcol_to_ind_ac(row, col)
-          CALL GR_PRODUCTION_TRANSFER_ODE_D(pn(k), pn_d(k), en(k), en_d(&
-&                                     k), ac_cp(k), ac_cp_d(k), ac_ct(k)&
-&                                     , ac_ct_d(k), ac_kexc(k), &
-&                                     ac_kexc_d(k), ac_hp(k), ac_hp_d(k)&
-&                                     , ac_ht(k), ac_ht_d(k), ac_qt(k), &
-&                                     ac_qt_d(k), l)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            input_layer_d(:) = (/ac_hp_d(k), ac_ht_d(k), pn_d(k), en_d(k&
+&             )/)
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP_D(weight_1, weight_1_d, bias_1, bias_1_d, &
+&                        weight_2, weight_2_d, bias_2, bias_2_d, &
+&                        weight_3, weight_3_d, bias_3, bias_3_d, &
+&                        input_layer, input_layer_d, output_layer(:, k)&
+&                        , output_layer_d(:, k))
+          ELSE
+            output_layer_d(:, k) = 0.0_4
+            output_layer(:, k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_PRODUCTION_D(output_layer(1, k), output_layer_d(1, k&
+&                          ), output_layer(2, k), output_layer_d(2, k), &
+&                          pn(k), pn_d(k), en(k), en_d(k), ac_cp(k), &
+&                          ac_cp_d(k), beta, ac_hp(k), ac_hp_d(k), pr, &
+&                          pr_d, perc, perc_d)
+            CALL GR_EXCHANGE_D(output_layer(4, k), output_layer_d(4, k)&
+&                        , ac_kexc(k), ac_kexc_d(k), ac_ht(k), ac_ht_d(k&
+&                        ), l, l_d)
+          ELSE
+            pr = 0._sp
+            perc = 0._sp
+            l = 0._sp
+            l_d = 0.0_4
+            perc_d = 0.0_4
+            pr_d = 0.0_4
+          END IF
+          temp = -(output_layer(3, k)*output_layer(3, k)) + 1._sp
+          prr_d = 0.9_sp*(temp*(pr_d+perc_d)-(pr+perc)*2*output_layer(3&
+&           , k)*output_layer_d(3, k)) + l_d
+          prr = 0.9_sp*(temp*(pr+perc)) + l
+          temp = 0.9_sp*(output_layer(3, k)*output_layer(3, k)) + 0.1_sp
+          prd_d = (pr+perc)*0.9_sp*2*output_layer(3, k)*output_layer_d(3&
+&           , k) + temp*(pr_d+perc_d)
+          prd = temp*(pr+perc)
+          CALL GR_TRANSFER_D(5._sp, ac_prcp(k), prr, prr_d, ac_ct(k), &
+&                      ac_ct_d(k), ac_ht(k), ac_ht_d(k), qr, qr_d)
+          IF (0._sp .LT. prd + l) THEN
+            qd_d = prd_d + l_d
+            qd = prd + l
+          ELSE
+            qd = 0._sp
+            qd_d = 0.0_4
+          END IF
+          ac_qt_d(k) = qr_d + qd_d
+          ac_qt(k) = qr + qd
 ! Transform from mm/dt to m3/s
           temp = 1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)
           ac_qt_d(k) = temp*ac_qt_d(k)/setup%dt
@@ -15407,17 +15533,21 @@ SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-  END SUBROUTINE GR4_ODE_TIME_STEP_D
+  END SUBROUTINE GR4_MLP_TIME_STEP_D
 
-!  Differentiation of gr4_ode_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context):
-!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt
-  SUBROUTINE GR4_ODE_TIME_STEP_B(setup, mesh, input_data, options, &
-&   returns, time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b&
-&   , ac_ct, ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b&
-&   , ac_ht, ac_ht_b, ac_qt, ac_qt_b)
+!  Differentiation of gr4_mlp_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+  SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
+&   returns, time_step, weight_1, weight_1_b, bias_1, bias_1_b, weight_2&
+&   , weight_2_b, bias_2, bias_2_b, weight_3, weight_3_b, bias_3, &
+&   bias_3_b, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
+&   ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b, ac_ht, &
+&   ac_ht_b, ac_qt, ac_qt_b)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15425,6 +15555,24 @@ SUBROUTINE GR4_ODE_TIME_STEP_B(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)) :: &
+&   weight_1_b
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(2)) :: bias_1_b
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)) :: &
+&   weight_2_b
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(3)) :: bias_2_b
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)) :: &
+&   weight_3_b
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
+    REAL(sp), DIMENSION(setup%neurons(4)) :: bias_3_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac) :: ac_mlt_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
@@ -15436,16 +15584,27 @@ SUBROUTINE GR4_ODE_TIME_STEP_B(setup, mesh, input_data, options, &
 &   ac_ht_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_b
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_b
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer_b
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b, pn_b, en_b
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: l
+    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b
+    INTRINSIC MAX
+    REAL(sp) :: temp_b
     INTEGER :: branch
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'pet', ac_pet)
     ac_prcp = ac_prcp + ac_mlt
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
 ! Interception with OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
@@ -15467,61 +15626,163 @@ SUBROUTINE GR4_ODE_TIME_STEP_B(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-! Production and transfer without OPENMP
+! Forward MLP without OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0) THEN
+          CALL PUSHCONTROL2B(0)
+        ELSE
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4ARRAY(input_layer, setup%neurons(1))
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
+&                      weight_3, bias_3, input_layer, output_layer(:, k)&
+&                     )
+            CALL PUSHCONTROL2B(2)
+          ELSE
+            output_layer(:, k) = 0._sp
+            CALL PUSHCONTROL2B(1)
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
 &           local_active_cell(row, col) .EQ. 0) THEN
           CALL PUSHCONTROL1B(0)
         ELSE
+          CALL PUSHINTEGER4(k)
           k = mesh%rowcol_to_ind_ac(row, col)
-          CALL PUSHREAL4(ac_qt(k))
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4(perc)
+            CALL PUSHREAL4(pr)
+            CALL PUSHREAL4(ac_hp(k))
+            CALL GR_PRODUCTION(output_layer(1, k), output_layer(2, k), &
+&                        pn(k), en(k), ac_cp(k), beta, ac_hp(k), pr, &
+&                        perc)
+            CALL GR_EXCHANGE(output_layer(4, k), ac_kexc(k), ac_ht(k), l&
+&                     )
+            CALL PUSHCONTROL1B(1)
+          ELSE
+            CALL PUSHREAL4(pr)
+            pr = 0._sp
+            CALL PUSHREAL4(perc)
+            perc = 0._sp
+            l = 0._sp
+            CALL PUSHCONTROL1B(0)
+          END IF
+          CALL PUSHREAL4(prr)
+          prr = 0.9_sp*(1._sp-output_layer(3, k)**2)*(pr+perc) + l
+          prd = (0.1_sp+0.9_sp*output_layer(3, k)**2)*(pr+perc)
           CALL PUSHREAL4(ac_ht(k))
-          CALL PUSHREAL4(ac_hp(k))
-          CALL GR_PRODUCTION_TRANSFER_ODE(pn(k), en(k), ac_cp(k), ac_ct(&
-&                                   k), ac_kexc(k), ac_hp(k), ac_ht(k), &
-&                                   ac_qt(k), l)
-! Transform from mm/dt to m3/s
+          CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
+&                    qr)
+          IF (0._sp .LT. prd + l) THEN
+            CALL PUSHCONTROL1B(0)
+          ELSE
+            CALL PUSHCONTROL1B(1)
+          END IF
           CALL PUSHCONTROL1B(1)
         END IF
       END DO
     END DO
+    output_layer_b = 0.0_4
     en_b = 0.0_4
     pn_b = 0.0_4
     DO col=mesh%ncol,1,-1
       DO row=mesh%nrow,1,-1
         CALL POPCONTROL1B(branch)
         IF (branch .NE. 0) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
           ac_qt_b(k) = mesh%dx(row, col)*1e-3_sp*mesh%dy(row, col)*&
 &           ac_qt_b(k)/setup%dt
-          CALL POPREAL4(ac_hp(k))
-          CALL POPREAL4(ac_ht(k))
-          CALL POPREAL4(ac_qt(k))
-          CALL GR_PRODUCTION_TRANSFER_ODE_B(pn(k), pn_b(k), en(k), en_b(&
-&                                     k), ac_cp(k), ac_cp_b(k), ac_ct(k)&
-&                                     , ac_ct_b(k), ac_kexc(k), &
-&                                     ac_kexc_b(k), ac_hp(k), ac_hp_b(k)&
-&                                     , ac_ht(k), ac_ht_b(k), ac_qt(k), &
-&                                     ac_qt_b(k), l)
+          qr_b = ac_qt_b(k)
+          qd_b = ac_qt_b(k)
           ac_qt_b(k) = 0.0_4
-        END IF
-      END DO
-    END DO
-    ac_prcp_b = 0.0_4
-    DO col=mesh%ncol,1,-1
-      DO row=mesh%nrow,1,-1
-        CALL POPCONTROL2B(branch)
-        IF (branch .NE. 0) THEN
-          IF (branch .EQ. 1) THEN
-            k = mesh%rowcol_to_ind_ac(row, col)
-            en_b(k) = 0.0_4
-            pn_b(k) = 0.0_4
+          CALL POPCONTROL1B(branch)
+          IF (branch .EQ. 0) THEN
+            prd_b = qd_b
+            l_b = qd_b
           ELSE
-            k = mesh%rowcol_to_ind_ac(row, col)
-            CALL POPREAL4(ac_hi(k))
-            CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
-&                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
+            l_b = 0.0_4
+            prd_b = 0.0_4
+          END IF
+          CALL POPREAL4(ac_ht(k))
+          CALL GR_TRANSFER_B(5._sp, ac_prcp(k), prr, prr_b, ac_ct(k), &
+&                      ac_ct_b(k), ac_ht(k), ac_ht_b(k), qr, qr_b)
+          output_layer_b(3, k) = output_layer_b(3, k) + 2*output_layer(3&
+&           , k)*0.9_sp*(pr+perc)*prd_b - 2*output_layer(3, k)*(pr+perc)&
+&           *0.9_sp*prr_b
+          temp_b = (0.9_sp*output_layer(3, k)**2+0.1_sp)*prd_b
+          pr_b = temp_b
+          perc_b = temp_b
+          CALL POPREAL4(prr)
+          temp_b = (1._sp-output_layer(3, k)**2)*0.9_sp*prr_b
+          l_b = l_b + prr_b
+          pr_b = pr_b + temp_b
+          perc_b = perc_b + temp_b
+          CALL POPCONTROL1B(branch)
+          IF (branch .EQ. 0) THEN
+            CALL POPREAL4(perc)
+            CALL POPREAL4(pr)
+          ELSE
+            CALL GR_EXCHANGE_B(output_layer(4, k), output_layer_b(4, k)&
+&                        , ac_kexc(k), ac_kexc_b(k), ac_ht(k), ac_ht_b(k&
+&                        ), l, l_b)
+            CALL POPREAL4(ac_hp(k))
+            CALL POPREAL4(pr)
+            CALL POPREAL4(perc)
+            CALL GR_PRODUCTION_B(output_layer(1, k), output_layer_b(1, k&
+&                          ), output_layer(2, k), output_layer_b(2, k), &
+&                          pn(k), pn_b(k), en(k), en_b(k), ac_cp(k), &
+&                          ac_cp_b(k), beta, ac_hp(k), ac_hp_b(k), pr, &
+&                          pr_b, perc, perc_b)
+          END IF
+          CALL POPINTEGER4(k)
+        END IF
+      END DO
+    END DO
+    DO col=mesh%ncol,1,-1
+      DO row=mesh%nrow,1,-1
+        CALL POPCONTROL2B(branch)
+        IF (branch .NE. 0) THEN
+          IF (branch .EQ. 1) THEN
+            k = mesh%rowcol_to_ind_ac(row, col)
+            output_layer_b(:, k) = 0.0_4
+          ELSE
+            k = mesh%rowcol_to_ind_ac(row, col)
+            CALL FORWARD_MLP_B(weight_1, weight_1_b, bias_1, bias_1_b, &
+&                        weight_2, weight_2_b, bias_2, bias_2_b, &
+&                        weight_3, weight_3_b, bias_3, bias_3_b, &
+&                        input_layer, input_layer_b, output_layer(:, k)&
+&                        , output_layer_b(:, k))
+            output_layer_b(:, k) = 0.0_4
+            CALL POPREAL4ARRAY(input_layer, setup%neurons(1))
+            ac_hp_b(k) = ac_hp_b(k) + input_layer_b(1)
+            ac_ht_b(k) = ac_ht_b(k) + input_layer_b(2)
+            pn_b(k) = pn_b(k) + input_layer_b(3)
+            en_b(k) = en_b(k) + input_layer_b(4)
+          END IF
+        END IF
+      END DO
+    END DO
+    ac_prcp_b = 0.0_4
+    DO col=mesh%ncol,1,-1
+      DO row=mesh%nrow,1,-1
+        CALL POPCONTROL2B(branch)
+        IF (branch .NE. 0) THEN
+          IF (branch .EQ. 1) THEN
+            k = mesh%rowcol_to_ind_ac(row, col)
+            en_b(k) = 0.0_4
+            pn_b(k) = 0.0_4
+          ELSE
+            k = mesh%rowcol_to_ind_ac(row, col)
+            CALL POPREAL4(ac_hi(k))
+            CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
+&                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
 &                            , pn(k), pn_b(k), en(k), en_b(k))
             pn_b(k) = 0.0_4
             en_b(k) = 0.0_4
@@ -15530,11 +15791,11 @@ SUBROUTINE GR4_ODE_TIME_STEP_B(setup, mesh, input_data, options, &
       END DO
     END DO
     ac_mlt_b = ac_mlt_b + ac_prcp_b
-  END SUBROUTINE GR4_ODE_TIME_STEP_B
+  END SUBROUTINE GR4_MLP_TIME_STEP_B
 
-  SUBROUTINE GR4_ODE_TIME_STEP(setup, mesh, input_data, options, returns&
-&   , time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, &
-&   ac_ht, ac_qt)
+  SUBROUTINE GR4_MLP_TIME_STEP(setup, mesh, input_data, options, returns&
+&   , time_step, weight_1, bias_1, weight_2, bias_2, weight_3, bias_3, &
+&   ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, ac_ht, ac_qt)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15542,19 +15803,34 @@ SUBROUTINE GR4_ODE_TIME_STEP(setup, mesh, input_data, options, returns&
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
 &   ac_kexc
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: l
+    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
+    INTRINSIC MAX
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'pet', ac_pet)
     ac_prcp = ac_prcp + ac_mlt
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
 ! Interception with OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
@@ -15571,34 +15847,66 @@ SUBROUTINE GR4_ODE_TIME_STEP(setup, mesh, input_data, options, returns&
         END IF
       END DO
     END DO
-! Production and transfer without OPENMP
+! Forward MLP without OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
 &           local_active_cell(row, col) .EQ. 0)) THEN
           k = mesh%rowcol_to_ind_ac(row, col)
-          CALL GR_PRODUCTION_TRANSFER_ODE(pn(k), en(k), ac_cp(k), ac_ct(&
-&                                   k), ac_kexc(k), ac_hp(k), ac_ht(k), &
-&                                   ac_qt(k), l)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
+&                      weight_3, bias_3, input_layer, output_layer(:, k)&
+&                     )
+          ELSE
+            output_layer(:, k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_PRODUCTION(output_layer(1, k), output_layer(2, k), &
+&                        pn(k), en(k), ac_cp(k), beta, ac_hp(k), pr, &
+&                        perc)
+            CALL GR_EXCHANGE(output_layer(4, k), ac_kexc(k), ac_ht(k), l&
+&                     )
+          ELSE
+            pr = 0._sp
+            perc = 0._sp
+            l = 0._sp
+          END IF
+          prr = 0.9_sp*(1._sp-output_layer(3, k)**2)*(pr+perc) + l
+          prd = (0.1_sp+0.9_sp*output_layer(3, k)**2)*(pr+perc)
+          CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
+&                    qr)
+          IF (0._sp .LT. prd + l) THEN
+            qd = prd + l
+          ELSE
+            qd = 0._sp
+          END IF
+          ac_qt(k) = qr + qd
 ! Transform from mm/dt to m3/s
           ac_qt(k) = ac_qt(k)*1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col&
 &           )/setup%dt
         END IF
       END DO
     END DO
-  END SUBROUTINE GR4_ODE_TIME_STEP
+  END SUBROUTINE GR4_MLP_TIME_STEP
 
-!  Differentiation of gr4_ode_mlp_time_step in forward (tangent) mode (with options fixinterface noISIZE context):
+!  Differentiation of gr4_ode_time_step in forward (tangent) mode (with options fixinterface noISIZE context):
 !   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-  SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
-&   returns, time_step, weight_1, weight_1_d, bias_1, bias_1_d, weight_2&
-&   , weight_2_d, bias_2, bias_2_d, weight_3, weight_3_d, bias_3, &
-&   bias_3_d, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
-&   ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d, ac_ht, &
-&   ac_ht_d, ac_qt, ac_qt_d)
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt
+  SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d&
+&   , ac_ct, ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d&
+&   , ac_ht, ac_ht_d, ac_qt, ac_qt_d)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15606,24 +15914,6 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1_d
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1_d
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2_d
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2_d
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3_d
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
@@ -15635,12 +15925,6 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
 &   ac_ht_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_d
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_d
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer_d
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d, pn_d, en_d
     INTEGER :: row, col, k, time_step_returns
@@ -15673,43 +15957,18 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-    output_layer_d = 0.0_4
-! Forward MLP without OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            input_layer_d(:) = (/ac_hp_d(k), ac_ht_d(k), pn_d(k), en_d(k&
-&             )/)
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP_D(weight_1, weight_1_d, bias_1, bias_1_d, &
-&                        weight_2, weight_2_d, bias_2, bias_2_d, &
-&                        weight_3, weight_3_d, bias_3, bias_3_d, &
-&                        input_layer, input_layer_d, output_layer(:, k)&
-&                        , output_layer_d(:, k))
-          ELSE
-            output_layer_d(:, k) = 0.0_4
-            output_layer(:, k) = 0._sp
-          END IF
-        END IF
-      END DO
-    END DO
-! Production and transfer with OPENMP
+! Production and transfer without OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
 &           local_active_cell(row, col) .EQ. 0)) THEN
           k = mesh%rowcol_to_ind_ac(row, col)
-          CALL GR_PRODUCTION_TRANSFER_ODE_MLP_D(output_layer(:, k), &
-&                                         output_layer_d(:, k), pn(k), &
-&                                         pn_d(k), en(k), en_d(k), ac_cp&
-&                                         (k), ac_cp_d(k), ac_ct(k), &
-&                                         ac_ct_d(k), ac_kexc(k), &
-&                                         ac_kexc_d(k), ac_hp(k), &
-&                                         ac_hp_d(k), ac_ht(k), ac_ht_d(&
-&                                         k), ac_qt(k), ac_qt_d(k), l)
+          CALL GR_PRODUCTION_TRANSFER_ODE_D(pn(k), pn_d(k), en(k), en_d(&
+&                                     k), ac_cp(k), ac_cp_d(k), ac_ct(k)&
+&                                     , ac_ct_d(k), ac_kexc(k), &
+&                                     ac_kexc_d(k), ac_hp(k), ac_hp_d(k)&
+&                                     , ac_ht(k), ac_ht_d(k), ac_qt(k), &
+&                                     ac_qt_d(k), l)
 ! Transform from mm/dt to m3/s
           temp = 1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)
           ac_qt_d(k) = temp*ac_qt_d(k)/setup%dt
@@ -15717,21 +15976,17 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-  END SUBROUTINE GR4_ODE_MLP_TIME_STEP_D
+  END SUBROUTINE GR4_ODE_TIME_STEP_D
 
-!  Differentiation of gr4_ode_mlp_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context):
-!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-  SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
-&   returns, time_step, weight_1, weight_1_b, bias_1, bias_1_b, weight_2&
-&   , weight_2_b, bias_2, bias_2_b, weight_3, weight_3_b, bias_3, &
-&   bias_3_b, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
-&   ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b, ac_ht, &
-&   ac_ht_b, ac_qt, ac_qt_b)
+!  Differentiation of gr4_ode_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt
+  SUBROUTINE GR4_ODE_TIME_STEP_B(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b&
+&   , ac_ct, ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b&
+&   , ac_ht, ac_ht_b, ac_qt, ac_qt_b)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15739,24 +15994,6 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)) :: &
-&   weight_1_b
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
-    REAL(sp), DIMENSION(setup%neurons(2)) :: bias_1_b
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)) :: &
-&   weight_2_b
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
-    REAL(sp), DIMENSION(setup%neurons(3)) :: bias_2_b
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)) :: &
-&   weight_3_b
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
-    REAL(sp), DIMENSION(setup%neurons(4)) :: bias_3_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac) :: ac_mlt_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
@@ -15768,12 +16005,6 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
 &   ac_ht_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_b
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_b
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer_b
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b, pn_b, en_b
     INTEGER :: row, col, k, time_step_returns
@@ -15805,29 +16036,7 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-! Forward MLP without OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0) THEN
-          CALL PUSHCONTROL2B(0)
-        ELSE
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            CALL PUSHREAL4ARRAY(input_layer, setup%neurons(1))
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
-&                      weight_3, bias_3, input_layer, output_layer(:, k)&
-&                     )
-            CALL PUSHCONTROL2B(2)
-          ELSE
-            output_layer(:, k) = 0._sp
-            CALL PUSHCONTROL2B(1)
-          END IF
-        END IF
-      END DO
-    END DO
-! Production and transfer with OPENMP
+! Production and transfer without OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -15838,16 +16047,14 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
           CALL PUSHREAL4(ac_qt(k))
           CALL PUSHREAL4(ac_ht(k))
           CALL PUSHREAL4(ac_hp(k))
-          CALL GR_PRODUCTION_TRANSFER_ODE_MLP(output_layer(:, k), pn(k)&
-&                                       , en(k), ac_cp(k), ac_ct(k), &
-&                                       ac_kexc(k), ac_hp(k), ac_ht(k), &
-&                                       ac_qt(k), l)
+          CALL GR_PRODUCTION_TRANSFER_ODE(pn(k), en(k), ac_cp(k), ac_ct(&
+&                                   k), ac_kexc(k), ac_hp(k), ac_ht(k), &
+&                                   ac_qt(k), l)
 ! Transform from mm/dt to m3/s
           CALL PUSHCONTROL1B(1)
         END IF
       END DO
     END DO
-    output_layer_b = 0.0_4
     en_b = 0.0_4
     pn_b = 0.0_4
     DO col=mesh%ncol,1,-1
@@ -15860,70 +16067,740 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
           CALL POPREAL4(ac_hp(k))
           CALL POPREAL4(ac_ht(k))
           CALL POPREAL4(ac_qt(k))
-          CALL GR_PRODUCTION_TRANSFER_ODE_MLP_B(output_layer(:, k), &
-&                                         output_layer_b(:, k), pn(k), &
-&                                         pn_b(k), en(k), en_b(k), ac_cp&
-&                                         (k), ac_cp_b(k), ac_ct(k), &
-&                                         ac_ct_b(k), ac_kexc(k), &
-&                                         ac_kexc_b(k), ac_hp(k), &
-&                                         ac_hp_b(k), ac_ht(k), ac_ht_b(&
-&                                         k), ac_qt(k), ac_qt_b(k), l)
+          CALL GR_PRODUCTION_TRANSFER_ODE_B(pn(k), pn_b(k), en(k), en_b(&
+&                                     k), ac_cp(k), ac_cp_b(k), ac_ct(k)&
+&                                     , ac_ct_b(k), ac_kexc(k), &
+&                                     ac_kexc_b(k), ac_hp(k), ac_hp_b(k)&
+&                                     , ac_ht(k), ac_ht_b(k), ac_qt(k), &
+&                                     ac_qt_b(k), l)
           ac_qt_b(k) = 0.0_4
         END IF
       END DO
     END DO
+    ac_prcp_b = 0.0_4
     DO col=mesh%ncol,1,-1
       DO row=mesh%nrow,1,-1
         CALL POPCONTROL2B(branch)
         IF (branch .NE. 0) THEN
           IF (branch .EQ. 1) THEN
             k = mesh%rowcol_to_ind_ac(row, col)
-            output_layer_b(:, k) = 0.0_4
+            en_b(k) = 0.0_4
+            pn_b(k) = 0.0_4
           ELSE
             k = mesh%rowcol_to_ind_ac(row, col)
-            CALL FORWARD_MLP_B(weight_1, weight_1_b, bias_1, bias_1_b, &
-&                        weight_2, weight_2_b, bias_2, bias_2_b, &
-&                        weight_3, weight_3_b, bias_3, bias_3_b, &
-&                        input_layer, input_layer_b, output_layer(:, k)&
-&                        , output_layer_b(:, k))
-            output_layer_b(:, k) = 0.0_4
-            CALL POPREAL4ARRAY(input_layer, setup%neurons(1))
-            ac_hp_b(k) = ac_hp_b(k) + input_layer_b(1)
-            ac_ht_b(k) = ac_ht_b(k) + input_layer_b(2)
-            pn_b(k) = pn_b(k) + input_layer_b(3)
-            en_b(k) = en_b(k) + input_layer_b(4)
+            CALL POPREAL4(ac_hi(k))
+            CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
+&                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
+&                            , pn(k), pn_b(k), en(k), en_b(k))
+            pn_b(k) = 0.0_4
+            en_b(k) = 0.0_4
           END IF
         END IF
       END DO
     END DO
-    ac_prcp_b = 0.0_4
-    DO col=mesh%ncol,1,-1
-      DO row=mesh%nrow,1,-1
-        CALL POPCONTROL2B(branch)
-        IF (branch .NE. 0) THEN
-          IF (branch .EQ. 1) THEN
-            k = mesh%rowcol_to_ind_ac(row, col)
-            en_b(k) = 0.0_4
-            pn_b(k) = 0.0_4
+    ac_mlt_b = ac_mlt_b + ac_prcp_b
+  END SUBROUTINE GR4_ODE_TIME_STEP_B
+
+  SUBROUTINE GR4_ODE_TIME_STEP(setup, mesh, input_data, options, returns&
+&   , time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, &
+&   ac_ht, ac_qt)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: l
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp = ac_prcp + ac_mlt
+! Interception with OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
+&                          k), pn(k), en(k))
+          ELSE
+            pn(k) = 0._sp
+            en(k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer without OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          CALL GR_PRODUCTION_TRANSFER_ODE(pn(k), en(k), ac_cp(k), ac_ct(&
+&                                   k), ac_kexc(k), ac_hp(k), ac_ht(k), &
+&                                   ac_qt(k), l)
+! Transform from mm/dt to m3/s
+          ac_qt(k) = ac_qt(k)*1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col&
+&           )/setup%dt
+        END IF
+      END DO
+    END DO
+  END SUBROUTINE GR4_ODE_TIME_STEP
+
+!  Differentiation of gr4_ode_mlp_time_step in forward (tangent) mode (with options fixinterface noISIZE context):
+!   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+  SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
+&   returns, time_step, weight_1, weight_1_d, bias_1, bias_1_d, weight_2&
+&   , weight_2_d, bias_2, bias_2_d, weight_3, weight_3_d, bias_3, &
+&   bias_3_d, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
+&   ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d, ac_ht, &
+&   ac_ht_d, ac_qt, ac_qt_d)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1_d
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1_d
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2_d
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2_d
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3_d
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci_d, ac_cp_d, &
+&   ac_ct_d, ac_kexc_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_d, ac_hp_d, &
+&   ac_ht_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_d
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_d
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer_d
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d, pn_d, en_d
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: l
+    REAL(sp) :: temp
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp_d = ac_mlt_d
+    ac_prcp = ac_prcp + ac_mlt
+    en_d = 0.0_4
+    pn_d = 0.0_4
+! Interception with OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_INTERCEPTION_D(ac_prcp(k), ac_prcp_d(k), ac_pet(k), &
+&                            ac_ci(k), ac_ci_d(k), ac_hi(k), ac_hi_d(k)&
+&                            , pn(k), pn_d(k), en(k), en_d(k))
+          ELSE
+            pn_d(k) = 0.0_4
+            pn(k) = 0._sp
+            en_d(k) = 0.0_4
+            en(k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+    output_layer_d = 0.0_4
+! Forward MLP without OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            input_layer_d(:) = (/ac_hp_d(k), ac_ht_d(k), pn_d(k), en_d(k&
+&             )/)
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP_D(weight_1, weight_1_d, bias_1, bias_1_d, &
+&                        weight_2, weight_2_d, bias_2, bias_2_d, &
+&                        weight_3, weight_3_d, bias_3, bias_3_d, &
+&                        input_layer, input_layer_d, output_layer(:, k)&
+&                        , output_layer_d(:, k))
+          ELSE
+            output_layer_d(:, k) = 0.0_4
+            output_layer(:, k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          CALL GR_PRODUCTION_TRANSFER_ODE_MLP_D(output_layer(:, k), &
+&                                         output_layer_d(:, k), pn(k), &
+&                                         pn_d(k), en(k), en_d(k), ac_cp&
+&                                         (k), ac_cp_d(k), ac_ct(k), &
+&                                         ac_ct_d(k), ac_kexc(k), &
+&                                         ac_kexc_d(k), ac_hp(k), &
+&                                         ac_hp_d(k), ac_ht(k), ac_ht_d(&
+&                                         k), ac_qt(k), ac_qt_d(k), l)
+! Transform from mm/dt to m3/s
+          temp = 1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)
+          ac_qt_d(k) = temp*ac_qt_d(k)/setup%dt
+          ac_qt(k) = temp*(ac_qt(k)/setup%dt)
+        END IF
+      END DO
+    END DO
+  END SUBROUTINE GR4_ODE_MLP_TIME_STEP_D
+
+!  Differentiation of gr4_ode_mlp_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+  SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
+&   returns, time_step, weight_1, weight_1_b, bias_1, bias_1_b, weight_2&
+&   , weight_2_b, bias_2, bias_2_b, weight_3, weight_3_b, bias_3, &
+&   bias_3_b, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
+&   ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b, ac_ht, &
+&   ac_ht_b, ac_qt, ac_qt_b)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)) :: &
+&   weight_1_b
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(2)) :: bias_1_b
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)) :: &
+&   weight_2_b
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(3)) :: bias_2_b
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)) :: &
+&   weight_3_b
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
+    REAL(sp), DIMENSION(setup%neurons(4)) :: bias_3_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac) :: ac_mlt_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac) :: ac_ci_b, ac_cp_b, ac_ct_b, &
+&   ac_kexc_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_b, ac_hp_b, &
+&   ac_ht_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_b
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_b
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer_b
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b, pn_b, en_b
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: l
+    INTEGER :: branch
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp = ac_prcp + ac_mlt
+! Interception with OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0) THEN
+          CALL PUSHCONTROL2B(0)
+        ELSE
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4(ac_hi(k))
+            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
+&                          k), pn(k), en(k))
+            CALL PUSHCONTROL2B(2)
+          ELSE
+            pn(k) = 0._sp
+            en(k) = 0._sp
+            CALL PUSHCONTROL2B(1)
+          END IF
+        END IF
+      END DO
+    END DO
+! Forward MLP without OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0) THEN
+          CALL PUSHCONTROL2B(0)
+        ELSE
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4ARRAY(input_layer, setup%neurons(1))
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
+&                      weight_3, bias_3, input_layer, output_layer(:, k)&
+&                     )
+            CALL PUSHCONTROL2B(2)
+          ELSE
+            output_layer(:, k) = 0._sp
+            CALL PUSHCONTROL2B(1)
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0) THEN
+          CALL PUSHCONTROL1B(0)
+        ELSE
+          k = mesh%rowcol_to_ind_ac(row, col)
+          CALL PUSHREAL4(ac_qt(k))
+          CALL PUSHREAL4(ac_ht(k))
+          CALL PUSHREAL4(ac_hp(k))
+          CALL GR_PRODUCTION_TRANSFER_ODE_MLP(output_layer(:, k), pn(k)&
+&                                       , en(k), ac_cp(k), ac_ct(k), &
+&                                       ac_kexc(k), ac_hp(k), ac_ht(k), &
+&                                       ac_qt(k), l)
+! Transform from mm/dt to m3/s
+          CALL PUSHCONTROL1B(1)
+        END IF
+      END DO
+    END DO
+    output_layer_b = 0.0_4
+    en_b = 0.0_4
+    pn_b = 0.0_4
+    DO col=mesh%ncol,1,-1
+      DO row=mesh%nrow,1,-1
+        CALL POPCONTROL1B(branch)
+        IF (branch .NE. 0) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          ac_qt_b(k) = mesh%dx(row, col)*1e-3_sp*mesh%dy(row, col)*&
+&           ac_qt_b(k)/setup%dt
+          CALL POPREAL4(ac_hp(k))
+          CALL POPREAL4(ac_ht(k))
+          CALL POPREAL4(ac_qt(k))
+          CALL GR_PRODUCTION_TRANSFER_ODE_MLP_B(output_layer(:, k), &
+&                                         output_layer_b(:, k), pn(k), &
+&                                         pn_b(k), en(k), en_b(k), ac_cp&
+&                                         (k), ac_cp_b(k), ac_ct(k), &
+&                                         ac_ct_b(k), ac_kexc(k), &
+&                                         ac_kexc_b(k), ac_hp(k), &
+&                                         ac_hp_b(k), ac_ht(k), ac_ht_b(&
+&                                         k), ac_qt(k), ac_qt_b(k), l)
+          ac_qt_b(k) = 0.0_4
+        END IF
+      END DO
+    END DO
+    DO col=mesh%ncol,1,-1
+      DO row=mesh%nrow,1,-1
+        CALL POPCONTROL2B(branch)
+        IF (branch .NE. 0) THEN
+          IF (branch .EQ. 1) THEN
+            k = mesh%rowcol_to_ind_ac(row, col)
+            output_layer_b(:, k) = 0.0_4
+          ELSE
+            k = mesh%rowcol_to_ind_ac(row, col)
+            CALL FORWARD_MLP_B(weight_1, weight_1_b, bias_1, bias_1_b, &
+&                        weight_2, weight_2_b, bias_2, bias_2_b, &
+&                        weight_3, weight_3_b, bias_3, bias_3_b, &
+&                        input_layer, input_layer_b, output_layer(:, k)&
+&                        , output_layer_b(:, k))
+            output_layer_b(:, k) = 0.0_4
+            CALL POPREAL4ARRAY(input_layer, setup%neurons(1))
+            ac_hp_b(k) = ac_hp_b(k) + input_layer_b(1)
+            ac_ht_b(k) = ac_ht_b(k) + input_layer_b(2)
+            pn_b(k) = pn_b(k) + input_layer_b(3)
+            en_b(k) = en_b(k) + input_layer_b(4)
+          END IF
+        END IF
+      END DO
+    END DO
+    ac_prcp_b = 0.0_4
+    DO col=mesh%ncol,1,-1
+      DO row=mesh%nrow,1,-1
+        CALL POPCONTROL2B(branch)
+        IF (branch .NE. 0) THEN
+          IF (branch .EQ. 1) THEN
+            k = mesh%rowcol_to_ind_ac(row, col)
+            en_b(k) = 0.0_4
+            pn_b(k) = 0.0_4
+          ELSE
+            k = mesh%rowcol_to_ind_ac(row, col)
+            CALL POPREAL4(ac_hi(k))
+            CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
+&                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
+&                            , pn(k), pn_b(k), en(k), en_b(k))
+            pn_b(k) = 0.0_4
+            en_b(k) = 0.0_4
+          END IF
+        END IF
+      END DO
+    END DO
+    ac_mlt_b = ac_mlt_b + ac_prcp_b
+  END SUBROUTINE GR4_ODE_MLP_TIME_STEP_B
+
+  SUBROUTINE GR4_ODE_MLP_TIME_STEP(setup, mesh, input_data, options, &
+&   returns, time_step, weight_1, bias_1, weight_2, bias_2, weight_3, &
+&   bias_3, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, ac_ht, &
+&   ac_qt)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: l
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp = ac_prcp + ac_mlt
+! Interception with OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
+&                          k), pn(k), en(k))
+          ELSE
+            pn(k) = 0._sp
+            en(k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Forward MLP without OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
+&                      weight_3, bias_3, input_layer, output_layer(:, k)&
+&                     )
+          ELSE
+            output_layer(:, k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          CALL GR_PRODUCTION_TRANSFER_ODE_MLP(output_layer(:, k), pn(k)&
+&                                       , en(k), ac_cp(k), ac_ct(k), &
+&                                       ac_kexc(k), ac_hp(k), ac_ht(k), &
+&                                       ac_qt(k), l)
+! Transform from mm/dt to m3/s
+          ac_qt(k) = ac_qt(k)*1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col&
+&           )/setup%dt
+        END IF
+      END DO
+    END DO
+  END SUBROUTINE GR4_ODE_MLP_TIME_STEP
+
+!  Differentiation of gr5_time_step in forward (tangent) mode (with options fixinterface noISIZE context):
+!   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
+  SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
+&   time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
+&   ac_ct_d, ac_kexc, ac_kexc_d, ac_aexc, ac_aexc_d, ac_hi, ac_hi_d, &
+&   ac_hp, ac_hp_d, ac_ht, ac_ht_d, ac_qt, ac_qt_d)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc, ac_aexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci_d, ac_cp_d, &
+&   ac_ct_d, ac_kexc_d, ac_aexc_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_d, ac_hp_d, &
+&   ac_ht_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_d
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: pn_d, en_d, pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d
+    INTRINSIC MAX
+    REAL(sp) :: temp
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp_d = ac_mlt_d
+    ac_prcp = ac_prcp + ac_mlt
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_INTERCEPTION_D(ac_prcp(k), ac_prcp_d(k), ac_pet(k), &
+&                            ac_ci(k), ac_ci_d(k), ac_hi(k), ac_hi_d(k)&
+&                            , pn, pn_d, en, en_d)
+            CALL GR_PRODUCTION_D(0._sp, 0.0_4, 0._sp, 0.0_4, pn, pn_d, &
+&                          en, en_d, ac_cp(k), ac_cp_d(k), beta, ac_hp(k&
+&                          ), ac_hp_d(k), pr, pr_d, perc, perc_d)
+            CALL GR_THRESHOLD_EXCHANGE_D(ac_kexc(k), ac_kexc_d(k), &
+&                                  ac_aexc(k), ac_aexc_d(k), ac_ht(k), &
+&                                  ac_ht_d(k), l, l_d)
+          ELSE
+            pr = 0._sp
+            perc = 0._sp
+            l = 0._sp
+            l_d = 0.0_4
+            perc_d = 0.0_4
+            pr_d = 0.0_4
+          END IF
+          prr_d = 0.9_sp*(pr_d+perc_d) + l_d
+          prr = 0.9_sp*(pr+perc) + l
+          prd_d = 0.1_sp*(pr_d+perc_d)
+          prd = 0.1_sp*(pr+perc)
+          CALL GR_TRANSFER_D(5._sp, ac_prcp(k), prr, prr_d, ac_ct(k), &
+&                      ac_ct_d(k), ac_ht(k), ac_ht_d(k), qr, qr_d)
+          IF (0._sp .LT. prd + l) THEN
+            qd_d = prd_d + l_d
+            qd = prd + l
+          ELSE
+            qd = 0._sp
+            qd_d = 0.0_4
+          END IF
+          ac_qt_d(k) = qr_d + qd_d
+          ac_qt(k) = qr + qd
+! Transform from mm/dt to m3/s
+          temp = 1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)
+          ac_qt_d(k) = temp*ac_qt_d(k)/setup%dt
+          ac_qt(k) = temp*(ac_qt(k)/setup%dt)
+        END IF
+      END DO
+    END DO
+  END SUBROUTINE GR5_TIME_STEP_D
+
+!  Differentiation of gr5_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
+  SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
+&   time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
+&   ac_ct_b, ac_kexc, ac_kexc_b, ac_aexc, ac_aexc_b, ac_hi, ac_hi_b, &
+&   ac_hp, ac_hp_b, ac_ht, ac_ht_b, ac_qt, ac_qt_b)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac) :: ac_mlt_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc, ac_aexc
+    REAL(sp), DIMENSION(mesh%nac) :: ac_ci_b, ac_cp_b, ac_ct_b, &
+&   ac_kexc_b, ac_aexc_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_b, ac_hp_b, &
+&   ac_ht_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_b
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: pn_b, en_b, pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b
+    INTRINSIC MAX
+    REAL(sp) :: dummydiff_b
+    REAL(sp) :: dummydiff_b0
+    INTEGER :: branch
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp = ac_prcp + ac_mlt
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0) THEN
+          CALL PUSHCONTROL1B(0)
+        ELSE
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4(en)
+            CALL PUSHREAL4(pn)
+            CALL PUSHREAL4(ac_hi(k))
+            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
+&                          k), pn, en)
+            CALL PUSHREAL4(ac_hp(k))
+            CALL GR_PRODUCTION(0._sp, 0._sp, pn, en, ac_cp(k), beta, &
+&                        ac_hp(k), pr, perc)
+            CALL GR_THRESHOLD_EXCHANGE(ac_kexc(k), ac_aexc(k), ac_ht(k)&
+&                                , l)
+            CALL PUSHCONTROL1B(1)
+          ELSE
+            CALL PUSHCONTROL1B(0)
+            pr = 0._sp
+            perc = 0._sp
+            l = 0._sp
+          END IF
+          CALL PUSHREAL4(prr)
+          prr = 0.9_sp*(pr+perc) + l
+          prd = 0.1_sp*(pr+perc)
+          CALL PUSHREAL4(ac_ht(k))
+          CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
+&                    qr)
+          IF (0._sp .LT. prd + l) THEN
+            CALL PUSHCONTROL1B(0)
+          ELSE
+            CALL PUSHCONTROL1B(1)
+          END IF
+          CALL PUSHCONTROL1B(1)
+        END IF
+      END DO
+    END DO
+    ac_prcp_b = 0.0_4
+    DO col=mesh%ncol,1,-1
+      DO row=mesh%nrow,1,-1
+        CALL POPCONTROL1B(branch)
+        IF (branch .NE. 0) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          ac_qt_b(k) = mesh%dx(row, col)*1e-3_sp*mesh%dy(row, col)*&
+&           ac_qt_b(k)/setup%dt
+          qr_b = ac_qt_b(k)
+          qd_b = ac_qt_b(k)
+          ac_qt_b(k) = 0.0_4
+          CALL POPCONTROL1B(branch)
+          IF (branch .EQ. 0) THEN
+            prd_b = qd_b
+            l_b = qd_b
           ELSE
-            k = mesh%rowcol_to_ind_ac(row, col)
+            l_b = 0.0_4
+            prd_b = 0.0_4
+          END IF
+          CALL POPREAL4(ac_ht(k))
+          CALL GR_TRANSFER_B(5._sp, ac_prcp(k), prr, prr_b, ac_ct(k), &
+&                      ac_ct_b(k), ac_ht(k), ac_ht_b(k), qr, qr_b)
+          pr_b = 0.1_sp*prd_b + 0.9_sp*prr_b
+          perc_b = 0.1_sp*prd_b + 0.9_sp*prr_b
+          CALL POPREAL4(prr)
+          l_b = l_b + prr_b
+          CALL POPCONTROL1B(branch)
+          IF (branch .NE. 0) THEN
+            CALL GR_THRESHOLD_EXCHANGE_B(ac_kexc(k), ac_kexc_b(k), &
+&                                  ac_aexc(k), ac_aexc_b(k), ac_ht(k), &
+&                                  ac_ht_b(k), l, l_b)
+            CALL POPREAL4(ac_hp(k))
+            pn_b = 0.0_4
+            en_b = 0.0_4
+            CALL GR_PRODUCTION_B(0._sp, dummydiff_b, 0._sp, dummydiff_b0&
+&                          , pn, pn_b, en, en_b, ac_cp(k), ac_cp_b(k), &
+&                          beta, ac_hp(k), ac_hp_b(k), pr, pr_b, perc, &
+&                          perc_b)
             CALL POPREAL4(ac_hi(k))
+            CALL POPREAL4(pn)
+            CALL POPREAL4(en)
             CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
 &                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
-&                            , pn(k), pn_b(k), en(k), en_b(k))
-            pn_b(k) = 0.0_4
-            en_b(k) = 0.0_4
+&                            , pn, pn_b, en, en_b)
           END IF
         END IF
       END DO
     END DO
     ac_mlt_b = ac_mlt_b + ac_prcp_b
-  END SUBROUTINE GR4_ODE_MLP_TIME_STEP_B
+  END SUBROUTINE GR5_TIME_STEP_B
 
-  SUBROUTINE GR4_ODE_MLP_TIME_STEP(setup, mesh, input_data, options, &
-&   returns, time_step, weight_1, bias_1, weight_2, bias_2, weight_3, &
-&   bias_3, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, ac_ht, &
-&   ac_qt)
+  SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
+&   time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_aexc, ac_hi, &
+&   ac_hp, ac_ht, ac_qt)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15931,32 +16808,22 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
-&   ac_kexc
+&   ac_kexc, ac_aexc
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: l
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
+    INTRINSIC MAX
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'pet', ac_pet)
     ac_prcp = ac_prcp + ac_mlt
-! Interception with OPENMP
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -15964,57 +16831,43 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP(setup, mesh, input_data, options, &
           k = mesh%rowcol_to_ind_ac(row, col)
           IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
             CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
-&                          k), pn(k), en(k))
+&                          k), pn, en)
+            CALL GR_PRODUCTION(0._sp, 0._sp, pn, en, ac_cp(k), beta, &
+&                        ac_hp(k), pr, perc)
+            CALL GR_THRESHOLD_EXCHANGE(ac_kexc(k), ac_aexc(k), ac_ht(k)&
+&                                , l)
           ELSE
-            pn(k) = 0._sp
-            en(k) = 0._sp
+            pr = 0._sp
+            perc = 0._sp
+            l = 0._sp
           END IF
-        END IF
-      END DO
-    END DO
-! Forward MLP without OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
-&                      weight_3, bias_3, input_layer, output_layer(:, k)&
-&                     )
+          prr = 0.9_sp*(pr+perc) + l
+          prd = 0.1_sp*(pr+perc)
+          CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
+&                    qr)
+          IF (0._sp .LT. prd + l) THEN
+            qd = prd + l
           ELSE
-            output_layer(:, k) = 0._sp
+            qd = 0._sp
           END IF
-        END IF
-      END DO
-    END DO
-! Production and transfer with OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          CALL GR_PRODUCTION_TRANSFER_ODE_MLP(output_layer(:, k), pn(k)&
-&                                       , en(k), ac_cp(k), ac_ct(k), &
-&                                       ac_kexc(k), ac_hp(k), ac_ht(k), &
-&                                       ac_qt(k), l)
+          ac_qt(k) = qr + qd
 ! Transform from mm/dt to m3/s
           ac_qt(k) = ac_qt(k)*1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col&
 &           )/setup%dt
         END IF
       END DO
     END DO
-  END SUBROUTINE GR4_ODE_MLP_TIME_STEP
+  END SUBROUTINE GR5_TIME_STEP
 
-!  Differentiation of gr5_time_step in forward (tangent) mode (with options fixinterface noISIZE context):
+!  Differentiation of gr5_ri_time_step in forward (tangent) mode (with options fixinterface noISIZE context):
 !   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
-  SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
-&   time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
-&   ac_ct_d, ac_kexc, ac_kexc_d, ac_aexc, ac_aexc_d, ac_hi, ac_hi_d, &
-&   ac_hp, ac_hp_d, ac_ht, ac_ht_d, ac_qt, ac_qt_d)
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt ac_aexc
+  SUBROUTINE GR5_RI_TIME_STEP_D(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d&
+&   , ac_ct, ac_ct_d, ac_alpha1, ac_alpha1_d, ac_alpha2, ac_alpha2_d, &
+&   ac_kexc, ac_kexc_d, ac_aexc, ac_aexc_d, ac_hi, ac_hi_d, ac_hp, &
+&   ac_hp_d, ac_ht, ac_ht_d, ac_qt, ac_qt_d)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -16028,6 +16881,9 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
 &   ac_kexc, ac_aexc
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci_d, ac_cp_d, &
 &   ac_ct_d, ac_kexc_d, ac_aexc_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1_d, &
+&   ac_alpha2_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_d, ac_hp_d, &
 &   ac_ht_d
@@ -16036,8 +16892,10 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
-    REAL(sp) :: pn_d, en_d, pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    REAL(sp) :: pn_d, en_d, pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d&
+&   , split_d
+    INTRINSIC TANH
     INTRINSIC MAX
     REAL(sp) :: temp
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
@@ -16048,6 +16906,7 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
     ac_prcp = ac_prcp + ac_mlt
 ! Beta percolation parameter is time step dependent
     beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
+    pn_d = 0.0_4
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -16057,9 +16916,10 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
             CALL GR_INTERCEPTION_D(ac_prcp(k), ac_prcp_d(k), ac_pet(k), &
 &                            ac_ci(k), ac_ci_d(k), ac_hi(k), ac_hi_d(k)&
 &                            , pn, pn_d, en, en_d)
-            CALL GR_PRODUCTION_D(0._sp, 0.0_4, 0._sp, 0.0_4, pn, pn_d, &
-&                          en, en_d, ac_cp(k), ac_cp_d(k), beta, ac_hp(k&
-&                          ), ac_hp_d(k), pr, pr_d, perc, perc_d)
+            CALL GR_RI_PRODUCTION_D(pn, pn_d, en, en_d, ac_cp(k), &
+&                             ac_cp_d(k), beta, ac_alpha1(k), &
+&                             ac_alpha1_d(k), ac_hp(k), ac_hp_d(k), pr, &
+&                             pr_d, perc, perc_d, setup%dt)
             CALL GR_THRESHOLD_EXCHANGE_D(ac_kexc(k), ac_kexc_d(k), &
 &                                  ac_aexc(k), ac_aexc_d(k), ac_ht(k), &
 &                                  ac_ht_d(k), l, l_d)
@@ -16071,10 +16931,13 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
             perc_d = 0.0_4
             pr_d = 0.0_4
           END IF
-          prr_d = 0.9_sp*(pr_d+perc_d) + l_d
-          prr = 0.9_sp*(pr+perc) + l
-          prd_d = 0.1_sp*(pr_d+perc_d)
-          prd = 0.1_sp*(pr+perc)
+          split_d = 0.9_sp*2*TANH(ac_alpha2(k)*pn)*(1.0-TANH(ac_alpha2(k&
+&           )*pn)**2)*(pn*ac_alpha2_d(k)+ac_alpha2(k)*pn_d)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
+          prr_d = (1._sp-split)*(pr_d+perc_d) - (pr+perc)*split_d + l_d
+          prr = (1._sp-split)*(pr+perc) + l
+          prd_d = (pr+perc)*split_d + split*(pr_d+perc_d)
+          prd = split*(pr+perc)
           CALL GR_TRANSFER_D(5._sp, ac_prcp(k), prr, prr_d, ac_ct(k), &
 &                      ac_ct_d(k), ac_ht(k), ac_ht_d(k), qr, qr_d)
           IF (0._sp .LT. prd + l) THEN
@@ -16093,17 +16956,18 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
         END IF
       END DO
     END DO
-  END SUBROUTINE GR5_TIME_STEP_D
+  END SUBROUTINE GR5_RI_TIME_STEP_D
 
-!  Differentiation of gr5_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context):
-!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
-  SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
-&   time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
-&   ac_ct_b, ac_kexc, ac_kexc_b, ac_aexc, ac_aexc_b, ac_hi, ac_hi_b, &
-&   ac_hp, ac_hp_b, ac_ht, ac_ht_b, ac_qt, ac_qt_b)
+!  Differentiation of gr5_ri_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt ac_aexc
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt ac_aexc
+  SUBROUTINE GR5_RI_TIME_STEP_B(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b&
+&   , ac_ct, ac_ct_b, ac_alpha1, ac_alpha1_b, ac_alpha2, ac_alpha2_b, &
+&   ac_kexc, ac_kexc_b, ac_aexc, ac_aexc_b, ac_hi, ac_hi_b, ac_hp, &
+&   ac_hp_b, ac_ht, ac_ht_b, ac_qt, ac_qt_b)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -16117,6 +16981,8 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
 &   ac_kexc, ac_aexc
     REAL(sp), DIMENSION(mesh%nac) :: ac_ci_b, ac_cp_b, ac_ct_b, &
 &   ac_kexc_b, ac_aexc_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
+    REAL(sp), DIMENSION(mesh%nac) :: ac_alpha1_b, ac_alpha2_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_b, ac_hp_b, &
 &   ac_ht_b
@@ -16125,11 +16991,12 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
-    REAL(sp) :: pn_b, en_b, pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    REAL(sp) :: pn_b, en_b, pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b&
+&   , split_b
+    INTRINSIC TANH
     INTRINSIC MAX
-    REAL(sp) :: dummydiff_b
-    REAL(sp) :: dummydiff_b0
+    REAL(sp) :: temp_b
     INTEGER :: branch
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
@@ -16151,21 +17018,27 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
             CALL PUSHREAL4(ac_hi(k))
             CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
 &                          k), pn, en)
+            CALL PUSHREAL4(perc)
+            CALL PUSHREAL4(pr)
             CALL PUSHREAL4(ac_hp(k))
-            CALL GR_PRODUCTION(0._sp, 0._sp, pn, en, ac_cp(k), beta, &
-&                        ac_hp(k), pr, perc)
+            CALL GR_RI_PRODUCTION(pn, en, ac_cp(k), beta, ac_alpha1(k), &
+&                           ac_hp(k), pr, perc, setup%dt)
             CALL GR_THRESHOLD_EXCHANGE(ac_kexc(k), ac_aexc(k), ac_ht(k)&
 &                                , l)
             CALL PUSHCONTROL1B(1)
           ELSE
-            CALL PUSHCONTROL1B(0)
+            CALL PUSHREAL4(pr)
             pr = 0._sp
+            CALL PUSHREAL4(perc)
             perc = 0._sp
             l = 0._sp
+            CALL PUSHCONTROL1B(0)
           END IF
+          CALL PUSHREAL4(split)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
           CALL PUSHREAL4(prr)
-          prr = 0.9_sp*(pr+perc) + l
-          prd = 0.1_sp*(pr+perc)
+          prr = (1._sp-split)*(pr+perc) + l
+          prd = split*(pr+perc)
           CALL PUSHREAL4(ac_ht(k))
           CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
 &                    qr)
@@ -16179,6 +17052,7 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
       END DO
     END DO
     ac_prcp_b = 0.0_4
+    pn_b = 0.0_4
     DO col=mesh%ncol,1,-1
       DO row=mesh%nrow,1,-1
         CALL POPCONTROL1B(branch)
@@ -16200,38 +17074,48 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
           CALL POPREAL4(ac_ht(k))
           CALL GR_TRANSFER_B(5._sp, ac_prcp(k), prr, prr_b, ac_ct(k), &
 &                      ac_ct_b(k), ac_ht(k), ac_ht_b(k), qr, qr_b)
-          pr_b = 0.1_sp*prd_b + 0.9_sp*prr_b
-          perc_b = 0.1_sp*prd_b + 0.9_sp*prr_b
+          split_b = (pr+perc)*prd_b - (pr+perc)*prr_b
+          pr_b = split*prd_b + (1._sp-split)*prr_b
+          perc_b = split*prd_b + (1._sp-split)*prr_b
           CALL POPREAL4(prr)
           l_b = l_b + prr_b
+          CALL POPREAL4(split)
+          temp_b = (1.0-TANH(ac_alpha2(k)*pn)**2)*2*TANH(ac_alpha2(k)*pn&
+&           )*0.9_sp*split_b
+          ac_alpha2_b(k) = ac_alpha2_b(k) + pn*temp_b
+          pn_b = pn_b + ac_alpha2(k)*temp_b
           CALL POPCONTROL1B(branch)
-          IF (branch .NE. 0) THEN
+          IF (branch .EQ. 0) THEN
+            CALL POPREAL4(perc)
+            CALL POPREAL4(pr)
+          ELSE
             CALL GR_THRESHOLD_EXCHANGE_B(ac_kexc(k), ac_kexc_b(k), &
 &                                  ac_aexc(k), ac_aexc_b(k), ac_ht(k), &
 &                                  ac_ht_b(k), l, l_b)
             CALL POPREAL4(ac_hp(k))
-            pn_b = 0.0_4
-            en_b = 0.0_4
-            CALL GR_PRODUCTION_B(0._sp, dummydiff_b, 0._sp, dummydiff_b0&
-&                          , pn, pn_b, en, en_b, ac_cp(k), ac_cp_b(k), &
-&                          beta, ac_hp(k), ac_hp_b(k), pr, pr_b, perc, &
-&                          perc_b)
+            CALL POPREAL4(pr)
+            CALL POPREAL4(perc)
+            CALL GR_RI_PRODUCTION_B(pn, pn_b, en, en_b, ac_cp(k), &
+&                             ac_cp_b(k), beta, ac_alpha1(k), &
+&                             ac_alpha1_b(k), ac_hp(k), ac_hp_b(k), pr, &
+&                             pr_b, perc, perc_b, setup%dt)
             CALL POPREAL4(ac_hi(k))
             CALL POPREAL4(pn)
             CALL POPREAL4(en)
             CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
 &                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
 &                            , pn, pn_b, en, en_b)
+            pn_b = 0.0_4
           END IF
         END IF
       END DO
     END DO
     ac_mlt_b = ac_mlt_b + ac_prcp_b
-  END SUBROUTINE GR5_TIME_STEP_B
+  END SUBROUTINE GR5_RI_TIME_STEP_B
 
-  SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
-&   time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_aexc, ac_hi, &
-&   ac_hp, ac_ht, ac_qt)
+  SUBROUTINE GR5_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&   , time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_alpha1, ac_alpha2, &
+&   ac_kexc, ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -16242,11 +17126,13 @@ SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
 &   ac_kexc, ac_aexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    INTRINSIC TANH
     INTRINSIC MAX
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
@@ -16263,8 +17149,8 @@ SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
           IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
             CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
 &                          k), pn, en)
-            CALL GR_PRODUCTION(0._sp, 0._sp, pn, en, ac_cp(k), beta, &
-&                        ac_hp(k), pr, perc)
+            CALL GR_RI_PRODUCTION(pn, en, ac_cp(k), beta, ac_alpha1(k), &
+&                           ac_hp(k), pr, perc, setup%dt)
             CALL GR_THRESHOLD_EXCHANGE(ac_kexc(k), ac_aexc(k), ac_ht(k)&
 &                                , l)
           ELSE
@@ -16272,8 +17158,9 @@ SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
             perc = 0._sp
             l = 0._sp
           END IF
-          prr = 0.9_sp*(pr+perc) + l
-          prd = 0.1_sp*(pr+perc)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
+          prr = (1._sp-split)*(pr+perc) + l
+          prd = split*(pr+perc)
           CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
 &                    qr)
           IF (0._sp .LT. prd + l) THEN
@@ -16288,7 +17175,7 @@ SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
         END IF
       END DO
     END DO
-  END SUBROUTINE GR5_TIME_STEP
+  END SUBROUTINE GR5_RI_TIME_STEP
 
 !  Differentiation of gr6_time_step in forward (tangent) mode (with options fixinterface noISIZE context):
 !   variations   of useful results: ac_qt ac_he ac_hi ac_hp ac_ht
@@ -20666,6 +21553,59 @@ SUBROUTINE SIMULATION_CHECKPOINT_D(setup, mesh, input_data, parameters&
         checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
         rr_parameters_inc = rr_parameters_inc + 4
         rr_states_inc = rr_states_inc + 3
+      CASE ('gr4_ri') 
+! 'gr4_ri' module
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+1)
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+2)
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+3)
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % hi
+! % hp
+! % ht
+        CALL GR4_RI_TIME_STEP_D(setup, mesh, input_data, options, &
+&                         returns, t, checkpoint_variable%ac_mlt, &
+&                         checkpoint_variable_d%ac_mlt, &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+1), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+1), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+2), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+2), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+3), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+3), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+4), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+4), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+5), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+5), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+6), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+6), h1, &
+&                         h1_d, h2, h2_d, h3, h3_d, checkpoint_variable%&
+&                         ac_qtz(:, setup%nqz), checkpoint_variable_d%&
+&                         ac_qtz(:, setup%nqz))
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+1) = h1_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+2) = h2_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+3) = h3_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        rr_parameters_inc = rr_parameters_inc + 6
+        rr_states_inc = rr_states_inc + 3
       CASE ('gr4_mlp') 
 ! 'gr4_mlp' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -20875,7 +21815,63 @@ SUBROUTINE SIMULATION_CHECKPOINT_D(setup, mesh, input_data, parameters&
         checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
         checkpoint_variable_d%ac_rr_states(:, rr_states_inc+3) = h3_d
         checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
-        rr_parameters_inc = rr_parameters_inc + 5
+        rr_parameters_inc = rr_parameters_inc + 5
+        rr_states_inc = rr_states_inc + 3
+      CASE ('gr5_ri') 
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+1)
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+2)
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+3)
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % aexc
+! % hi
+! % hp
+! % ht
+        CALL GR5_RI_TIME_STEP_D(setup, mesh, input_data, options, &
+&                         returns, t, checkpoint_variable%ac_mlt, &
+&                         checkpoint_variable_d%ac_mlt, &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+1), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+1), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+2), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+2), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+3), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+3), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+4), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+4), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+5), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+5), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+6), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+6), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+7), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+7), h1, &
+&                         h1_d, h2, h2_d, h3, h3_d, checkpoint_variable%&
+&                         ac_qtz(:, setup%nqz), checkpoint_variable_d%&
+&                         ac_qtz(:, setup%nqz))
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+1) = h1_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+2) = h2_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+3) = h3_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        rr_parameters_inc = rr_parameters_inc + 7
         rr_states_inc = rr_states_inc + 3
       CASE ('gr6') 
 ! 'gr6' module
@@ -21301,6 +22297,49 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 3
         CALL PUSHCONTROL4B(1)
+      CASE ('gr4_ri') 
+! 'gr4_ri' module
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % hi
+! % hp
+! % ht
+        CALL PUSHREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                     SIZE(checkpoint_variable%ac_qtz, 1))
+        CALL PUSHREAL4ARRAY(h3, mesh%nac)
+        CALL PUSHREAL4ARRAY(h2, mesh%nac)
+        CALL PUSHREAL4ARRAY(h1, mesh%nac)
+        CALL GR4_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&                       , t, checkpoint_variable%ac_mlt, &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+1), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+2), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+3), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+4), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+5), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+6), h1, h2&
+&                       , h3, checkpoint_variable%ac_qtz(:, setup%nqz))
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        CALL PUSHINTEGER4(rr_parameters_inc)
+        rr_parameters_inc = rr_parameters_inc + 6
+        CALL PUSHINTEGER4(rr_states_inc)
+        rr_states_inc = rr_states_inc + 3
+        CALL PUSHCONTROL4B(2)
       CASE ('gr4_mlp') 
 ! 'gr4_mlp' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -21343,7 +22382,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 4
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 3
-        CALL PUSHCONTROL4B(2)
+        CALL PUSHCONTROL4B(3)
       CASE ('gr4_ode') 
 ! 'gr4_ode' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -21382,7 +22421,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 4
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 3
-        CALL PUSHCONTROL4B(3)
+        CALL PUSHCONTROL4B(4)
       CASE ('gr4_ode_mlp') 
 ! 'gr4_ode_mlp' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -21426,7 +22465,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 4
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 3
-        CALL PUSHCONTROL4B(4)
+        CALL PUSHCONTROL4B(5)
       CASE ('gr5') 
 ! 'gr5' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -21466,7 +22505,52 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 5
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 3
-        CALL PUSHCONTROL4B(5)
+        CALL PUSHCONTROL4B(6)
+      CASE ('gr5_ri') 
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % aexc
+! % hi
+! % hp
+! % ht
+        CALL PUSHREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                     SIZE(checkpoint_variable%ac_qtz, 1))
+        CALL PUSHREAL4ARRAY(h3, mesh%nac)
+        CALL PUSHREAL4ARRAY(h2, mesh%nac)
+        CALL PUSHREAL4ARRAY(h1, mesh%nac)
+        CALL GR5_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&                       , t, checkpoint_variable%ac_mlt, &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+1), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+2), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+3), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+4), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+5), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+6), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+7), h1, h2, h3, &
+&                       checkpoint_variable%ac_qtz(:, setup%nqz))
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        CALL PUSHINTEGER4(rr_parameters_inc)
+        rr_parameters_inc = rr_parameters_inc + 7
+        CALL PUSHINTEGER4(rr_states_inc)
+        rr_states_inc = rr_states_inc + 3
+        CALL PUSHCONTROL4B(7)
       CASE ('gr6') 
 ! 'gr6' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -21514,7 +22598,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 6
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 4
-        CALL PUSHCONTROL4B(6)
+        CALL PUSHCONTROL4B(8)
       CASE ('grc') 
 ! 'grc' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -21559,7 +22643,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 5
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 4
-        CALL PUSHCONTROL4B(7)
+        CALL PUSHCONTROL4B(9)
       CASE ('grd') 
 ! 'grd' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -21587,7 +22671,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 2
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 2
-        CALL PUSHCONTROL4B(8)
+        CALL PUSHCONTROL4B(10)
       CASE ('loieau') 
 ! 'loieau' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -21618,7 +22702,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 3
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 2
-        CALL PUSHCONTROL4B(9)
+        CALL PUSHCONTROL4B(11)
       CASE ('vic3l') 
 ! 'vic3l' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -21674,7 +22758,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 9
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 4
-        CALL PUSHCONTROL4B(10)
+        CALL PUSHCONTROL4B(12)
       CASE DEFAULT
         CALL PUSHCONTROL4B(0)
       END SELECT
@@ -21771,60 +22855,125 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
 &                     ac_qz)
       END IF
       CALL POPCONTROL4B(branch)
-      IF (branch .LT. 5) THEN
-        IF (branch .LT. 2) THEN
+      IF (branch .LT. 6) THEN
+        IF (branch .LT. 3) THEN
           IF (branch .NE. 0) THEN
-            CALL POPINTEGER4(rr_states_inc)
-            CALL POPINTEGER4(rr_parameters_inc)
-            h3_b = 0.0_4
-            h3_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3&
-&             )
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
-&             0.0_4
-            h2_b = 0.0_4
-            h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2&
-&             )
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
-&             0.0_4
-            h1_b = 0.0_4
-            h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1&
-&             )
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
-&             0.0_4
-            CALL POPREAL4ARRAY(h1, mesh%nac)
-            CALL POPREAL4ARRAY(h2, mesh%nac)
-            CALL POPREAL4ARRAY(h3, mesh%nac)
-            CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz)&
-&                        , SIZE(checkpoint_variable%ac_qtz, 1))
-            CALL GR4_TIME_STEP_B(setup, mesh, input_data, options, &
-&                          returns, t, checkpoint_variable%ac_mlt, &
-&                          checkpoint_variable_b%ac_mlt, &
-&                          checkpoint_variable%ac_rr_parameters(:, &
-&                          rr_parameters_inc+1), checkpoint_variable_b%&
-&                          ac_rr_parameters(:, rr_parameters_inc+1), &
-&                          checkpoint_variable%ac_rr_parameters(:, &
-&                          rr_parameters_inc+2), checkpoint_variable_b%&
-&                          ac_rr_parameters(:, rr_parameters_inc+2), &
-&                          checkpoint_variable%ac_rr_parameters(:, &
-&                          rr_parameters_inc+3), checkpoint_variable_b%&
-&                          ac_rr_parameters(:, rr_parameters_inc+3), &
-&                          checkpoint_variable%ac_rr_parameters(:, &
-&                          rr_parameters_inc+4), checkpoint_variable_b%&
-&                          ac_rr_parameters(:, rr_parameters_inc+4), h1&
-&                          , h1_b, h2, h2_b, h3, h3_b, &
-&                          checkpoint_variable%ac_qtz(:, setup%nqz), &
-&                          checkpoint_variable_b%ac_qtz(:, setup%nqz))
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
-&             checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) + &
-&             h3_b
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
-&             checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) + &
-&             h2_b
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
-&             checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
-&             h1_b
+            IF (branch .EQ. 1) THEN
+              CALL POPINTEGER4(rr_states_inc)
+              CALL POPINTEGER4(rr_parameters_inc)
+              h3_b = 0.0_4
+              h3_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +3)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
+&               0.0_4
+              h2_b = 0.0_4
+              h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +2)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&               0.0_4
+              h1_b = 0.0_4
+              h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +1)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&               0.0_4
+              CALL POPREAL4ARRAY(h1, mesh%nac)
+              CALL POPREAL4ARRAY(h2, mesh%nac)
+              CALL POPREAL4ARRAY(h3, mesh%nac)
+              CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz&
+&                          ), SIZE(checkpoint_variable%ac_qtz, 1))
+              CALL GR4_TIME_STEP_B(setup, mesh, input_data, options, &
+&                            returns, t, checkpoint_variable%ac_mlt, &
+&                            checkpoint_variable_b%ac_mlt, &
+&                            checkpoint_variable%ac_rr_parameters(:, &
+&                            rr_parameters_inc+1), checkpoint_variable_b&
+&                            %ac_rr_parameters(:, rr_parameters_inc+1), &
+&                            checkpoint_variable%ac_rr_parameters(:, &
+&                            rr_parameters_inc+2), checkpoint_variable_b&
+&                            %ac_rr_parameters(:, rr_parameters_inc+2), &
+&                            checkpoint_variable%ac_rr_parameters(:, &
+&                            rr_parameters_inc+3), checkpoint_variable_b&
+&                            %ac_rr_parameters(:, rr_parameters_inc+3), &
+&                            checkpoint_variable%ac_rr_parameters(:, &
+&                            rr_parameters_inc+4), checkpoint_variable_b&
+&                            %ac_rr_parameters(:, rr_parameters_inc+4), &
+&                            h1, h1_b, h2, h2_b, h3, h3_b, &
+&                            checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                            checkpoint_variable_b%ac_qtz(:, setup%nqz))
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) +&
+&               h3_b
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) +&
+&               h2_b
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) +&
+&               h1_b
+            ELSE
+              CALL POPINTEGER4(rr_states_inc)
+              CALL POPINTEGER4(rr_parameters_inc)
+              h3_b = 0.0_4
+              h3_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +3)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
+&               0.0_4
+              h2_b = 0.0_4
+              h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +2)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&               0.0_4
+              h1_b = 0.0_4
+              h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +1)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&               0.0_4
+              CALL POPREAL4ARRAY(h1, mesh%nac)
+              CALL POPREAL4ARRAY(h2, mesh%nac)
+              CALL POPREAL4ARRAY(h3, mesh%nac)
+              CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz&
+&                          ), SIZE(checkpoint_variable%ac_qtz, 1))
+              CALL GR4_RI_TIME_STEP_B(setup, mesh, input_data, options, &
+&                               returns, t, checkpoint_variable%ac_mlt, &
+&                               checkpoint_variable_b%ac_mlt, &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+1), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+1), &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+2), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+2), &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+3), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+3), &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+4), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+4), &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+5), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+5), &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+6), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+6), h1, h1_b, h2, &
+&                               h2_b, h3, h3_b, checkpoint_variable%&
+&                               ac_qtz(:, setup%nqz), &
+&                               checkpoint_variable_b%ac_qtz(:, setup%&
+&                               nqz))
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) +&
+&               h3_b
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) +&
+&               h2_b
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) +&
+&               h1_b
+            END IF
           END IF
-        ELSE IF (branch .EQ. 2) THEN
+        ELSE IF (branch .EQ. 3) THEN
           CALL POPINTEGER4(rr_states_inc)
           CALL POPINTEGER4(rr_parameters_inc)
           h3_b = 0.0_4
@@ -21880,7 +23029,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
 &           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
 &           h1_b
-        ELSE IF (branch .EQ. 3) THEN
+        ELSE IF (branch .EQ. 4) THEN
           CALL POPINTEGER4(rr_states_inc)
           CALL POPINTEGER4(rr_parameters_inc)
           h3_b = 0.0_4
@@ -21986,8 +23135,8 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
 &           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
 &           h1_b
         END IF
-      ELSE IF (branch .LT. 8) THEN
-        IF (branch .EQ. 5) THEN
+      ELSE IF (branch .LT. 9) THEN
+        IF (branch .EQ. 6) THEN
           CALL POPINTEGER4(rr_states_inc)
           CALL POPINTEGER4(rr_parameters_inc)
           h3_b = 0.0_4
@@ -22034,7 +23183,60 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
 &           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
 &           h1_b
-        ELSE IF (branch .EQ. 6) THEN
+        ELSE IF (branch .EQ. 7) THEN
+          CALL POPINTEGER4(rr_states_inc)
+          CALL POPINTEGER4(rr_parameters_inc)
+          h3_b = 0.0_4
+          h3_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3)
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = 0.0_4
+          h2_b = 0.0_4
+          h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2)
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = 0.0_4
+          h1_b = 0.0_4
+          h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1)
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = 0.0_4
+          CALL POPREAL4ARRAY(h1, mesh%nac)
+          CALL POPREAL4ARRAY(h2, mesh%nac)
+          CALL POPREAL4ARRAY(h3, mesh%nac)
+          CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                      SIZE(checkpoint_variable%ac_qtz, 1))
+          CALL GR5_RI_TIME_STEP_B(setup, mesh, input_data, options, &
+&                           returns, t, checkpoint_variable%ac_mlt, &
+&                           checkpoint_variable_b%ac_mlt, &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+1), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+1), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+2), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+2), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+3), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+3), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+4), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+4), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+5), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+5), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+6), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+6), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+7), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+7), h1&
+&                           , h1_b, h2, h2_b, h3, h3_b, &
+&                           checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                           checkpoint_variable_b%ac_qtz(:, setup%nqz))
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
+&           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) + &
+&           h3_b
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) + &
+&           h2_b
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
+&           h1_b
+        ELSE
           CALL POPINTEGER4(rr_states_inc)
           CALL POPINTEGER4(rr_parameters_inc)
           h4_b = 0.0_4
@@ -22091,7 +23293,9 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
 &           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
 &           h1_b
-        ELSE
+        END IF
+      ELSE IF (branch .LT. 11) THEN
+        IF (branch .EQ. 9) THEN
           CALL POPINTEGER4(rr_states_inc)
           CALL POPINTEGER4(rr_parameters_inc)
           h4_b = 0.0_4
@@ -22144,37 +23348,40 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
 &           h2_b
           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
 &           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
+&           h1_b
+        ELSE
+          CALL POPINTEGER4(rr_states_inc)
+          CALL POPINTEGER4(rr_parameters_inc)
+          h2_b = 0.0_4
+          h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2)
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = 0.0_4
+          h1_b = 0.0_4
+          h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1)
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = 0.0_4
+          CALL POPREAL4ARRAY(h1, mesh%nac)
+          CALL POPREAL4ARRAY(h2, mesh%nac)
+          CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                      SIZE(checkpoint_variable%ac_qtz, 1))
+          CALL GRD_TIME_STEP_B(setup, mesh, input_data, options, returns&
+&                        , t, checkpoint_variable%ac_mlt, &
+&                        checkpoint_variable_b%ac_mlt, &
+&                        checkpoint_variable%ac_rr_parameters(:, &
+&                        rr_parameters_inc+1), checkpoint_variable_b%&
+&                        ac_rr_parameters(:, rr_parameters_inc+1), &
+&                        checkpoint_variable%ac_rr_parameters(:, &
+&                        rr_parameters_inc+2), checkpoint_variable_b%&
+&                        ac_rr_parameters(:, rr_parameters_inc+2), h1, &
+&                        h1_b, h2, h2_b, checkpoint_variable%ac_qtz(:, &
+&                        setup%nqz), checkpoint_variable_b%ac_qtz(:, &
+&                        setup%nqz))
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) + &
+&           h2_b
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
 &           h1_b
         END IF
-      ELSE IF (branch .EQ. 8) THEN
-        CALL POPINTEGER4(rr_states_inc)
-        CALL POPINTEGER4(rr_parameters_inc)
-        h2_b = 0.0_4
-        h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2)
-        checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = 0.0_4
-        h1_b = 0.0_4
-        h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1)
-        checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = 0.0_4
-        CALL POPREAL4ARRAY(h1, mesh%nac)
-        CALL POPREAL4ARRAY(h2, mesh%nac)
-        CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz), &
-&                    SIZE(checkpoint_variable%ac_qtz, 1))
-        CALL GRD_TIME_STEP_B(setup, mesh, input_data, options, returns, &
-&                      t, checkpoint_variable%ac_mlt, &
-&                      checkpoint_variable_b%ac_mlt, checkpoint_variable&
-&                      %ac_rr_parameters(:, rr_parameters_inc+1), &
-&                      checkpoint_variable_b%ac_rr_parameters(:, &
-&                      rr_parameters_inc+1), checkpoint_variable%&
-&                      ac_rr_parameters(:, rr_parameters_inc+2), &
-&                      checkpoint_variable_b%ac_rr_parameters(:, &
-&                      rr_parameters_inc+2), h1, h1_b, h2, h2_b, &
-&                      checkpoint_variable%ac_qtz(:, setup%nqz), &
-&                      checkpoint_variable_b%ac_qtz(:, setup%nqz))
-        checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
-&         checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) + h2_b
-        checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
-&         checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + h1_b
-      ELSE IF (branch .EQ. 9) THEN
+      ELSE IF (branch .EQ. 11) THEN
         CALL POPINTEGER4(rr_states_inc)
         CALL POPINTEGER4(rr_parameters_inc)
         h2_b = 0.0_4
@@ -22376,6 +23583,41 @@ SUBROUTINE SIMULATION_CHECKPOINT(setup, mesh, input_data, parameters, &
         checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
         rr_parameters_inc = rr_parameters_inc + 4
         rr_states_inc = rr_states_inc + 3
+      CASE ('gr4_ri') 
+! 'gr4_ri' module
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % hi
+! % hp
+! % ht
+        CALL GR4_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&                       , t, checkpoint_variable%ac_mlt, &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+1), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+2), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+3), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+4), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+5), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+6), h1, h2&
+&                       , h3, checkpoint_variable%ac_qtz(:, setup%nqz))
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        rr_parameters_inc = rr_parameters_inc + 6
+        rr_states_inc = rr_states_inc + 3
       CASE ('gr4_mlp') 
 ! 'gr4_mlp' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -22510,6 +23752,43 @@ SUBROUTINE SIMULATION_CHECKPOINT(setup, mesh, input_data, parameters, &
         checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
         rr_parameters_inc = rr_parameters_inc + 5
         rr_states_inc = rr_states_inc + 3
+      CASE ('gr5_ri') 
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % aexc
+! % hi
+! % hp
+! % ht
+        CALL GR5_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&                       , t, checkpoint_variable%ac_mlt, &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+1), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+2), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+3), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+4), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+5), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+6), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+7), h1, h2, h3, &
+&                       checkpoint_variable%ac_qtz(:, setup%nqz))
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        rr_parameters_inc = rr_parameters_inc + 7
+        rr_states_inc = rr_states_inc + 3
       CASE ('gr6') 
 ! 'gr6' module
 ! % To avoid potential aliasing tapenade warning (DF02)
diff --git a/smash/fcore/forward/forward_openmp_db.f90 b/smash/fcore/forward/forward_openmp_db.f90
index 6a006d4b..3f08c6d5 100644
--- a/smash/fcore/forward/forward_openmp_db.f90
+++ b/smash/fcore/forward/forward_openmp_db.f90
@@ -13492,6 +13492,267 @@ SUBROUTINE GR_PRODUCTION(fq_ps, fq_es, pn, en, cp, beta, hp, pr, perc)
     hp = hp_imd - perc*inv_cp
   END SUBROUTINE GR_PRODUCTION
 
+!  Differentiation of gr_ri_production in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
+!   variations   of useful results: hp perc pr
+!   with respect to varying inputs: alpha1 hp en cp pn
+  SUBROUTINE GR_RI_PRODUCTION_D(pn, pn_d, en, en_d, cp, cp_d, beta, &
+&   alpha1, alpha1_d, hp, hp_d, pr, pr_d, perc, perc_d, dt)
+    IMPLICIT NONE
+    REAL(sp), INTENT(IN) :: pn, en, cp, beta, alpha1
+    REAL(sp), INTENT(IN) :: pn_d, en_d, cp_d, alpha1_d
+    REAL(sp), INTENT(IN) :: dt
+    REAL(sp), INTENT(INOUT) :: hp
+    REAL(sp), INTENT(INOUT) :: hp_d
+    REAL(sp), INTENT(OUT) :: pr, perc
+    REAL(sp), INTENT(OUT) :: pr_d, perc_d
+    REAL(sp) :: inv_cp, ps, es, hp_imd
+    REAL(sp) :: inv_cp_d, ps_d, es_d, hp_imd_d
+    REAL(sp) :: lambda, gam, inv_lambda
+    REAL(sp) :: lambda_d, gam_d, inv_lambda_d
+    INTRINSIC EXP
+    INTRINSIC SQRT
+    INTRINSIC TANH
+    REAL(sp) :: arg1
+    REAL(sp) :: arg1_d
+    REAL(sp) :: arg2
+    REAL(sp) :: arg2_d
+    REAL(sp) :: pwx1
+    REAL(sp) :: pwx1_d
+    REAL(sp) :: pwr1
+    REAL(sp) :: pwr1_d
+    REAL(sp) :: temp
+    REAL(sp) :: temp0
+    REAL(sp) :: temp1
+    REAL(sp) :: temp2
+    REAL(sp) :: temp3
+    REAL(sp) :: temp4
+    inv_cp_d = -(cp_d/cp**2)
+    inv_cp = 1._sp/cp
+    pr = 0._sp
+    gam_d = EXP(-(pn*alpha1))*(alpha1*pn_d+pn*alpha1_d)
+    gam = 1._sp - EXP(-(pn*alpha1))
+    temp = SQRT(-gam + 1._sp)
+    IF (1._sp - gam .EQ. 0.0) THEN
+      lambda_d = 0.0_4
+    ELSE
+      lambda_d = -(gam_d/(2.0*temp))
+    END IF
+    lambda = temp
+    inv_lambda_d = -(lambda_d/lambda**2)
+    inv_lambda = 1._sp/lambda
+    arg1_d = inv_cp*(pn*lambda_d+lambda*pn_d) + lambda*pn*inv_cp_d
+    arg1 = lambda*pn*inv_cp
+    arg2_d = inv_cp*(pn*lambda_d+lambda*pn_d) + lambda*pn*inv_cp_d
+    arg2 = lambda*pn*inv_cp
+    temp = TANH(arg2)
+    temp0 = lambda*hp*temp + 1._sp
+    temp1 = -(lambda*hp*(lambda*hp)) + 1._sp
+    temp2 = cp*inv_lambda*temp1
+    temp3 = TANH(arg1)
+    temp4 = temp3*temp2/temp0
+    ps_d = (temp2*(1.0-TANH(arg1)**2)*arg1_d+temp3*(temp1*(inv_lambda*&
+&     cp_d+cp*inv_lambda_d)-cp*inv_lambda*2*lambda*hp*(hp*lambda_d+&
+&     lambda*hp_d))-temp4*(temp*(hp*lambda_d+lambda*hp_d)+lambda*hp*(1.0&
+&     -TANH(arg2)**2)*arg2_d))/temp0 - dt*gam_d
+    ps = temp4 - dt*gam
+    temp4 = TANH(en*inv_cp)
+    temp3 = TANH(en*inv_cp)
+    temp2 = hp*cp*(-hp+2._sp)
+    temp1 = temp2*temp3/((-hp+1._sp)*temp4+1._sp)
+    es_d = (temp3*((2._sp-hp)*(cp*hp_d+hp*cp_d)-hp*cp*hp_d)+temp2*(1.0-&
+&     TANH(en*inv_cp)**2)*(inv_cp*en_d+en*inv_cp_d)-temp1*((1._sp-hp)*(&
+&     1.0-TANH(en*inv_cp)**2)*(inv_cp*en_d+en*inv_cp_d)-temp4*hp_d))/((&
+&     1._sp-hp)*temp4+1._sp)
+    es = temp1
+    hp_imd_d = hp_d + inv_cp*(ps_d-es_d) + (ps-es)*inv_cp_d
+    hp_imd = hp + (ps-es)*inv_cp
+    IF (pn .GT. 0) THEN
+      pr_d = pn_d - cp*(hp_imd_d-hp_d) - (hp_imd-hp)*cp_d
+      pr = pn - (hp_imd-hp)*cp
+    ELSE
+      pr_d = 0.0_4
+    END IF
+    pwx1_d = 4*hp_imd**3*hp_imd_d/beta**4
+    pwx1 = 1._sp + (hp_imd/beta)**4
+    pwr1_d = -(0.25_sp*pwx1**(-1.25)*pwx1_d)
+    pwr1 = pwx1**(-0.25_sp)
+    perc_d = (1._sp-pwr1)*(cp*hp_imd_d+hp_imd*cp_d) - hp_imd*cp*pwr1_d
+    perc = hp_imd*cp*(1._sp-pwr1)
+    hp_d = hp_imd_d - inv_cp*perc_d - perc*inv_cp_d
+    hp = hp_imd - perc*inv_cp
+  END SUBROUTINE GR_RI_PRODUCTION_D
+
+!  Differentiation of gr_ri_production in reverse (adjoint) mode (with options fixinterface noISIZE context OpenMP):
+!   gradient     of useful results: alpha1 hp cp pn perc pr
+!   with respect to varying inputs: alpha1 hp en cp pn
+  SUBROUTINE GR_RI_PRODUCTION_B(pn, pn_b, en, en_b, cp, cp_b, beta, &
+&   alpha1, alpha1_b, hp, hp_b, pr, pr_b, perc, perc_b, dt)
+    IMPLICIT NONE
+    REAL(sp), INTENT(IN) :: pn, en, cp, beta, alpha1
+    REAL(sp) :: pn_b, en_b, cp_b, alpha1_b
+    REAL(sp), INTENT(IN) :: dt
+    REAL(sp), INTENT(INOUT) :: hp
+    REAL(sp), INTENT(INOUT) :: hp_b
+    REAL(sp) :: pr, perc
+    REAL(sp) :: pr_b, perc_b
+    REAL(sp) :: inv_cp, ps, es, hp_imd
+    REAL(sp) :: inv_cp_b, ps_b, es_b, hp_imd_b
+    REAL(sp) :: lambda, gam, inv_lambda
+    REAL(sp) :: lambda_b, gam_b, inv_lambda_b
+    INTRINSIC EXP
+    INTRINSIC SQRT
+    INTRINSIC TANH
+    REAL(sp) :: arg1
+    REAL(sp) :: arg1_b
+    REAL(sp) :: arg2
+    REAL(sp) :: arg2_b
+    REAL(sp) :: pwx1
+    REAL(sp) :: pwx1_b
+    REAL(sp) :: pwr1
+    REAL(sp) :: pwr1_b
+    REAL(sp) :: temp
+    REAL(sp) :: temp_b
+    REAL(sp) :: temp0
+    REAL(sp) :: temp_b0
+    REAL(sp) :: temp1
+    REAL(sp) :: temp2
+    REAL(sp) :: temp3
+    REAL(sp) :: temp_b1
+    REAL(sp) :: temp4
+    REAL(sp) :: temp_b2
+    REAL(sp) :: temp_b3
+    REAL(sp) :: temp_b4
+    REAL(sp) :: temp_b5
+    INTEGER :: branch
+    inv_cp = 1._sp/cp
+    gam = 1._sp - EXP(-(pn*alpha1))
+    lambda = SQRT(1._sp - gam)
+    inv_lambda = 1._sp/lambda
+    arg1 = lambda*pn*inv_cp
+    arg2 = lambda*pn*inv_cp
+    ps = cp*inv_lambda*TANH(arg1)*(1._sp-(lambda*hp)**2)/(1._sp+lambda*&
+&     hp*TANH(arg2)) - gam*dt
+    es = hp*cp*(2._sp-hp)*TANH(en*inv_cp)/(1._sp+(1._sp-hp)*TANH(en*&
+&     inv_cp))
+    hp_imd = hp + (ps-es)*inv_cp
+    IF (pn .GT. 0) THEN
+      CALL PUSHCONTROL1B(0)
+    ELSE
+      CALL PUSHCONTROL1B(1)
+    END IF
+    pwx1 = 1._sp + (hp_imd/beta)**4
+    pwr1 = pwx1**(-0.25_sp)
+    CALL PUSHREAL4(perc)
+    perc = hp_imd*cp*(1._sp-pwr1)
+    pwx1 = 1._sp + (hp_imd/beta)**4
+    pwr1 = pwx1**(-0.25_sp)
+    inv_cp = 1._sp/cp
+    perc_b = perc_b - inv_cp*hp_b
+    inv_cp_b = -(perc*hp_b)
+    CALL POPREAL4(perc)
+!$OMP ATOMIC update
+    cp_b = cp_b + hp_imd*(1._sp-pwr1)*perc_b
+    pwr1_b = -(hp_imd*cp*perc_b)
+    pwx1_b = -(0.25_sp*pwx1**(-1.25)*pwr1_b)
+    hp_imd_b = hp_b + cp*(1._sp-pwr1)*perc_b + 4*hp_imd**3*pwx1_b/beta**&
+&     4
+    CALL POPCONTROL1B(branch)
+    IF (branch .EQ. 0) THEN
+      pn_b = pn_b + pr_b
+      hp_imd_b = hp_imd_b - cp*pr_b
+      hp_b = cp*pr_b
+!$OMP ATOMIC update
+      cp_b = cp_b - (hp_imd-hp)*pr_b
+    ELSE
+      hp_b = 0.0_4
+    END IF
+    es_b = -(inv_cp*hp_imd_b)
+    temp4 = TANH(en*inv_cp)
+    temp3 = (-hp+1._sp)*temp4 + 1._sp
+    temp1 = TANH(en*inv_cp)
+    temp0 = hp*cp*(-hp+2._sp)
+    temp_b1 = es_b/temp3
+    temp_b0 = (2._sp-hp)*temp1*temp_b1
+    temp_b4 = -(temp0*temp1*temp_b1/temp3)
+!$OMP ATOMIC update
+    hp_b = hp_b + hp_imd_b + cp*temp_b0 - hp*cp*temp1*temp_b1 - temp4*&
+&     temp_b4
+    ps_b = inv_cp*hp_imd_b
+    temp_b = (1.0-TANH(en*inv_cp)**2)*temp0*temp_b1
+    temp_b5 = (1.0-TANH(en*inv_cp)**2)*(1._sp-hp)*temp_b4
+    en_b = inv_cp*temp_b5 + inv_cp*temp_b
+!$OMP ATOMIC update
+    cp_b = cp_b + hp*temp_b0
+    arg1 = lambda*pn*inv_cp
+    arg2 = lambda*pn*inv_cp
+    inv_lambda = 1._sp/lambda
+    temp = TANH(arg2)
+    temp0 = lambda*hp*temp + 1._sp
+    temp1 = -(lambda*hp*(lambda*hp)) + 1._sp
+    temp2 = cp*inv_lambda*temp1
+    temp3 = TANH(arg1)
+    temp_b0 = ps_b/temp0
+    arg1_b = (1.0-TANH(arg1)**2)*temp2*temp_b0
+    temp_b1 = temp3*temp_b0
+    temp_b3 = -(temp3*temp2*temp_b0/temp0)
+    arg2_b = (1.0-TANH(arg2)**2)*lambda*hp*temp_b3
+    inv_cp_b = inv_cp_b + (ps-es)*hp_imd_b + en*temp_b5 + en*temp_b + &
+&     lambda*pn*arg2_b + lambda*pn*arg1_b
+!$OMP ATOMIC update
+    cp_b = cp_b + inv_lambda*temp1*temp_b1 - inv_cp_b/cp**2
+    inv_lambda_b = cp*temp1*temp_b1
+    temp_b2 = -(2*lambda*hp*cp*inv_lambda*temp_b1)
+    lambda_b = hp*temp*temp_b3 + hp*temp_b2 + pn*inv_cp*arg2_b + pn*&
+&     inv_cp*arg1_b - inv_lambda_b/lambda**2
+    IF (1._sp - gam .EQ. 0.0) THEN
+      gam_b = -(dt*ps_b)
+    ELSE
+      gam_b = -(dt*ps_b) - lambda_b/(2.0*SQRT(1._sp-gam))
+    END IF
+!$OMP ATOMIC update
+    hp_b = hp_b + lambda*temp*temp_b3 + lambda*temp_b2
+    temp_b = -(EXP(-(pn*alpha1))*gam_b)
+    pn_b = pn_b + lambda*inv_cp*arg2_b + lambda*inv_cp*arg1_b - alpha1*&
+&     temp_b
+!$OMP ATOMIC update
+    alpha1_b = alpha1_b - pn*temp_b
+  END SUBROUTINE GR_RI_PRODUCTION_B
+
+  SUBROUTINE GR_RI_PRODUCTION(pn, en, cp, beta, alpha1, hp, pr, perc, dt&
+& )
+    IMPLICIT NONE
+    REAL(sp), INTENT(IN) :: pn, en, cp, beta, alpha1
+    REAL(sp), INTENT(IN) :: dt
+    REAL(sp), INTENT(INOUT) :: hp
+    REAL(sp), INTENT(OUT) :: pr, perc
+    REAL(sp) :: inv_cp, ps, es, hp_imd
+    REAL(sp) :: lambda, gam, inv_lambda
+    INTRINSIC EXP
+    INTRINSIC SQRT
+    INTRINSIC TANH
+    REAL(sp) :: arg1
+    REAL(sp) :: arg2
+    REAL(sp) :: pwx1
+    REAL(sp) :: pwr1
+    inv_cp = 1._sp/cp
+    pr = 0._sp
+    gam = 1._sp - EXP(-(pn*alpha1))
+    lambda = SQRT(1._sp - gam)
+    inv_lambda = 1._sp/lambda
+    arg1 = lambda*pn*inv_cp
+    arg2 = lambda*pn*inv_cp
+    ps = cp*inv_lambda*TANH(arg1)*(1._sp-(lambda*hp)**2)/(1._sp+lambda*&
+&     hp*TANH(arg2)) - gam*dt
+    es = hp*cp*(2._sp-hp)*TANH(en*inv_cp)/(1._sp+(1._sp-hp)*TANH(en*&
+&     inv_cp))
+    hp_imd = hp + (ps-es)*inv_cp
+    IF (pn .GT. 0) pr = pn - (hp_imd-hp)*cp
+    pwx1 = 1._sp + (hp_imd/beta)**4
+    pwr1 = pwx1**(-0.25_sp)
+    perc = hp_imd*cp*(1._sp-pwr1)
+    hp = hp_imd - perc*inv_cp
+  END SUBROUTINE GR_RI_PRODUCTION
+
 !  Differentiation of gr_exchange in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
 !   variations   of useful results: l
 !   with respect to varying inputs: kexc fq_l ht
@@ -14914,17 +15175,15 @@ SUBROUTINE GR4_TIME_STEP(setup, mesh, input_data, options, returns, &
     END DO
   END SUBROUTINE GR4_TIME_STEP
 
-!  Differentiation of gr4_mlp_time_step in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
+!  Differentiation of gr4_ri_time_step in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
 !   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-  SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
-&   returns, time_step, weight_1, weight_1_d, bias_1, bias_1_d, weight_2&
-&   , weight_2_d, bias_2, bias_2_d, weight_3, weight_3_d, bias_3, &
-&   bias_3_d, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
-&   ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d, ac_ht, &
-&   ac_ht_d, ac_qt, ac_qt_d)
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt
+  SUBROUTINE GR4_RI_TIME_STEP_D(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d&
+&   , ac_ct, ac_ct_d, ac_alpha1, ac_alpha1_d, ac_alpha2, ac_alpha2_d, &
+&   ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d, ac_ht, ac_ht_d, &
+&   ac_qt, ac_qt_d)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -14932,46 +15191,27 @@ SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1_d
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1_d
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2_d
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2_d
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3_d
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
 &   ac_kexc
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci_d, ac_cp_d, &
 &   ac_ct_d, ac_kexc_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1_d, &
+&   ac_alpha2_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_d, ac_hp_d, &
 &   ac_ht_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_d
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_d
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer_d
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d, pn_d, en_d
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
-    REAL(sp) :: pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    REAL(sp) :: pn_d, en_d, pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d&
+&   , split_d
+    INTRINSIC TANH
     INTRINSIC MAX
     REAL(sp) :: temp
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
@@ -14982,12 +15222,14 @@ SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
     ac_prcp = ac_prcp + ac_mlt
 ! Beta percolation parameter is time step dependent
     beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
-    en_d = 0.0_4
     pn_d = 0.0_4
-! Interception with OPENMP
-!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
-!$OMP&ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_d, ac_ci_d, ac_hi_d&
-!$OMP&, pn_d, en_d), PRIVATE(row, col, k), SCHEDULE(static)
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
+!$OMP&ac_hi, ac_hp, ac_ht, ac_qt), SHARED(ac_prcp_d, ac_ci_d, ac_cp_d, &
+!$OMP&ac_ct_d, ac_kexc_d, ac_hi_d, ac_hp_d, ac_ht_d, ac_qt_d), PRIVATE(&
+!$OMP&row, col, k, time_step_returns, pn, en, pr, perc, l, prr, prd, qr&
+!$OMP&, qd, split), PRIVATE(pn_d, en_d, pr_d, perc_d, l_d, prr_d, prd_d&
+!$OMP&, qr_d, qd_d, split_d), PRIVATE(temp), SCHEDULE(static)
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -14996,110 +15238,58 @@ SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
           IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
             CALL GR_INTERCEPTION_D(ac_prcp(k), ac_prcp_d(k), ac_pet(k), &
 &                            ac_ci(k), ac_ci_d(k), ac_hi(k), ac_hi_d(k)&
-&                            , pn(k), pn_d(k), en(k), en_d(k))
+&                            , pn, pn_d, en, en_d)
+            CALL GR_RI_PRODUCTION_D(pn, pn_d, en, en_d, ac_cp(k), &
+&                             ac_cp_d(k), beta, ac_alpha1(k), &
+&                             ac_alpha1_d(k), ac_hp(k), ac_hp_d(k), pr, &
+&                             pr_d, perc, perc_d, setup%dt)
+            CALL GR_EXCHANGE_D(0._sp, 0.0_4, ac_kexc(k), ac_kexc_d(k), &
+&                        ac_ht(k), ac_ht_d(k), l, l_d)
           ELSE
-            pn_d(k) = 0.0_4
-            pn(k) = 0._sp
-            en_d(k) = 0.0_4
-            en(k) = 0._sp
+            pr = 0._sp
+            perc = 0._sp
+            l = 0._sp
+            l_d = 0.0_4
+            perc_d = 0.0_4
+            pr_d = 0.0_4
           END IF
-        END IF
-      END DO
-    END DO
-    output_layer_d = 0.0_4
-! Forward MLP without OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            input_layer_d(:) = (/ac_hp_d(k), ac_ht_d(k), pn_d(k), en_d(k&
-&             )/)
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP_D(weight_1, weight_1_d, bias_1, bias_1_d, &
-&                        weight_2, weight_2_d, bias_2, bias_2_d, &
-&                        weight_3, weight_3_d, bias_3, bias_3_d, &
-&                        input_layer, input_layer_d, output_layer(:, k)&
-&                        , output_layer_d(:, k))
+          split_d = 0.9_sp*2*TANH(ac_alpha2(k)*pn)*(1.0-TANH(ac_alpha2(k&
+&           )*pn)**2)*(pn*ac_alpha2_d(k)+ac_alpha2(k)*pn_d)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
+          prr_d = (1._sp-split)*(pr_d+perc_d) - (pr+perc)*split_d + l_d
+          prr = (1._sp-split)*(pr+perc) + l
+          prd_d = (pr+perc)*split_d + split*(pr_d+perc_d)
+          prd = split*(pr+perc)
+          CALL GR_TRANSFER_D(5._sp, ac_prcp(k), prr, prr_d, ac_ct(k), &
+&                      ac_ct_d(k), ac_ht(k), ac_ht_d(k), qr, qr_d)
+          IF (0._sp .LT. prd + l) THEN
+            qd_d = prd_d + l_d
+            qd = prd + l
           ELSE
-            output_layer_d(:, k) = 0.0_4
-            output_layer(:, k) = 0._sp
+            qd = 0._sp
+            qd_d = 0.0_4
           END IF
+          ac_qt_d(k) = qr_d + qd_d
+          ac_qt(k) = qr + qd
+! Transform from mm/dt to m3/s
+          temp = 1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)
+          ac_qt_d(k) = temp*ac_qt_d(k)/setup%dt
+          ac_qt(k) = temp*(ac_qt(k)/setup%dt)
         END IF
       END DO
     END DO
-! Production and transfer with OPENMP
-!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, output_layer, ac_prcp, ac_pet, ac_cp, beta, ac_ct, &
-!$OMP&ac_kexc, ac_hp, ac_ht, ac_qt, pn, en), SHARED(output_layer_d, &
-!$OMP&ac_prcp_d, ac_cp_d, ac_ct_d, ac_kexc_d, ac_hp_d, ac_ht_d, ac_qt_d&
-!$OMP&, pn_d, en_d), PRIVATE(row, col, k, time_step_returns, pr, perc, l&
-!$OMP&, prr, prd, qr, qd), PRIVATE(pr_d, perc_d, l_d, prr_d, prd_d, qr_d&
-!$OMP&, qd_d), PRIVATE(temp), SCHEDULE(static)
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            CALL GR_PRODUCTION_D(output_layer(1, k), output_layer_d(1, k&
-&                          ), output_layer(2, k), output_layer_d(2, k), &
-&                          pn(k), pn_d(k), en(k), en_d(k), ac_cp(k), &
-&                          ac_cp_d(k), beta, ac_hp(k), ac_hp_d(k), pr, &
-&                          pr_d, perc, perc_d)
-            CALL GR_EXCHANGE_D(output_layer(4, k), output_layer_d(4, k)&
-&                        , ac_kexc(k), ac_kexc_d(k), ac_ht(k), ac_ht_d(k&
-&                        ), l, l_d)
-          ELSE
-            pr = 0._sp
-            perc = 0._sp
-            l = 0._sp
-            l_d = 0.0_4
-            perc_d = 0.0_4
-            pr_d = 0.0_4
-          END IF
-          temp = -(output_layer(3, k)*output_layer(3, k)) + 1._sp
-          prr_d = 0.9_sp*(temp*(pr_d+perc_d)-(pr+perc)*2*output_layer(3&
-&           , k)*output_layer_d(3, k)) + l_d
-          prr = 0.9_sp*(temp*(pr+perc)) + l
-          temp = 0.9_sp*(output_layer(3, k)*output_layer(3, k)) + 0.1_sp
-          prd_d = (pr+perc)*0.9_sp*2*output_layer(3, k)*output_layer_d(3&
-&           , k) + temp*(pr_d+perc_d)
-          prd = temp*(pr+perc)
-          CALL GR_TRANSFER_D(5._sp, ac_prcp(k), prr, prr_d, ac_ct(k), &
-&                      ac_ct_d(k), ac_ht(k), ac_ht_d(k), qr, qr_d)
-          IF (0._sp .LT. prd + l) THEN
-            qd_d = prd_d + l_d
-            qd = prd + l
-          ELSE
-            qd = 0._sp
-            qd_d = 0.0_4
-          END IF
-          ac_qt_d(k) = qr_d + qd_d
-          ac_qt(k) = qr + qd
-! Transform from mm/dt to m3/s
-          temp = 1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)
-          ac_qt_d(k) = temp*ac_qt_d(k)/setup%dt
-          ac_qt(k) = temp*(ac_qt(k)/setup%dt)
-        END IF
-      END DO
-    END DO
-  END SUBROUTINE GR4_MLP_TIME_STEP_D
+  END SUBROUTINE GR4_RI_TIME_STEP_D
 
-!  Differentiation of gr4_mlp_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context OpenMP):
-!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-  SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
-&   returns, time_step, weight_1, weight_1_b, bias_1, bias_1_b, weight_2&
-&   , weight_2_b, bias_2, bias_2_b, weight_3, weight_3_b, bias_3, &
-&   bias_3_b, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
-&   ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b, ac_ht, &
-&   ac_ht_b, ac_qt, ac_qt_b)
+!  Differentiation of gr4_ri_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context OpenMP):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt
+  SUBROUTINE GR4_RI_TIME_STEP_B(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b&
+&   , ac_ct, ac_ct_b, ac_alpha1, ac_alpha1_b, ac_alpha2, ac_alpha2_b, &
+&   ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b, ac_ht, ac_ht_b, &
+&   ac_qt, ac_qt_b)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15107,51 +15297,32 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)) :: &
-&   weight_1_b
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
-    REAL(sp), DIMENSION(setup%neurons(2)) :: bias_1_b
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)) :: &
-&   weight_2_b
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
-    REAL(sp), DIMENSION(setup%neurons(3)) :: bias_2_b
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)) :: &
-&   weight_3_b
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
-    REAL(sp), DIMENSION(setup%neurons(4)) :: bias_3_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac) :: ac_mlt_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
 &   ac_kexc
     REAL(sp), DIMENSION(mesh%nac) :: ac_ci_b, ac_cp_b, ac_ct_b, &
 &   ac_kexc_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
+    REAL(sp), DIMENSION(mesh%nac) :: ac_alpha1_b, ac_alpha2_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_b, ac_hp_b, &
 &   ac_ht_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_b
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_b
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer_b
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b, pn_b, en_b
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
-    REAL(sp) :: pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    REAL(sp) :: pn_b, en_b, pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b&
+&   , split_b
+    INTRINSIC TANH
     INTRINSIC MAX
+    REAL(sp) :: dummydiff_b
+    REAL(sp) :: temp_b
     INTEGER :: branch
     INTEGER :: chunk_start
     INTEGER :: chunk_end
-    REAL(sp) :: temp_b
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
@@ -15159,60 +15330,11 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
     ac_prcp = ac_prcp + ac_mlt
 ! Beta percolation parameter is time step dependent
     beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
-! Interception with OPENMP
-!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
-!$OMP&ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), PRIVATE(&
-!$OMP&chunk_start, chunk_end)
-    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
-    DO col=chunk_start,chunk_end
-      DO row=1,mesh%nrow
-        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0) THEN
-          CALL PUSHCONTROL2B(0)
-        ELSE
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            CALL PUSHREAL4(ac_hi(k))
-            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
-&                          k), pn(k), en(k))
-            CALL PUSHCONTROL2B(2)
-          ELSE
-            pn(k) = 0._sp
-            en(k) = 0._sp
-            CALL PUSHCONTROL2B(1)
-          END IF
-        END IF
-      END DO
-    END DO
-!$OMP END PARALLEL
-! Forward MLP without OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0) THEN
-          CALL PUSHCONTROL2B(0)
-        ELSE
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            CALL PUSHREAL4ARRAY(input_layer, setup%neurons(1))
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
-&                      weight_3, bias_3, input_layer, output_layer(:, k)&
-&                     )
-            CALL PUSHCONTROL2B(2)
-          ELSE
-            output_layer(:, k) = 0._sp
-            CALL PUSHCONTROL2B(1)
-          END IF
-        END IF
-      END DO
-    END DO
-! Production and transfer with OPENMP
 !$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, output_layer, ac_prcp, ac_pet, ac_cp, beta, ac_ct, &
-!$OMP&ac_kexc, ac_hp, ac_ht, ac_qt, pn, en), PRIVATE(row, col, k, &
-!$OMP&time_step_returns, pr, perc, l, prr, prd, qr, qd), PRIVATE(&
-!$OMP&chunk_start, chunk_end)
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
+!$OMP&ac_hi, ac_hp, ac_ht, ac_qt), PRIVATE(row, col, k, &
+!$OMP&time_step_returns, pn, en, pr, perc, l, prr, prd, qr, qd, split), &
+!$OMP&PRIVATE(chunk_start, chunk_end)
     CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
     DO col=chunk_start,chunk_end
       DO row=1,mesh%nrow
@@ -15220,17 +15342,19 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
 &           local_active_cell(row, col) .EQ. 0) THEN
           CALL PUSHCONTROL1B(0)
         ELSE
-          CALL PUSHINTEGER4(k)
           k = mesh%rowcol_to_ind_ac(row, col)
           IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4(en)
+            CALL PUSHREAL4(pn)
+            CALL PUSHREAL4(ac_hi(k))
+            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
+&                          k), pn, en)
             CALL PUSHREAL4(perc)
             CALL PUSHREAL4(pr)
             CALL PUSHREAL4(ac_hp(k))
-            CALL GR_PRODUCTION(output_layer(1, k), output_layer(2, k), &
-&                        pn(k), en(k), ac_cp(k), beta, ac_hp(k), pr, &
-&                        perc)
-            CALL GR_EXCHANGE(output_layer(4, k), ac_kexc(k), ac_ht(k), l&
-&                     )
+            CALL GR_RI_PRODUCTION(pn, en, ac_cp(k), beta, ac_alpha1(k), &
+&                           ac_hp(k), pr, perc, setup%dt)
+            CALL GR_EXCHANGE(0._sp, ac_kexc(k), ac_ht(k), l)
             CALL PUSHCONTROL1B(1)
           ELSE
             CALL PUSHREAL4(pr)
@@ -15240,9 +15364,11 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
             l = 0._sp
             CALL PUSHCONTROL1B(0)
           END IF
+          CALL PUSHREAL4(split)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
           CALL PUSHREAL4(prr)
-          prr = 0.9_sp*(1._sp-output_layer(3, k)**2)*(pr+perc) + l
-          prd = (0.1_sp+0.9_sp*output_layer(3, k)**2)*(pr+perc)
+          prr = (1._sp-split)*(pr+perc) + l
+          prd = split*(pr+perc)
           CALL PUSHREAL4(ac_ht(k))
           CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
 &                    qr)
@@ -15257,23 +15383,29 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
     END DO
     CALL PUSHREAL4(pr)
     CALL PUSHREAL4(perc)
+    CALL PUSHREAL4(pn)
     CALL PUSHREAL4(prr)
-    CALL PUSHINTEGER4(k)
+    CALL PUSHREAL4(split)
+    CALL PUSHREAL4(en)
 !$OMP END PARALLEL
-    output_layer_b = 0.0_4
-    en_b = 0.0_4
+    ac_prcp_b = 0.0_4
     pn_b = 0.0_4
 !$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, output_layer, ac_prcp, ac_pet, ac_cp, beta, ac_ct, &
-!$OMP&ac_kexc, ac_hp, ac_ht, ac_qt, pn, en), SHARED(output_layer_b, &
-!$OMP&ac_prcp_b, ac_cp_b, ac_ct_b, ac_kexc_b, ac_hp_b, ac_ht_b, ac_qt_b&
-!$OMP&, pn_b, en_b), PRIVATE(row, col, k, time_step_returns, pr, perc, l&
-!$OMP&, prr, prd, qr, qd), PRIVATE(pr_b, perc_b, l_b, prr_b, prd_b, qr_b&
-!$OMP&, qd_b), PRIVATE(branch, chunk_end, chunk_start), PRIVATE(temp_b)
-    CALL POPINTEGER4(k)
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
+!$OMP&ac_hi, ac_hp, ac_ht, ac_qt), SHARED(ac_prcp_b, ac_ci_b, ac_cp_b, &
+!$OMP&ac_ct_b, ac_kexc_b, ac_hi_b, ac_hp_b, ac_ht_b, ac_qt_b), PRIVATE(&
+!$OMP&row, col, k, time_step_returns, pn, en, pr, perc, l, prr, prd, qr&
+!$OMP&, qd, split), PRIVATE(pn_b, en_b, pr_b, perc_b, l_b, prr_b, prd_b&
+!$OMP&, qr_b, qd_b, split_b), PRIVATE(branch, chunk_end, chunk_start), &
+!$OMP&PRIVATE(temp_b)
+    CALL POPREAL4(en)
+    CALL POPREAL4(split)
     CALL POPREAL4(prr)
+    CALL POPREAL4(pn)
     CALL POPREAL4(perc)
     CALL POPREAL4(pr)
+    pn_b = 0.0_4
+    en_b = 0.0_4
     pr_b = 0.0_4
     perc_b = 0.0_4
     l_b = 0.0_4
@@ -15281,11 +15413,14 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
     prd_b = 0.0_4
     qr_b = 0.0_4
     qd_b = 0.0_4
+    split_b = 0.0_4
+    pn_b = 0.0_4
     CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
     DO col=chunk_end,chunk_start,-1
       DO row=mesh%nrow,1,-1
         CALL POPCONTROL1B(branch)
         IF (branch .NE. 0) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
           ac_qt_b(k) = mesh%dx(row, col)*1e-3_sp*mesh%dy(row, col)*&
 &           ac_qt_b(k)/setup%dt
           qr_b = ac_qt_b(k)
@@ -15302,99 +15437,49 @@ SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
           CALL POPREAL4(ac_ht(k))
           CALL GR_TRANSFER_B(5._sp, ac_prcp(k), prr, prr_b, ac_ct(k), &
 &                      ac_ct_b(k), ac_ht(k), ac_ht_b(k), qr, qr_b)
-!$OMP     ATOMIC update
-          output_layer_b(3, k) = output_layer_b(3, k) + 2*output_layer(3&
-&           , k)*0.9_sp*(pr+perc)*prd_b - 2*output_layer(3, k)*(pr+perc)&
-&           *0.9_sp*prr_b
-          temp_b = (0.9_sp*output_layer(3, k)**2+0.1_sp)*prd_b
-          pr_b = temp_b
-          perc_b = temp_b
+          split_b = (pr+perc)*prd_b - (pr+perc)*prr_b
+          pr_b = split*prd_b + (1._sp-split)*prr_b
+          perc_b = split*prd_b + (1._sp-split)*prr_b
           CALL POPREAL4(prr)
-          temp_b = (1._sp-output_layer(3, k)**2)*0.9_sp*prr_b
           l_b = l_b + prr_b
-          pr_b = pr_b + temp_b
-          perc_b = perc_b + temp_b
+          CALL POPREAL4(split)
+          temp_b = (1.0-TANH(ac_alpha2(k)*pn)**2)*2*TANH(ac_alpha2(k)*pn&
+&           )*0.9_sp*split_b
+!$OMP     ATOMIC update
+          ac_alpha2_b(k) = ac_alpha2_b(k) + pn*temp_b
+          pn_b = pn_b + ac_alpha2(k)*temp_b
           CALL POPCONTROL1B(branch)
           IF (branch .EQ. 0) THEN
             CALL POPREAL4(perc)
             CALL POPREAL4(pr)
           ELSE
-            CALL GR_EXCHANGE_B(output_layer(4, k), output_layer_b(4, k)&
-&                        , ac_kexc(k), ac_kexc_b(k), ac_ht(k), ac_ht_b(k&
-&                        ), l, l_b)
+            CALL GR_EXCHANGE_B(0._sp, dummydiff_b, ac_kexc(k), ac_kexc_b&
+&                        (k), ac_ht(k), ac_ht_b(k), l, l_b)
             CALL POPREAL4(ac_hp(k))
             CALL POPREAL4(pr)
             CALL POPREAL4(perc)
-            CALL GR_PRODUCTION_B(output_layer(1, k), output_layer_b(1, k&
-&                          ), output_layer(2, k), output_layer_b(2, k), &
-&                          pn(k), pn_b(k), en(k), en_b(k), ac_cp(k), &
-&                          ac_cp_b(k), beta, ac_hp(k), ac_hp_b(k), pr, &
-&                          pr_b, perc, perc_b)
-          END IF
-          CALL POPINTEGER4(k)
-        END IF
-      END DO
-    END DO
-!$OMP END PARALLEL
-    DO col=mesh%ncol,1,-1
-      DO row=mesh%nrow,1,-1
-        CALL POPCONTROL2B(branch)
-        IF (branch .NE. 0) THEN
-          IF (branch .EQ. 1) THEN
-            k = mesh%rowcol_to_ind_ac(row, col)
-            output_layer_b(:, k) = 0.0_4
-          ELSE
-            k = mesh%rowcol_to_ind_ac(row, col)
-            CALL FORWARD_MLP_B(weight_1, weight_1_b, bias_1, bias_1_b, &
-&                        weight_2, weight_2_b, bias_2, bias_2_b, &
-&                        weight_3, weight_3_b, bias_3, bias_3_b, &
-&                        input_layer, input_layer_b, output_layer(:, k)&
-&                        , output_layer_b(:, k))
-            output_layer_b(:, k) = 0.0_4
-            CALL POPREAL4ARRAY(input_layer, setup%neurons(1))
-            ac_hp_b(k) = ac_hp_b(k) + input_layer_b(1)
-            ac_ht_b(k) = ac_ht_b(k) + input_layer_b(2)
-            pn_b(k) = pn_b(k) + input_layer_b(3)
-            en_b(k) = en_b(k) + input_layer_b(4)
-          END IF
-        END IF
-      END DO
-    END DO
-    ac_prcp_b = 0.0_4
-!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
-!$OMP&ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_b, ac_ci_b, ac_hi_b&
-!$OMP&, pn_b, en_b), PRIVATE(row, col, k), PRIVATE(branch, chunk_end, &
-!$OMP&chunk_start)
-    en_b = 0.0_4
-    pn_b = 0.0_4
-    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
-    DO col=chunk_end,chunk_start,-1
-      DO row=mesh%nrow,1,-1
-        CALL POPCONTROL2B(branch)
-        IF (branch .NE. 0) THEN
-          IF (branch .EQ. 1) THEN
-            k = mesh%rowcol_to_ind_ac(row, col)
-            en_b(k) = 0.0_4
-            pn_b(k) = 0.0_4
-          ELSE
-            k = mesh%rowcol_to_ind_ac(row, col)
-            CALL POPREAL4(ac_hi(k))
-            CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
-&                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
-&                            , pn(k), pn_b(k), en(k), en_b(k))
-            pn_b(k) = 0.0_4
-            en_b(k) = 0.0_4
+            CALL GR_RI_PRODUCTION_B(pn, pn_b, en, en_b, ac_cp(k), &
+&                             ac_cp_b(k), beta, ac_alpha1(k), &
+&                             ac_alpha1_b(k), ac_hp(k), ac_hp_b(k), pr, &
+&                             pr_b, perc, perc_b, setup%dt)
+            CALL POPREAL4(ac_hi(k))
+            CALL POPREAL4(pn)
+            CALL POPREAL4(en)
+            CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
+&                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
+&                            , pn, pn_b, en, en_b)
+            pn_b = 0.0_4
           END IF
         END IF
       END DO
     END DO
 !$OMP END PARALLEL
     ac_mlt_b = ac_mlt_b + ac_prcp_b
-  END SUBROUTINE GR4_MLP_TIME_STEP_B
+  END SUBROUTINE GR4_RI_TIME_STEP_B
 
-  SUBROUTINE GR4_MLP_TIME_STEP(setup, mesh, input_data, options, returns&
-&   , time_step, weight_1, bias_1, weight_2, bias_2, weight_3, bias_3, &
-&   ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, ac_ht, ac_qt)
+  SUBROUTINE GR4_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&   , time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_alpha1, ac_alpha2, &
+&   ac_kexc, ac_hi, ac_hp, ac_ht, ac_qt)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15402,26 +15487,16 @@ SUBROUTINE GR4_MLP_TIME_STEP(setup, mesh, input_data, options, returns&
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
 &   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    INTRINSIC TANH
     INTRINSIC MAX
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
@@ -15430,65 +15505,30 @@ SUBROUTINE GR4_MLP_TIME_STEP(setup, mesh, input_data, options, returns&
     ac_prcp = ac_prcp + ac_mlt
 ! Beta percolation parameter is time step dependent
     beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
-! Interception with OPENMP
-!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
-!$OMP&ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), SCHEDULE(&
-!$OMP&                                     static)
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
-&                          k), pn(k), en(k))
-          ELSE
-            pn(k) = 0._sp
-            en(k) = 0._sp
-          END IF
-        END IF
-      END DO
-    END DO
-! Forward MLP without OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
-&                      weight_3, bias_3, input_layer, output_layer(:, k)&
-&                     )
-          ELSE
-            output_layer(:, k) = 0._sp
-          END IF
-        END IF
-      END DO
-    END DO
-! Production and transfer with OPENMP
 !$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, output_layer, ac_prcp, ac_pet, ac_cp, beta, ac_ct, &
-!$OMP&ac_kexc, ac_hp, ac_ht, ac_qt, pn, en), PRIVATE(row, col, k, &
-!$OMP&time_step_returns, pr, perc, l, prr, prd, qr, qd), SCHEDULE(static)
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
+!$OMP&ac_hi, ac_hp, ac_ht, ac_qt), PRIVATE(row, col, k, &
+!$OMP&time_step_returns, pn, en, pr, perc, l, prr, prd, qr, qd, split), &
+!$OMP&                                                  SCHEDULE(static)
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
 &           local_active_cell(row, col) .EQ. 0)) THEN
           k = mesh%rowcol_to_ind_ac(row, col)
           IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            CALL GR_PRODUCTION(output_layer(1, k), output_layer(2, k), &
-&                        pn(k), en(k), ac_cp(k), beta, ac_hp(k), pr, &
-&                        perc)
-            CALL GR_EXCHANGE(output_layer(4, k), ac_kexc(k), ac_ht(k), l&
-&                     )
+            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
+&                          k), pn, en)
+            CALL GR_RI_PRODUCTION(pn, en, ac_cp(k), beta, ac_alpha1(k), &
+&                           ac_hp(k), pr, perc, setup%dt)
+            CALL GR_EXCHANGE(0._sp, ac_kexc(k), ac_ht(k), l)
           ELSE
             pr = 0._sp
             perc = 0._sp
             l = 0._sp
           END IF
-          prr = 0.9_sp*(1._sp-output_layer(3, k)**2)*(pr+perc) + l
-          prd = (0.1_sp+0.9_sp*output_layer(3, k)**2)*(pr+perc)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
+          prr = (1._sp-split)*(pr+perc) + l
+          prd = split*(pr+perc)
           CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
 &                    qr)
           IF (0._sp .LT. prd + l) THEN
@@ -15503,16 +15543,19 @@ SUBROUTINE GR4_MLP_TIME_STEP(setup, mesh, input_data, options, returns&
         END IF
       END DO
     END DO
-  END SUBROUTINE GR4_MLP_TIME_STEP
+  END SUBROUTINE GR4_RI_TIME_STEP
 
-!  Differentiation of gr4_ode_time_step in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
+!  Differentiation of gr4_mlp_time_step in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
 !   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt
-  SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
-&   returns, time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d&
-&   , ac_ct, ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d&
-&   , ac_ht, ac_ht_d, ac_qt, ac_qt_d)
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+  SUBROUTINE GR4_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
+&   returns, time_step, weight_1, weight_1_d, bias_1, bias_1_d, weight_2&
+&   , weight_2_d, bias_2, bias_2_d, weight_3, weight_3_d, bias_3, &
+&   bias_3_d, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
+&   ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d, ac_ht, &
+&   ac_ht_d, ac_qt, ac_qt_d)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15520,6 +15563,24 @@ SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1_d
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1_d
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2_d
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2_d
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3_d
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
@@ -15531,10 +15592,18 @@ SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
 &   ac_ht_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_d
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_d
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer_d
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d, pn_d, en_d
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: l
+    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d
+    INTRINSIC MAX
     REAL(sp) :: temp
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
@@ -15542,12 +15611,14 @@ SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
 &                              , 'pet', ac_pet)
     ac_prcp_d = ac_mlt_d
     ac_prcp = ac_prcp + ac_mlt
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
     en_d = 0.0_4
     pn_d = 0.0_4
 ! Interception with OPENMP
-!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&ac_prcp, ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_d, ac_ci_d&
-!$OMP&, ac_hi_d, pn_d, en_d), PRIVATE(row, col, k), SCHEDULE(static)
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
+!$OMP&ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_d, ac_ci_d, ac_hi_d&
+!$OMP&, pn_d, en_d), PRIVATE(row, col, k), SCHEDULE(static)
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -15566,18 +15637,78 @@ SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-! Production and transfer without OPENMP
+    output_layer_d = 0.0_4
+! Forward MLP without OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
 &           local_active_cell(row, col) .EQ. 0)) THEN
           k = mesh%rowcol_to_ind_ac(row, col)
-          CALL GR_PRODUCTION_TRANSFER_ODE_D(pn(k), pn_d(k), en(k), en_d(&
-&                                     k), ac_cp(k), ac_cp_d(k), ac_ct(k)&
-&                                     , ac_ct_d(k), ac_kexc(k), &
-&                                     ac_kexc_d(k), ac_hp(k), ac_hp_d(k)&
-&                                     , ac_ht(k), ac_ht_d(k), ac_qt(k), &
-&                                     ac_qt_d(k), l)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            input_layer_d(:) = (/ac_hp_d(k), ac_ht_d(k), pn_d(k), en_d(k&
+&             )/)
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP_D(weight_1, weight_1_d, bias_1, bias_1_d, &
+&                        weight_2, weight_2_d, bias_2, bias_2_d, &
+&                        weight_3, weight_3_d, bias_3, bias_3_d, &
+&                        input_layer, input_layer_d, output_layer(:, k)&
+&                        , output_layer_d(:, k))
+          ELSE
+            output_layer_d(:, k) = 0.0_4
+            output_layer(:, k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, output_layer, ac_prcp, ac_pet, ac_cp, beta, ac_ct, &
+!$OMP&ac_kexc, ac_hp, ac_ht, ac_qt, pn, en), SHARED(output_layer_d, &
+!$OMP&ac_prcp_d, ac_cp_d, ac_ct_d, ac_kexc_d, ac_hp_d, ac_ht_d, ac_qt_d&
+!$OMP&, pn_d, en_d), PRIVATE(row, col, k, time_step_returns, pr, perc, l&
+!$OMP&, prr, prd, qr, qd), PRIVATE(pr_d, perc_d, l_d, prr_d, prd_d, qr_d&
+!$OMP&, qd_d), PRIVATE(temp), SCHEDULE(static)
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_PRODUCTION_D(output_layer(1, k), output_layer_d(1, k&
+&                          ), output_layer(2, k), output_layer_d(2, k), &
+&                          pn(k), pn_d(k), en(k), en_d(k), ac_cp(k), &
+&                          ac_cp_d(k), beta, ac_hp(k), ac_hp_d(k), pr, &
+&                          pr_d, perc, perc_d)
+            CALL GR_EXCHANGE_D(output_layer(4, k), output_layer_d(4, k)&
+&                        , ac_kexc(k), ac_kexc_d(k), ac_ht(k), ac_ht_d(k&
+&                        ), l, l_d)
+          ELSE
+            pr = 0._sp
+            perc = 0._sp
+            l = 0._sp
+            l_d = 0.0_4
+            perc_d = 0.0_4
+            pr_d = 0.0_4
+          END IF
+          temp = -(output_layer(3, k)*output_layer(3, k)) + 1._sp
+          prr_d = 0.9_sp*(temp*(pr_d+perc_d)-(pr+perc)*2*output_layer(3&
+&           , k)*output_layer_d(3, k)) + l_d
+          prr = 0.9_sp*(temp*(pr+perc)) + l
+          temp = 0.9_sp*(output_layer(3, k)*output_layer(3, k)) + 0.1_sp
+          prd_d = (pr+perc)*0.9_sp*2*output_layer(3, k)*output_layer_d(3&
+&           , k) + temp*(pr_d+perc_d)
+          prd = temp*(pr+perc)
+          CALL GR_TRANSFER_D(5._sp, ac_prcp(k), prr, prr_d, ac_ct(k), &
+&                      ac_ct_d(k), ac_ht(k), ac_ht_d(k), qr, qr_d)
+          IF (0._sp .LT. prd + l) THEN
+            qd_d = prd_d + l_d
+            qd = prd + l
+          ELSE
+            qd = 0._sp
+            qd_d = 0.0_4
+          END IF
+          ac_qt_d(k) = qr_d + qd_d
+          ac_qt(k) = qr + qd
 ! Transform from mm/dt to m3/s
           temp = 1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)
           ac_qt_d(k) = temp*ac_qt_d(k)/setup%dt
@@ -15585,17 +15716,21 @@ SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-  END SUBROUTINE GR4_ODE_TIME_STEP_D
+  END SUBROUTINE GR4_MLP_TIME_STEP_D
 
-!  Differentiation of gr4_ode_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context OpenMP):
-!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt
-  SUBROUTINE GR4_ODE_TIME_STEP_B(setup, mesh, input_data, options, &
-&   returns, time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b&
-&   , ac_ct, ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b&
-&   , ac_ht, ac_ht_b, ac_qt, ac_qt_b)
+!  Differentiation of gr4_mlp_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context OpenMP):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+  SUBROUTINE GR4_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
+&   returns, time_step, weight_1, weight_1_b, bias_1, bias_1_b, weight_2&
+&   , weight_2_b, bias_2, bias_2_b, weight_3, weight_3_b, bias_3, &
+&   bias_3_b, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
+&   ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b, ac_ht, &
+&   ac_ht_b, ac_qt, ac_qt_b)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15603,33 +15738,62 @@ SUBROUTINE GR4_ODE_TIME_STEP_B(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
-    REAL(sp), DIMENSION(mesh%nac) :: ac_mlt_b
-    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
-&   ac_kexc
-    REAL(sp), DIMENSION(mesh%nac) :: ac_ci_b, ac_cp_b, ac_ct_b, &
-&   ac_kexc_b
-    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
-    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_b, ac_hp_b, &
-&   ac_ht_b
-    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
-    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_b
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)) :: &
+&   weight_1_b
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(2)) :: bias_1_b
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)) :: &
+&   weight_2_b
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(3)) :: bias_2_b
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)) :: &
+&   weight_3_b
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
+    REAL(sp), DIMENSION(setup%neurons(4)) :: bias_3_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac) :: ac_mlt_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac) :: ac_ci_b, ac_cp_b, ac_ct_b, &
+&   ac_kexc_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_b, ac_hp_b, &
+&   ac_ht_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_b
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_b
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer_b
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b, pn_b, en_b
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: l
+    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b
+    INTRINSIC MAX
     INTEGER :: branch
     INTEGER :: chunk_start
     INTEGER :: chunk_end
+    REAL(sp) :: temp_b
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'pet', ac_pet)
     ac_prcp = ac_prcp + ac_mlt
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
 ! Interception with OPENMP
-!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&ac_prcp, ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), &
-!$OMP&PRIVATE(chunk_start, chunk_end)
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
+!$OMP&ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), PRIVATE(&
+!$OMP&chunk_start, chunk_end)
     CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
     DO col=chunk_start,chunk_end
       DO row=1,mesh%nrow
@@ -15651,57 +15815,187 @@ SUBROUTINE GR4_ODE_TIME_STEP_B(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-    CALL PUSHREAL4ARRAY(ac_prcp, mesh%nac)
-    CALL PUSHINTEGER4(time_step)
 !$OMP END PARALLEL
-! Production and transfer without OPENMP
+! Forward MLP without OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0) THEN
+          CALL PUSHCONTROL2B(0)
+        ELSE
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4ARRAY(input_layer, setup%neurons(1))
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
+&                      weight_3, bias_3, input_layer, output_layer(:, k)&
+&                     )
+            CALL PUSHCONTROL2B(2)
+          ELSE
+            output_layer(:, k) = 0._sp
+            CALL PUSHCONTROL2B(1)
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, output_layer, ac_prcp, ac_pet, ac_cp, beta, ac_ct, &
+!$OMP&ac_kexc, ac_hp, ac_ht, ac_qt, pn, en), PRIVATE(row, col, k, &
+!$OMP&time_step_returns, pr, perc, l, prr, prd, qr, qd), PRIVATE(&
+!$OMP&chunk_start, chunk_end)
+    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
+    DO col=chunk_start,chunk_end
+      DO row=1,mesh%nrow
+        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
 &           local_active_cell(row, col) .EQ. 0) THEN
           CALL PUSHCONTROL1B(0)
         ELSE
+          CALL PUSHINTEGER4(k)
           k = mesh%rowcol_to_ind_ac(row, col)
-          CALL PUSHREAL4(ac_qt(k))
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4(perc)
+            CALL PUSHREAL4(pr)
+            CALL PUSHREAL4(ac_hp(k))
+            CALL GR_PRODUCTION(output_layer(1, k), output_layer(2, k), &
+&                        pn(k), en(k), ac_cp(k), beta, ac_hp(k), pr, &
+&                        perc)
+            CALL GR_EXCHANGE(output_layer(4, k), ac_kexc(k), ac_ht(k), l&
+&                     )
+            CALL PUSHCONTROL1B(1)
+          ELSE
+            CALL PUSHREAL4(pr)
+            pr = 0._sp
+            CALL PUSHREAL4(perc)
+            perc = 0._sp
+            l = 0._sp
+            CALL PUSHCONTROL1B(0)
+          END IF
+          CALL PUSHREAL4(prr)
+          prr = 0.9_sp*(1._sp-output_layer(3, k)**2)*(pr+perc) + l
+          prd = (0.1_sp+0.9_sp*output_layer(3, k)**2)*(pr+perc)
           CALL PUSHREAL4(ac_ht(k))
-          CALL PUSHREAL4(ac_hp(k))
-          CALL GR_PRODUCTION_TRANSFER_ODE(pn(k), en(k), ac_cp(k), ac_ct(&
-&                                   k), ac_kexc(k), ac_hp(k), ac_ht(k), &
-&                                   ac_qt(k), l)
-! Transform from mm/dt to m3/s
+          CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
+&                    qr)
+          IF (0._sp .LT. prd + l) THEN
+            CALL PUSHCONTROL1B(0)
+          ELSE
+            CALL PUSHCONTROL1B(1)
+          END IF
           CALL PUSHCONTROL1B(1)
         END IF
       END DO
     END DO
+    CALL PUSHREAL4(pr)
+    CALL PUSHREAL4(perc)
+    CALL PUSHREAL4(prr)
+    CALL PUSHINTEGER4(k)
+!$OMP END PARALLEL
+    output_layer_b = 0.0_4
     en_b = 0.0_4
     pn_b = 0.0_4
-    DO col=mesh%ncol,1,-1
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, output_layer, ac_prcp, ac_pet, ac_cp, beta, ac_ct, &
+!$OMP&ac_kexc, ac_hp, ac_ht, ac_qt, pn, en), SHARED(output_layer_b, &
+!$OMP&ac_prcp_b, ac_cp_b, ac_ct_b, ac_kexc_b, ac_hp_b, ac_ht_b, ac_qt_b&
+!$OMP&, pn_b, en_b), PRIVATE(row, col, k, time_step_returns, pr, perc, l&
+!$OMP&, prr, prd, qr, qd), PRIVATE(pr_b, perc_b, l_b, prr_b, prd_b, qr_b&
+!$OMP&, qd_b), PRIVATE(branch, chunk_end, chunk_start), PRIVATE(temp_b)
+    CALL POPINTEGER4(k)
+    CALL POPREAL4(prr)
+    CALL POPREAL4(perc)
+    CALL POPREAL4(pr)
+    pr_b = 0.0_4
+    perc_b = 0.0_4
+    l_b = 0.0_4
+    prr_b = 0.0_4
+    prd_b = 0.0_4
+    qr_b = 0.0_4
+    qd_b = 0.0_4
+    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
+    DO col=chunk_end,chunk_start,-1
       DO row=mesh%nrow,1,-1
         CALL POPCONTROL1B(branch)
         IF (branch .NE. 0) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
           ac_qt_b(k) = mesh%dx(row, col)*1e-3_sp*mesh%dy(row, col)*&
 &           ac_qt_b(k)/setup%dt
-          CALL POPREAL4(ac_hp(k))
-          CALL POPREAL4(ac_ht(k))
-          CALL POPREAL4(ac_qt(k))
-          CALL GR_PRODUCTION_TRANSFER_ODE_B(pn(k), pn_b(k), en(k), en_b(&
-&                                     k), ac_cp(k), ac_cp_b(k), ac_ct(k)&
-&                                     , ac_ct_b(k), ac_kexc(k), &
-&                                     ac_kexc_b(k), ac_hp(k), ac_hp_b(k)&
-&                                     , ac_ht(k), ac_ht_b(k), ac_qt(k), &
-&                                     ac_qt_b(k), l)
+          qr_b = ac_qt_b(k)
+          qd_b = ac_qt_b(k)
           ac_qt_b(k) = 0.0_4
+          CALL POPCONTROL1B(branch)
+          IF (branch .EQ. 0) THEN
+            prd_b = qd_b
+            l_b = qd_b
+          ELSE
+            l_b = 0.0_4
+            prd_b = 0.0_4
+          END IF
+          CALL POPREAL4(ac_ht(k))
+          CALL GR_TRANSFER_B(5._sp, ac_prcp(k), prr, prr_b, ac_ct(k), &
+&                      ac_ct_b(k), ac_ht(k), ac_ht_b(k), qr, qr_b)
+!$OMP     ATOMIC update
+          output_layer_b(3, k) = output_layer_b(3, k) + 2*output_layer(3&
+&           , k)*0.9_sp*(pr+perc)*prd_b - 2*output_layer(3, k)*(pr+perc)&
+&           *0.9_sp*prr_b
+          temp_b = (0.9_sp*output_layer(3, k)**2+0.1_sp)*prd_b
+          pr_b = temp_b
+          perc_b = temp_b
+          CALL POPREAL4(prr)
+          temp_b = (1._sp-output_layer(3, k)**2)*0.9_sp*prr_b
+          l_b = l_b + prr_b
+          pr_b = pr_b + temp_b
+          perc_b = perc_b + temp_b
+          CALL POPCONTROL1B(branch)
+          IF (branch .EQ. 0) THEN
+            CALL POPREAL4(perc)
+            CALL POPREAL4(pr)
+          ELSE
+            CALL GR_EXCHANGE_B(output_layer(4, k), output_layer_b(4, k)&
+&                        , ac_kexc(k), ac_kexc_b(k), ac_ht(k), ac_ht_b(k&
+&                        ), l, l_b)
+            CALL POPREAL4(ac_hp(k))
+            CALL POPREAL4(pr)
+            CALL POPREAL4(perc)
+            CALL GR_PRODUCTION_B(output_layer(1, k), output_layer_b(1, k&
+&                          ), output_layer(2, k), output_layer_b(2, k), &
+&                          pn(k), pn_b(k), en(k), en_b(k), ac_cp(k), &
+&                          ac_cp_b(k), beta, ac_hp(k), ac_hp_b(k), pr, &
+&                          pr_b, perc, perc_b)
+          END IF
+          CALL POPINTEGER4(k)
+        END IF
+      END DO
+    END DO
+!$OMP END PARALLEL
+    DO col=mesh%ncol,1,-1
+      DO row=mesh%nrow,1,-1
+        CALL POPCONTROL2B(branch)
+        IF (branch .NE. 0) THEN
+          IF (branch .EQ. 1) THEN
+            k = mesh%rowcol_to_ind_ac(row, col)
+            output_layer_b(:, k) = 0.0_4
+          ELSE
+            k = mesh%rowcol_to_ind_ac(row, col)
+            CALL FORWARD_MLP_B(weight_1, weight_1_b, bias_1, bias_1_b, &
+&                        weight_2, weight_2_b, bias_2, bias_2_b, &
+&                        weight_3, weight_3_b, bias_3, bias_3_b, &
+&                        input_layer, input_layer_b, output_layer(:, k)&
+&                        , output_layer_b(:, k))
+            output_layer_b(:, k) = 0.0_4
+            CALL POPREAL4ARRAY(input_layer, setup%neurons(1))
+            ac_hp_b(k) = ac_hp_b(k) + input_layer_b(1)
+            ac_ht_b(k) = ac_ht_b(k) + input_layer_b(2)
+            pn_b(k) = pn_b(k) + input_layer_b(3)
+            en_b(k) = en_b(k) + input_layer_b(4)
+          END IF
         END IF
       END DO
     END DO
     ac_prcp_b = 0.0_4
-!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&ac_prcp, ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_b, ac_ci_b&
-!$OMP&, ac_hi_b, pn_b, en_b), PRIVATE(row, col, k), PRIVATE(branch, &
-!$OMP&chunk_end, chunk_start)
-    CALL POPINTEGER4(time_step)
-    CALL POPREAL4ARRAY(ac_prcp, mesh%nac)
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
+!$OMP&ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_b, ac_ci_b, ac_hi_b&
+!$OMP&, pn_b, en_b), PRIVATE(row, col, k), PRIVATE(branch, chunk_end, &
+!$OMP&chunk_start)
     en_b = 0.0_4
     pn_b = 0.0_4
     CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
@@ -15727,11 +16021,11 @@ SUBROUTINE GR4_ODE_TIME_STEP_B(setup, mesh, input_data, options, &
     END DO
 !$OMP END PARALLEL
     ac_mlt_b = ac_mlt_b + ac_prcp_b
-  END SUBROUTINE GR4_ODE_TIME_STEP_B
+  END SUBROUTINE GR4_MLP_TIME_STEP_B
 
-  SUBROUTINE GR4_ODE_TIME_STEP(setup, mesh, input_data, options, returns&
-&   , time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, &
-&   ac_ht, ac_qt)
+  SUBROUTINE GR4_MLP_TIME_STEP(setup, mesh, input_data, options, returns&
+&   , time_step, weight_1, bias_1, weight_2, bias_2, weight_3, bias_3, &
+&   ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, ac_ht, ac_qt)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15739,23 +16033,38 @@ SUBROUTINE GR4_ODE_TIME_STEP(setup, mesh, input_data, options, returns&
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
 &   ac_kexc
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: l
+    REAL(sp) :: beta, pr, perc, l, prr, prd, qr, qd
+    INTRINSIC MAX
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'pet', ac_pet)
     ac_prcp = ac_prcp + ac_mlt
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
 ! Interception with OPENMP
-!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&ac_prcp, ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), &
-!$OMP&                                              SCHEDULE(static)
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
+!$OMP&ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), SCHEDULE(&
+!$OMP&                                     static)
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -15771,34 +16080,70 @@ SUBROUTINE GR4_ODE_TIME_STEP(setup, mesh, input_data, options, returns&
         END IF
       END DO
     END DO
-! Production and transfer without OPENMP
+! Forward MLP without OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
 &           local_active_cell(row, col) .EQ. 0)) THEN
           k = mesh%rowcol_to_ind_ac(row, col)
-          CALL GR_PRODUCTION_TRANSFER_ODE(pn(k), en(k), ac_cp(k), ac_ct(&
-&                                   k), ac_kexc(k), ac_hp(k), ac_ht(k), &
-&                                   ac_qt(k), l)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
+&                      weight_3, bias_3, input_layer, output_layer(:, k)&
+&                     )
+          ELSE
+            output_layer(:, k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, output_layer, ac_prcp, ac_pet, ac_cp, beta, ac_ct, &
+!$OMP&ac_kexc, ac_hp, ac_ht, ac_qt, pn, en), PRIVATE(row, col, k, &
+!$OMP&time_step_returns, pr, perc, l, prr, prd, qr, qd), SCHEDULE(static)
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_PRODUCTION(output_layer(1, k), output_layer(2, k), &
+&                        pn(k), en(k), ac_cp(k), beta, ac_hp(k), pr, &
+&                        perc)
+            CALL GR_EXCHANGE(output_layer(4, k), ac_kexc(k), ac_ht(k), l&
+&                     )
+          ELSE
+            pr = 0._sp
+            perc = 0._sp
+            l = 0._sp
+          END IF
+          prr = 0.9_sp*(1._sp-output_layer(3, k)**2)*(pr+perc) + l
+          prd = (0.1_sp+0.9_sp*output_layer(3, k)**2)*(pr+perc)
+          CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
+&                    qr)
+          IF (0._sp .LT. prd + l) THEN
+            qd = prd + l
+          ELSE
+            qd = 0._sp
+          END IF
+          ac_qt(k) = qr + qd
 ! Transform from mm/dt to m3/s
           ac_qt(k) = ac_qt(k)*1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col&
 &           )/setup%dt
         END IF
       END DO
     END DO
-  END SUBROUTINE GR4_ODE_TIME_STEP
+  END SUBROUTINE GR4_MLP_TIME_STEP
 
-!  Differentiation of gr4_ode_mlp_time_step in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
+!  Differentiation of gr4_ode_time_step in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
 !   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-  SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
-&   returns, time_step, weight_1, weight_1_d, bias_1, bias_1_d, weight_2&
-&   , weight_2_d, bias_2, bias_2_d, weight_3, weight_3_d, bias_3, &
-&   bias_3_d, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
-&   ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d, ac_ht, &
-&   ac_ht_d, ac_qt, ac_qt_d)
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt
+  SUBROUTINE GR4_ODE_TIME_STEP_D(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d&
+&   , ac_ct, ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d&
+&   , ac_ht, ac_ht_d, ac_qt, ac_qt_d)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15806,24 +16151,6 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1_d
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1_d
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2_d
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2_d
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3_d
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
@@ -15835,12 +16162,6 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
 &   ac_ht_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_d
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_d
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer_d
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d, pn_d, en_d
     INTEGER :: row, col, k, time_step_returns
@@ -15855,9 +16176,9 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
     en_d = 0.0_4
     pn_d = 0.0_4
 ! Interception with OPENMP
-!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
-!$OMP&ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_d, ac_ci_d, ac_hi_d&
-!$OMP&, pn_d, en_d), PRIVATE(row, col, k), SCHEDULE(static)
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&ac_prcp, ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_d, ac_ci_d&
+!$OMP&, ac_hi_d, pn_d, en_d), PRIVATE(row, col, k), SCHEDULE(static)
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -15876,48 +16197,18 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-    output_layer_d = 0.0_4
-! Forward MLP without OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            input_layer_d(:) = (/ac_hp_d(k), ac_ht_d(k), pn_d(k), en_d(k&
-&             )/)
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP_D(weight_1, weight_1_d, bias_1, bias_1_d, &
-&                        weight_2, weight_2_d, bias_2, bias_2_d, &
-&                        weight_3, weight_3_d, bias_3, bias_3_d, &
-&                        input_layer, input_layer_d, output_layer(:, k)&
-&                        , output_layer_d(:, k))
-          ELSE
-            output_layer_d(:, k) = 0.0_4
-            output_layer(:, k) = 0._sp
-          END IF
-        END IF
-      END DO
-    END DO
-! Production and transfer with OPENMP
-!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, output_layer, ac_cp, ac_ct, ac_kexc, ac_hp, ac_ht, ac_qt&
-!$OMP&, pn, en), SHARED(output_layer_d, ac_cp_d, ac_ct_d, ac_kexc_d, &
-!$OMP&ac_hp_d, ac_ht_d, ac_qt_d, pn_d, en_d), PRIVATE(row, col, k, &
-!$OMP&time_step_returns, l), PRIVATE(temp), SCHEDULE(static)
+! Production and transfer without OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
 &           local_active_cell(row, col) .EQ. 0)) THEN
           k = mesh%rowcol_to_ind_ac(row, col)
-          CALL GR_PRODUCTION_TRANSFER_ODE_MLP_D(output_layer(:, k), &
-&                                         output_layer_d(:, k), pn(k), &
-&                                         pn_d(k), en(k), en_d(k), ac_cp&
-&                                         (k), ac_cp_d(k), ac_ct(k), &
-&                                         ac_ct_d(k), ac_kexc(k), &
-&                                         ac_kexc_d(k), ac_hp(k), &
-&                                         ac_hp_d(k), ac_ht(k), ac_ht_d(&
-&                                         k), ac_qt(k), ac_qt_d(k), l)
+          CALL GR_PRODUCTION_TRANSFER_ODE_D(pn(k), pn_d(k), en(k), en_d(&
+&                                     k), ac_cp(k), ac_cp_d(k), ac_ct(k)&
+&                                     , ac_ct_d(k), ac_kexc(k), &
+&                                     ac_kexc_d(k), ac_hp(k), ac_hp_d(k)&
+&                                     , ac_ht(k), ac_ht_d(k), ac_qt(k), &
+&                                     ac_qt_d(k), l)
 ! Transform from mm/dt to m3/s
           temp = 1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)
           ac_qt_d(k) = temp*ac_qt_d(k)/setup%dt
@@ -15925,21 +16216,17 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
-  END SUBROUTINE GR4_ODE_MLP_TIME_STEP_D
+  END SUBROUTINE GR4_ODE_TIME_STEP_D
 
-!  Differentiation of gr4_ode_mlp_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context OpenMP):
-!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
-!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
-!                weight_3 ac_ht ac_mlt
-  SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
-&   returns, time_step, weight_1, weight_1_b, bias_1, bias_1_b, weight_2&
-&   , weight_2_b, bias_2, bias_2_b, weight_3, weight_3_b, bias_3, &
-&   bias_3_b, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
-&   ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b, ac_ht, &
-&   ac_ht_b, ac_qt, ac_qt_b)
+!  Differentiation of gr4_ode_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context OpenMP):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt
+  SUBROUTINE GR4_ODE_TIME_STEP_B(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b&
+&   , ac_ct, ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b&
+&   , ac_ht, ac_ht_b, ac_qt, ac_qt_b)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -15947,24 +16234,6 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
     TYPE(OPTIONSDT), INTENT(IN) :: options
     TYPE(RETURNSDT), INTENT(INOUT) :: returns
     INTEGER, INTENT(IN) :: time_step
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
-&   :: weight_1
-    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)) :: &
-&   weight_1_b
-    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
-    REAL(sp), DIMENSION(setup%neurons(2)) :: bias_1_b
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
-&   :: weight_2
-    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)) :: &
-&   weight_2_b
-    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
-    REAL(sp), DIMENSION(setup%neurons(3)) :: bias_2_b
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
-&   :: weight_3
-    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)) :: &
-&   weight_3_b
-    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
-    REAL(sp), DIMENSION(setup%neurons(4)) :: bias_3_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac) :: ac_mlt_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
@@ -15976,12 +16245,6 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
 &   ac_ht_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_b
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_b
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer_b
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b, pn_b, en_b
     INTEGER :: row, col, k, time_step_returns
@@ -15995,9 +16258,9 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
 &                              , 'pet', ac_pet)
     ac_prcp = ac_prcp + ac_mlt
 ! Interception with OPENMP
-!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
-!$OMP&ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), PRIVATE(&
-!$OMP&chunk_start, chunk_end)
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&ac_prcp, ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), &
+!$OMP&PRIVATE(chunk_start, chunk_end)
     CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
     DO col=chunk_start,chunk_end
       DO row=1,mesh%nrow
@@ -16019,38 +16282,13 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
         END IF
       END DO
     END DO
+    CALL PUSHREAL4ARRAY(ac_prcp, mesh%nac)
+    CALL PUSHINTEGER4(time_step)
 !$OMP END PARALLEL
-! Forward MLP without OPENMP
+! Production and transfer without OPENMP
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0) THEN
-          CALL PUSHCONTROL2B(0)
-        ELSE
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            CALL PUSHREAL4ARRAY(input_layer, setup%neurons(1))
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
-&                      weight_3, bias_3, input_layer, output_layer(:, k)&
-&                     )
-            CALL PUSHCONTROL2B(2)
-          ELSE
-            output_layer(:, k) = 0._sp
-            CALL PUSHCONTROL2B(1)
-          END IF
-        END IF
-      END DO
-    END DO
-! Production and transfer with OPENMP
-!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, output_layer, ac_cp, ac_ct, ac_kexc, ac_hp, ac_ht, ac_qt&
-!$OMP&, pn, en), PRIVATE(row, col, k, time_step_returns, l), PRIVATE(&
-!$OMP&chunk_start, chunk_end)
-    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
-    DO col=chunk_start,chunk_end
-      DO row=1,mesh%nrow
-        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
 &           local_active_cell(row, col) .EQ. 0) THEN
           CALL PUSHCONTROL1B(0)
         ELSE
@@ -16058,26 +16296,17 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
           CALL PUSHREAL4(ac_qt(k))
           CALL PUSHREAL4(ac_ht(k))
           CALL PUSHREAL4(ac_hp(k))
-          CALL GR_PRODUCTION_TRANSFER_ODE_MLP(output_layer(:, k), pn(k)&
-&                                       , en(k), ac_cp(k), ac_ct(k), &
-&                                       ac_kexc(k), ac_hp(k), ac_ht(k), &
-&                                       ac_qt(k), l)
+          CALL GR_PRODUCTION_TRANSFER_ODE(pn(k), en(k), ac_cp(k), ac_ct(&
+&                                   k), ac_kexc(k), ac_hp(k), ac_ht(k), &
+&                                   ac_qt(k), l)
 ! Transform from mm/dt to m3/s
           CALL PUSHCONTROL1B(1)
         END IF
       END DO
     END DO
-!$OMP END PARALLEL
-    output_layer_b = 0.0_4
     en_b = 0.0_4
     pn_b = 0.0_4
-!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, output_layer, ac_cp, ac_ct, ac_kexc, ac_hp, ac_ht, ac_qt&
-!$OMP&, pn, en), SHARED(output_layer_b, ac_cp_b, ac_ct_b, ac_kexc_b, &
-!$OMP&ac_hp_b, ac_ht_b, ac_qt_b, pn_b, en_b), PRIVATE(row, col, k, &
-!$OMP&time_step_returns, l), PRIVATE(branch, chunk_end, chunk_start)
-    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
-    DO col=chunk_end,chunk_start,-1
+    DO col=mesh%ncol,1,-1
       DO row=mesh%nrow,1,-1
         CALL POPCONTROL1B(branch)
         IF (branch .NE. 0) THEN
@@ -16087,48 +16316,23 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
           CALL POPREAL4(ac_hp(k))
           CALL POPREAL4(ac_ht(k))
           CALL POPREAL4(ac_qt(k))
-          CALL GR_PRODUCTION_TRANSFER_ODE_MLP_B(output_layer(:, k), &
-&                                         output_layer_b(:, k), pn(k), &
-&                                         pn_b(k), en(k), en_b(k), ac_cp&
-&                                         (k), ac_cp_b(k), ac_ct(k), &
-&                                         ac_ct_b(k), ac_kexc(k), &
-&                                         ac_kexc_b(k), ac_hp(k), &
-&                                         ac_hp_b(k), ac_ht(k), ac_ht_b(&
-&                                         k), ac_qt(k), ac_qt_b(k), l)
+          CALL GR_PRODUCTION_TRANSFER_ODE_B(pn(k), pn_b(k), en(k), en_b(&
+&                                     k), ac_cp(k), ac_cp_b(k), ac_ct(k)&
+&                                     , ac_ct_b(k), ac_kexc(k), &
+&                                     ac_kexc_b(k), ac_hp(k), ac_hp_b(k)&
+&                                     , ac_ht(k), ac_ht_b(k), ac_qt(k), &
+&                                     ac_qt_b(k), l)
           ac_qt_b(k) = 0.0_4
         END IF
       END DO
     END DO
-!$OMP END PARALLEL
-    DO col=mesh%ncol,1,-1
-      DO row=mesh%nrow,1,-1
-        CALL POPCONTROL2B(branch)
-        IF (branch .NE. 0) THEN
-          IF (branch .EQ. 1) THEN
-            k = mesh%rowcol_to_ind_ac(row, col)
-            output_layer_b(:, k) = 0.0_4
-          ELSE
-            k = mesh%rowcol_to_ind_ac(row, col)
-            CALL FORWARD_MLP_B(weight_1, weight_1_b, bias_1, bias_1_b, &
-&                        weight_2, weight_2_b, bias_2, bias_2_b, &
-&                        weight_3, weight_3_b, bias_3, bias_3_b, &
-&                        input_layer, input_layer_b, output_layer(:, k)&
-&                        , output_layer_b(:, k))
-            output_layer_b(:, k) = 0.0_4
-            CALL POPREAL4ARRAY(input_layer, setup%neurons(1))
-            ac_hp_b(k) = ac_hp_b(k) + input_layer_b(1)
-            ac_ht_b(k) = ac_ht_b(k) + input_layer_b(2)
-            pn_b(k) = pn_b(k) + input_layer_b(3)
-            en_b(k) = en_b(k) + input_layer_b(4)
-          END IF
-        END IF
-      END DO
-    END DO
     ac_prcp_b = 0.0_4
-!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
-!$OMP&ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_b, ac_ci_b, ac_hi_b&
-!$OMP&, pn_b, en_b), PRIVATE(row, col, k), PRIVATE(branch, chunk_end, &
-!$OMP&chunk_start)
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&ac_prcp, ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_b, ac_ci_b&
+!$OMP&, ac_hi_b, pn_b, en_b), PRIVATE(row, col, k), PRIVATE(branch, &
+!$OMP&chunk_end, chunk_start)
+    CALL POPINTEGER4(time_step)
+    CALL POPREAL4ARRAY(ac_prcp, mesh%nac)
     en_b = 0.0_4
     pn_b = 0.0_4
     CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
@@ -16154,12 +16358,78 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
     END DO
 !$OMP END PARALLEL
     ac_mlt_b = ac_mlt_b + ac_prcp_b
-  END SUBROUTINE GR4_ODE_MLP_TIME_STEP_B
+  END SUBROUTINE GR4_ODE_TIME_STEP_B
 
-  SUBROUTINE GR4_ODE_MLP_TIME_STEP(setup, mesh, input_data, options, &
-&   returns, time_step, weight_1, bias_1, weight_2, bias_2, weight_3, &
-&   bias_3, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, ac_ht, &
-&   ac_qt)
+  SUBROUTINE GR4_ODE_TIME_STEP(setup, mesh, input_data, options, returns&
+&   , time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, &
+&   ac_ht, ac_qt)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: l
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp = ac_prcp + ac_mlt
+! Interception with OPENMP
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&ac_prcp, ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), &
+!$OMP&                                              SCHEDULE(static)
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
+&                          k), pn(k), en(k))
+          ELSE
+            pn(k) = 0._sp
+            en(k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer without OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          CALL GR_PRODUCTION_TRANSFER_ODE(pn(k), en(k), ac_cp(k), ac_ct(&
+&                                   k), ac_kexc(k), ac_hp(k), ac_ht(k), &
+&                                   ac_qt(k), l)
+! Transform from mm/dt to m3/s
+          ac_qt(k) = ac_qt(k)*1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col&
+&           )/setup%dt
+        END IF
+      END DO
+    END DO
+  END SUBROUTINE GR4_ODE_TIME_STEP
+
+!  Differentiation of gr4_ode_mlp_time_step in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
+!   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+  SUBROUTINE GR4_ODE_MLP_TIME_STEP_D(setup, mesh, input_data, options, &
+&   returns, time_step, weight_1, weight_1_d, bias_1, bias_1_d, weight_2&
+&   , weight_2_d, bias_2, bias_2_d, weight_3, weight_3_d, bias_3, &
+&   bias_3_d, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
+&   ac_ct_d, ac_kexc, ac_kexc_d, ac_hi, ac_hi_d, ac_hp, ac_hp_d, ac_ht, &
+&   ac_ht_d, ac_qt, ac_qt_d)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -16169,33 +16439,742 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP(setup, mesh, input_data, options, &
     INTEGER, INTENT(IN) :: time_step
     REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
 &   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1_d
     REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1_d
     REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
 &   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2_d
     REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2_d
     REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
 &   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3_d
     REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci_d, ac_cp_d, &
+&   ac_ct_d, ac_kexc_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_d, ac_hp_d, &
+&   ac_ht_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_d
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_d
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer_d
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d, pn_d, en_d
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: l
+    REAL(sp) :: temp
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp_d = ac_mlt_d
+    ac_prcp = ac_prcp + ac_mlt
+    en_d = 0.0_4
+    pn_d = 0.0_4
+! Interception with OPENMP
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
+!$OMP&ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_d, ac_ci_d, ac_hi_d&
+!$OMP&, pn_d, en_d), PRIVATE(row, col, k), SCHEDULE(static)
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_INTERCEPTION_D(ac_prcp(k), ac_prcp_d(k), ac_pet(k), &
+&                            ac_ci(k), ac_ci_d(k), ac_hi(k), ac_hi_d(k)&
+&                            , pn(k), pn_d(k), en(k), en_d(k))
+          ELSE
+            pn_d(k) = 0.0_4
+            pn(k) = 0._sp
+            en_d(k) = 0.0_4
+            en(k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+    output_layer_d = 0.0_4
+! Forward MLP without OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            input_layer_d(:) = (/ac_hp_d(k), ac_ht_d(k), pn_d(k), en_d(k&
+&             )/)
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP_D(weight_1, weight_1_d, bias_1, bias_1_d, &
+&                        weight_2, weight_2_d, bias_2, bias_2_d, &
+&                        weight_3, weight_3_d, bias_3, bias_3_d, &
+&                        input_layer, input_layer_d, output_layer(:, k)&
+&                        , output_layer_d(:, k))
+          ELSE
+            output_layer_d(:, k) = 0.0_4
+            output_layer(:, k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, output_layer, ac_cp, ac_ct, ac_kexc, ac_hp, ac_ht, ac_qt&
+!$OMP&, pn, en), SHARED(output_layer_d, ac_cp_d, ac_ct_d, ac_kexc_d, &
+!$OMP&ac_hp_d, ac_ht_d, ac_qt_d, pn_d, en_d), PRIVATE(row, col, k, &
+!$OMP&time_step_returns, l), PRIVATE(temp), SCHEDULE(static)
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          CALL GR_PRODUCTION_TRANSFER_ODE_MLP_D(output_layer(:, k), &
+&                                         output_layer_d(:, k), pn(k), &
+&                                         pn_d(k), en(k), en_d(k), ac_cp&
+&                                         (k), ac_cp_d(k), ac_ct(k), &
+&                                         ac_ct_d(k), ac_kexc(k), &
+&                                         ac_kexc_d(k), ac_hp(k), &
+&                                         ac_hp_d(k), ac_ht(k), ac_ht_d(&
+&                                         k), ac_qt(k), ac_qt_d(k), l)
+! Transform from mm/dt to m3/s
+          temp = 1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)
+          ac_qt_d(k) = temp*ac_qt_d(k)/setup%dt
+          ac_qt(k) = temp*(ac_qt(k)/setup%dt)
+        END IF
+      END DO
+    END DO
+  END SUBROUTINE GR4_ODE_MLP_TIME_STEP_D
+
+!  Differentiation of gr4_ode_mlp_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context OpenMP):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct bias_1
+!                bias_2 bias_3 ac_qt ac_hi ac_hp weight_1 weight_2
+!                weight_3 ac_ht ac_mlt
+  SUBROUTINE GR4_ODE_MLP_TIME_STEP_B(setup, mesh, input_data, options, &
+&   returns, time_step, weight_1, weight_1_b, bias_1, bias_1_b, weight_2&
+&   , weight_2_b, bias_2, bias_2_b, weight_3, weight_3_b, bias_3, &
+&   bias_3_b, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
+&   ac_ct_b, ac_kexc, ac_kexc_b, ac_hi, ac_hi_b, ac_hp, ac_hp_b, ac_ht, &
+&   ac_ht_b, ac_qt, ac_qt_b)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)) :: &
+&   weight_1_b
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(2)) :: bias_1_b
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)) :: &
+&   weight_2_b
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(3)) :: bias_2_b
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)) :: &
+&   weight_3_b
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
+    REAL(sp), DIMENSION(setup%neurons(4)) :: bias_3_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac) :: ac_mlt_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac) :: ac_ci_b, ac_cp_b, ac_ct_b, &
+&   ac_kexc_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_b, ac_hp_b, &
+&   ac_ht_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_b
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer_b
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer_b
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b, pn_b, en_b
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: l
+    INTEGER :: branch
+    INTEGER :: chunk_start
+    INTEGER :: chunk_end
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp = ac_prcp + ac_mlt
+! Interception with OPENMP
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
+!$OMP&ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), PRIVATE(&
+!$OMP&chunk_start, chunk_end)
+    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
+    DO col=chunk_start,chunk_end
+      DO row=1,mesh%nrow
+        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0) THEN
+          CALL PUSHCONTROL2B(0)
+        ELSE
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4(ac_hi(k))
+            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
+&                          k), pn(k), en(k))
+            CALL PUSHCONTROL2B(2)
+          ELSE
+            pn(k) = 0._sp
+            en(k) = 0._sp
+            CALL PUSHCONTROL2B(1)
+          END IF
+        END IF
+      END DO
+    END DO
+!$OMP END PARALLEL
+! Forward MLP without OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0) THEN
+          CALL PUSHCONTROL2B(0)
+        ELSE
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4ARRAY(input_layer, setup%neurons(1))
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
+&                      weight_3, bias_3, input_layer, output_layer(:, k)&
+&                     )
+            CALL PUSHCONTROL2B(2)
+          ELSE
+            output_layer(:, k) = 0._sp
+            CALL PUSHCONTROL2B(1)
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, output_layer, ac_cp, ac_ct, ac_kexc, ac_hp, ac_ht, ac_qt&
+!$OMP&, pn, en), PRIVATE(row, col, k, time_step_returns, l), PRIVATE(&
+!$OMP&chunk_start, chunk_end)
+    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
+    DO col=chunk_start,chunk_end
+      DO row=1,mesh%nrow
+        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0) THEN
+          CALL PUSHCONTROL1B(0)
+        ELSE
+          k = mesh%rowcol_to_ind_ac(row, col)
+          CALL PUSHREAL4(ac_qt(k))
+          CALL PUSHREAL4(ac_ht(k))
+          CALL PUSHREAL4(ac_hp(k))
+          CALL GR_PRODUCTION_TRANSFER_ODE_MLP(output_layer(:, k), pn(k)&
+&                                       , en(k), ac_cp(k), ac_ct(k), &
+&                                       ac_kexc(k), ac_hp(k), ac_ht(k), &
+&                                       ac_qt(k), l)
+! Transform from mm/dt to m3/s
+          CALL PUSHCONTROL1B(1)
+        END IF
+      END DO
+    END DO
+!$OMP END PARALLEL
+    output_layer_b = 0.0_4
+    en_b = 0.0_4
+    pn_b = 0.0_4
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, output_layer, ac_cp, ac_ct, ac_kexc, ac_hp, ac_ht, ac_qt&
+!$OMP&, pn, en), SHARED(output_layer_b, ac_cp_b, ac_ct_b, ac_kexc_b, &
+!$OMP&ac_hp_b, ac_ht_b, ac_qt_b, pn_b, en_b), PRIVATE(row, col, k, &
+!$OMP&time_step_returns, l), PRIVATE(branch, chunk_end, chunk_start)
+    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
+    DO col=chunk_end,chunk_start,-1
+      DO row=mesh%nrow,1,-1
+        CALL POPCONTROL1B(branch)
+        IF (branch .NE. 0) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          ac_qt_b(k) = mesh%dx(row, col)*1e-3_sp*mesh%dy(row, col)*&
+&           ac_qt_b(k)/setup%dt
+          CALL POPREAL4(ac_hp(k))
+          CALL POPREAL4(ac_ht(k))
+          CALL POPREAL4(ac_qt(k))
+          CALL GR_PRODUCTION_TRANSFER_ODE_MLP_B(output_layer(:, k), &
+&                                         output_layer_b(:, k), pn(k), &
+&                                         pn_b(k), en(k), en_b(k), ac_cp&
+&                                         (k), ac_cp_b(k), ac_ct(k), &
+&                                         ac_ct_b(k), ac_kexc(k), &
+&                                         ac_kexc_b(k), ac_hp(k), &
+&                                         ac_hp_b(k), ac_ht(k), ac_ht_b(&
+&                                         k), ac_qt(k), ac_qt_b(k), l)
+          ac_qt_b(k) = 0.0_4
+        END IF
+      END DO
+    END DO
+!$OMP END PARALLEL
+    DO col=mesh%ncol,1,-1
+      DO row=mesh%nrow,1,-1
+        CALL POPCONTROL2B(branch)
+        IF (branch .NE. 0) THEN
+          IF (branch .EQ. 1) THEN
+            k = mesh%rowcol_to_ind_ac(row, col)
+            output_layer_b(:, k) = 0.0_4
+          ELSE
+            k = mesh%rowcol_to_ind_ac(row, col)
+            CALL FORWARD_MLP_B(weight_1, weight_1_b, bias_1, bias_1_b, &
+&                        weight_2, weight_2_b, bias_2, bias_2_b, &
+&                        weight_3, weight_3_b, bias_3, bias_3_b, &
+&                        input_layer, input_layer_b, output_layer(:, k)&
+&                        , output_layer_b(:, k))
+            output_layer_b(:, k) = 0.0_4
+            CALL POPREAL4ARRAY(input_layer, setup%neurons(1))
+            ac_hp_b(k) = ac_hp_b(k) + input_layer_b(1)
+            ac_ht_b(k) = ac_ht_b(k) + input_layer_b(2)
+            pn_b(k) = pn_b(k) + input_layer_b(3)
+            en_b(k) = en_b(k) + input_layer_b(4)
+          END IF
+        END IF
+      END DO
+    END DO
+    ac_prcp_b = 0.0_4
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
+!$OMP&ac_pet, ac_ci, ac_hi, pn, en), SHARED(ac_prcp_b, ac_ci_b, ac_hi_b&
+!$OMP&, pn_b, en_b), PRIVATE(row, col, k), PRIVATE(branch, chunk_end, &
+!$OMP&chunk_start)
+    en_b = 0.0_4
+    pn_b = 0.0_4
+    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
+    DO col=chunk_end,chunk_start,-1
+      DO row=mesh%nrow,1,-1
+        CALL POPCONTROL2B(branch)
+        IF (branch .NE. 0) THEN
+          IF (branch .EQ. 1) THEN
+            k = mesh%rowcol_to_ind_ac(row, col)
+            en_b(k) = 0.0_4
+            pn_b(k) = 0.0_4
+          ELSE
+            k = mesh%rowcol_to_ind_ac(row, col)
+            CALL POPREAL4(ac_hi(k))
+            CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
+&                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
+&                            , pn(k), pn_b(k), en(k), en_b(k))
+            pn_b(k) = 0.0_4
+            en_b(k) = 0.0_4
+          END IF
+        END IF
+      END DO
+    END DO
+!$OMP END PARALLEL
+    ac_mlt_b = ac_mlt_b + ac_prcp_b
+  END SUBROUTINE GR4_ODE_MLP_TIME_STEP_B
+
+  SUBROUTINE GR4_ODE_MLP_TIME_STEP(setup, mesh, input_data, options, &
+&   returns, time_step, weight_1, bias_1, weight_2, bias_2, weight_3, &
+&   bias_3, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, ac_ht, &
+&   ac_qt)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(setup%neurons(2), setup%neurons(1)), INTENT(IN) &
+&   :: weight_1
+    REAL(sp), DIMENSION(setup%neurons(2)), INTENT(IN) :: bias_1
+    REAL(sp), DIMENSION(setup%neurons(3), setup%neurons(2)), INTENT(IN) &
+&   :: weight_2
+    REAL(sp), DIMENSION(setup%neurons(3)), INTENT(IN) :: bias_2
+    REAL(sp), DIMENSION(setup%neurons(4), setup%neurons(3)), INTENT(IN) &
+&   :: weight_3
+    REAL(sp), DIMENSION(setup%neurons(4)), INTENT(IN) :: bias_3
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
+    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
+&   output_layer
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: l
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp = ac_prcp + ac_mlt
+! Interception with OPENMP
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
+!$OMP&ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), SCHEDULE(&
+!$OMP&                                     static)
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
+&                          k), pn(k), en(k))
+          ELSE
+            pn(k) = 0._sp
+            en(k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Forward MLP without OPENMP
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
+            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
+&                      weight_3, bias_3, input_layer, output_layer(:, k)&
+&                     )
+          ELSE
+            output_layer(:, k) = 0._sp
+          END IF
+        END IF
+      END DO
+    END DO
+! Production and transfer with OPENMP
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, output_layer, ac_cp, ac_ct, ac_kexc, ac_hp, ac_ht, ac_qt&
+!$OMP&, pn, en), PRIVATE(row, col, k, time_step_returns, l), SCHEDULE(&
+!$OMP&                                       static)
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          CALL GR_PRODUCTION_TRANSFER_ODE_MLP(output_layer(:, k), pn(k)&
+&                                       , en(k), ac_cp(k), ac_ct(k), &
+&                                       ac_kexc(k), ac_hp(k), ac_ht(k), &
+&                                       ac_qt(k), l)
+! Transform from mm/dt to m3/s
+          ac_qt(k) = ac_qt(k)*1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col&
+&           )/setup%dt
+        END IF
+      END DO
+    END DO
+  END SUBROUTINE GR4_ODE_MLP_TIME_STEP
+
+!  Differentiation of gr5_time_step in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
+!   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
+  SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
+&   time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
+&   ac_ct_d, ac_kexc, ac_kexc_d, ac_aexc, ac_aexc_d, ac_hi, ac_hi_d, &
+&   ac_hp, ac_hp_d, ac_ht, ac_ht_d, ac_qt, ac_qt_d)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc, ac_aexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci_d, ac_cp_d, &
+&   ac_ct_d, ac_kexc_d, ac_aexc_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_d, ac_hp_d, &
+&   ac_ht_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_d
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: pn_d, en_d, pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d
+    INTRINSIC MAX
+    REAL(sp) :: temp
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp_d = ac_mlt_d
+    ac_prcp = ac_prcp + ac_mlt
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
+!$OMP&ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), SHARED(ac_prcp_d, ac_ci_d, &
+!$OMP&ac_cp_d, ac_ct_d, ac_kexc_d, ac_aexc_d, ac_hi_d, ac_hp_d, ac_ht_d&
+!$OMP&, ac_qt_d), PRIVATE(row, col, k, time_step_returns, pn, en, pr, &
+!$OMP&perc, l, prr, prd, qr, qd), PRIVATE(pn_d, en_d, pr_d, perc_d, l_d&
+!$OMP&, prr_d, prd_d, qr_d, qd_d), PRIVATE(temp), SCHEDULE(static)
+    DO col=1,mesh%ncol
+      DO row=1,mesh%nrow
+        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0)) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL GR_INTERCEPTION_D(ac_prcp(k), ac_prcp_d(k), ac_pet(k), &
+&                            ac_ci(k), ac_ci_d(k), ac_hi(k), ac_hi_d(k)&
+&                            , pn, pn_d, en, en_d)
+            CALL GR_PRODUCTION_D(0._sp, 0.0_4, 0._sp, 0.0_4, pn, pn_d, &
+&                          en, en_d, ac_cp(k), ac_cp_d(k), beta, ac_hp(k&
+&                          ), ac_hp_d(k), pr, pr_d, perc, perc_d)
+            CALL GR_THRESHOLD_EXCHANGE_D(ac_kexc(k), ac_kexc_d(k), &
+&                                  ac_aexc(k), ac_aexc_d(k), ac_ht(k), &
+&                                  ac_ht_d(k), l, l_d)
+          ELSE
+            pr = 0._sp
+            perc = 0._sp
+            l = 0._sp
+            l_d = 0.0_4
+            perc_d = 0.0_4
+            pr_d = 0.0_4
+          END IF
+          prr_d = 0.9_sp*(pr_d+perc_d) + l_d
+          prr = 0.9_sp*(pr+perc) + l
+          prd_d = 0.1_sp*(pr_d+perc_d)
+          prd = 0.1_sp*(pr+perc)
+          CALL GR_TRANSFER_D(5._sp, ac_prcp(k), prr, prr_d, ac_ct(k), &
+&                      ac_ct_d(k), ac_ht(k), ac_ht_d(k), qr, qr_d)
+          IF (0._sp .LT. prd + l) THEN
+            qd_d = prd_d + l_d
+            qd = prd + l
+          ELSE
+            qd = 0._sp
+            qd_d = 0.0_4
+          END IF
+          ac_qt_d(k) = qr_d + qd_d
+          ac_qt(k) = qr + qd
+! Transform from mm/dt to m3/s
+          temp = 1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)
+          ac_qt_d(k) = temp*ac_qt_d(k)/setup%dt
+          ac_qt(k) = temp*(ac_qt(k)/setup%dt)
+        END IF
+      END DO
+    END DO
+  END SUBROUTINE GR5_TIME_STEP_D
+
+!  Differentiation of gr5_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context OpenMP):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
+!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
+  SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
+&   time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
+&   ac_ct_b, ac_kexc, ac_kexc_b, ac_aexc, ac_aexc_b, ac_hi, ac_hi_b, &
+&   ac_hp, ac_hp_b, ac_ht, ac_ht_b, ac_qt, ac_qt_b)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
+    REAL(sp), DIMENSION(mesh%nac) :: ac_mlt_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
+&   ac_kexc, ac_aexc
+    REAL(sp), DIMENSION(mesh%nac) :: ac_ci_b, ac_cp_b, ac_ct_b, &
+&   ac_kexc_b, ac_aexc_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_b, ac_hp_b, &
+&   ac_ht_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
+    REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt_b
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b
+    INTEGER :: row, col, k, time_step_returns
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: pn_b, en_b, pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b
+    INTRINSIC MAX
+    REAL(sp) :: dummydiff_b
+    REAL(sp) :: dummydiff_b0
+    INTEGER :: branch
+    INTEGER :: chunk_start
+    INTEGER :: chunk_end
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'prcp', ac_prcp)
+    CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
+&                              , 'pet', ac_pet)
+    ac_prcp = ac_prcp + ac_mlt
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
+!$OMP&ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), PRIVATE(row, col, k, &
+!$OMP&time_step_returns, pn, en, pr, perc, l, prr, prd, qr, qd), PRIVATE&
+!$OMP&(chunk_start, chunk_end)
+    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
+    DO col=chunk_start,chunk_end
+      DO row=1,mesh%nrow
+        IF (mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
+&           local_active_cell(row, col) .EQ. 0) THEN
+          CALL PUSHCONTROL1B(0)
+        ELSE
+          k = mesh%rowcol_to_ind_ac(row, col)
+          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
+            CALL PUSHREAL4(en)
+            CALL PUSHREAL4(pn)
+            CALL PUSHREAL4(ac_hi(k))
+            CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
+&                          k), pn, en)
+            CALL PUSHREAL4(ac_hp(k))
+            CALL GR_PRODUCTION(0._sp, 0._sp, pn, en, ac_cp(k), beta, &
+&                        ac_hp(k), pr, perc)
+            CALL GR_THRESHOLD_EXCHANGE(ac_kexc(k), ac_aexc(k), ac_ht(k)&
+&                                , l)
+            CALL PUSHCONTROL1B(1)
+          ELSE
+            CALL PUSHCONTROL1B(0)
+            pr = 0._sp
+            perc = 0._sp
+            l = 0._sp
+          END IF
+          CALL PUSHREAL4(prr)
+          prr = 0.9_sp*(pr+perc) + l
+          prd = 0.1_sp*(pr+perc)
+          CALL PUSHREAL4(ac_ht(k))
+          CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
+&                    qr)
+          IF (0._sp .LT. prd + l) THEN
+            CALL PUSHCONTROL1B(0)
+          ELSE
+            CALL PUSHCONTROL1B(1)
+          END IF
+          CALL PUSHCONTROL1B(1)
+        END IF
+      END DO
+    END DO
+    CALL PUSHREAL4(pn)
+    CALL PUSHREAL4(prr)
+    CALL PUSHREAL4(en)
+!$OMP END PARALLEL
+    ac_prcp_b = 0.0_4
+!$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
+!$OMP&ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), SHARED(ac_prcp_b, ac_ci_b, &
+!$OMP&ac_cp_b, ac_ct_b, ac_kexc_b, ac_aexc_b, ac_hi_b, ac_hp_b, ac_ht_b&
+!$OMP&, ac_qt_b), PRIVATE(row, col, k, time_step_returns, pn, en, pr, &
+!$OMP&perc, l, prr, prd, qr, qd), PRIVATE(pn_b, en_b, pr_b, perc_b, l_b&
+!$OMP&, prr_b, prd_b, qr_b, qd_b), PRIVATE(branch, chunk_end, &
+!$OMP&chunk_start)
+    CALL POPREAL4(en)
+    CALL POPREAL4(prr)
+    CALL POPREAL4(pn)
+    pn_b = 0.0_4
+    en_b = 0.0_4
+    pr_b = 0.0_4
+    perc_b = 0.0_4
+    l_b = 0.0_4
+    prr_b = 0.0_4
+    prd_b = 0.0_4
+    qr_b = 0.0_4
+    qd_b = 0.0_4
+    CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
+    DO col=chunk_end,chunk_start,-1
+      DO row=mesh%nrow,1,-1
+        CALL POPCONTROL1B(branch)
+        IF (branch .NE. 0) THEN
+          k = mesh%rowcol_to_ind_ac(row, col)
+          ac_qt_b(k) = mesh%dx(row, col)*1e-3_sp*mesh%dy(row, col)*&
+&           ac_qt_b(k)/setup%dt
+          qr_b = ac_qt_b(k)
+          qd_b = ac_qt_b(k)
+          ac_qt_b(k) = 0.0_4
+          CALL POPCONTROL1B(branch)
+          IF (branch .EQ. 0) THEN
+            prd_b = qd_b
+            l_b = qd_b
+          ELSE
+            l_b = 0.0_4
+            prd_b = 0.0_4
+          END IF
+          CALL POPREAL4(ac_ht(k))
+          CALL GR_TRANSFER_B(5._sp, ac_prcp(k), prr, prr_b, ac_ct(k), &
+&                      ac_ct_b(k), ac_ht(k), ac_ht_b(k), qr, qr_b)
+          pr_b = 0.1_sp*prd_b + 0.9_sp*prr_b
+          perc_b = 0.1_sp*prd_b + 0.9_sp*prr_b
+          CALL POPREAL4(prr)
+          l_b = l_b + prr_b
+          CALL POPCONTROL1B(branch)
+          IF (branch .NE. 0) THEN
+            CALL GR_THRESHOLD_EXCHANGE_B(ac_kexc(k), ac_kexc_b(k), &
+&                                  ac_aexc(k), ac_aexc_b(k), ac_ht(k), &
+&                                  ac_ht_b(k), l, l_b)
+            CALL POPREAL4(ac_hp(k))
+            pn_b = 0.0_4
+            en_b = 0.0_4
+            CALL GR_PRODUCTION_B(0._sp, dummydiff_b, 0._sp, dummydiff_b0&
+&                          , pn, pn_b, en, en_b, ac_cp(k), ac_cp_b(k), &
+&                          beta, ac_hp(k), ac_hp_b(k), pr, pr_b, perc, &
+&                          perc_b)
+            CALL POPREAL4(ac_hi(k))
+            CALL POPREAL4(pn)
+            CALL POPREAL4(en)
+            CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
+&                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
+&                            , pn, pn_b, en, en_b)
+          END IF
+        END IF
+      END DO
+    END DO
+!$OMP END PARALLEL
+    ac_mlt_b = ac_mlt_b + ac_prcp_b
+  END SUBROUTINE GR5_TIME_STEP_B
+
+  SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
+&   time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_aexc, ac_hi, &
+&   ac_hp, ac_ht, ac_qt)
+    IMPLICIT NONE
+    TYPE(SETUPDT), INTENT(IN) :: setup
+    TYPE(MESHDT), INTENT(IN) :: mesh
+    TYPE(INPUT_DATADT), INTENT(IN) :: input_data
+    TYPE(OPTIONSDT), INTENT(IN) :: options
+    TYPE(RETURNSDT), INTENT(INOUT) :: returns
+    INTEGER, INTENT(IN) :: time_step
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
-&   ac_kexc
+&   ac_kexc, ac_aexc
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
-    REAL(sp), DIMENSION(setup%neurons(1)) :: input_layer
-    REAL(sp), DIMENSION(setup%neurons(setup%n_layers+1), mesh%nac) :: &
-&   output_layer
-    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet, pn, en
+    REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: l
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
+    INTRINSIC MAX
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'pet', ac_pet)
     ac_prcp = ac_prcp + ac_mlt
-! Interception with OPENMP
-!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(mesh, ac_prcp, &
-!$OMP&ac_pet, ac_ci, ac_hi, pn, en), PRIVATE(row, col, k), SCHEDULE(&
-!$OMP&                                     static)
+! Beta percolation parameter is time step dependent
+    beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
+!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
+!$OMP&ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), PRIVATE(row, col, k, &
+!$OMP&time_step_returns, pn, en, pr, perc, l, prr, prd, qr, qd), SCHEDULE&
+!$OMP&                                           (static)
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -16203,61 +17182,43 @@ SUBROUTINE GR4_ODE_MLP_TIME_STEP(setup, mesh, input_data, options, &
           k = mesh%rowcol_to_ind_ac(row, col)
           IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
             CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
-&                          k), pn(k), en(k))
+&                          k), pn, en)
+            CALL GR_PRODUCTION(0._sp, 0._sp, pn, en, ac_cp(k), beta, &
+&                        ac_hp(k), pr, perc)
+            CALL GR_THRESHOLD_EXCHANGE(ac_kexc(k), ac_aexc(k), ac_ht(k)&
+&                                , l)
           ELSE
-            pn(k) = 0._sp
-            en(k) = 0._sp
+            pr = 0._sp
+            perc = 0._sp
+            l = 0._sp
           END IF
-        END IF
-      END DO
-    END DO
-! Forward MLP without OPENMP
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
-            input_layer(:) = (/ac_hp(k), ac_ht(k), pn(k), en(k)/)
-            CALL FORWARD_MLP(weight_1, bias_1, weight_2, bias_2, &
-&                      weight_3, bias_3, input_layer, output_layer(:, k)&
-&                     )
+          prr = 0.9_sp*(pr+perc) + l
+          prd = 0.1_sp*(pr+perc)
+          CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
+&                    qr)
+          IF (0._sp .LT. prd + l) THEN
+            qd = prd + l
           ELSE
-            output_layer(:, k) = 0._sp
+            qd = 0._sp
           END IF
-        END IF
-      END DO
-    END DO
-! Production and transfer with OPENMP
-!$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, output_layer, ac_cp, ac_ct, ac_kexc, ac_hp, ac_ht, ac_qt&
-!$OMP&, pn, en), PRIVATE(row, col, k, time_step_returns, l), SCHEDULE(&
-!$OMP&                                       static)
-    DO col=1,mesh%ncol
-      DO row=1,mesh%nrow
-        IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
-&           local_active_cell(row, col) .EQ. 0)) THEN
-          k = mesh%rowcol_to_ind_ac(row, col)
-          CALL GR_PRODUCTION_TRANSFER_ODE_MLP(output_layer(:, k), pn(k)&
-&                                       , en(k), ac_cp(k), ac_ct(k), &
-&                                       ac_kexc(k), ac_hp(k), ac_ht(k), &
-&                                       ac_qt(k), l)
+          ac_qt(k) = qr + qd
 ! Transform from mm/dt to m3/s
           ac_qt(k) = ac_qt(k)*1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col&
 &           )/setup%dt
         END IF
       END DO
     END DO
-  END SUBROUTINE GR4_ODE_MLP_TIME_STEP
+  END SUBROUTINE GR5_TIME_STEP
 
-!  Differentiation of gr5_time_step in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
+!  Differentiation of gr5_ri_time_step in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
 !   variations   of useful results: ac_qt ac_hi ac_hp ac_ht
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
-  SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
-&   time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d, ac_ct, &
-&   ac_ct_d, ac_kexc, ac_kexc_d, ac_aexc, ac_aexc_d, ac_hi, ac_hi_d, &
-&   ac_hp, ac_hp_d, ac_ht, ac_ht_d, ac_qt, ac_qt_d)
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt ac_aexc
+  SUBROUTINE GR5_RI_TIME_STEP_D(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_d, ac_ci, ac_ci_d, ac_cp, ac_cp_d&
+&   , ac_ct, ac_ct_d, ac_alpha1, ac_alpha1_d, ac_alpha2, ac_alpha2_d, &
+&   ac_kexc, ac_kexc_d, ac_aexc, ac_aexc_d, ac_hi, ac_hi_d, ac_hp, &
+&   ac_hp_d, ac_ht, ac_ht_d, ac_qt, ac_qt_d)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -16271,6 +17232,9 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
 &   ac_kexc, ac_aexc
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci_d, ac_cp_d, &
 &   ac_ct_d, ac_kexc_d, ac_aexc_d
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1_d, &
+&   ac_alpha2_d
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_d, ac_hp_d, &
 &   ac_ht_d
@@ -16279,8 +17243,10 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_d
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
-    REAL(sp) :: pn_d, en_d, pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    REAL(sp) :: pn_d, en_d, pr_d, perc_d, l_d, prr_d, prd_d, qr_d, qd_d&
+&   , split_d
+    INTRINSIC TANH
     INTRINSIC MAX
     REAL(sp) :: temp
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
@@ -16291,13 +17257,15 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
     ac_prcp = ac_prcp + ac_mlt
 ! Beta percolation parameter is time step dependent
     beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
+    pn_d = 0.0_4
 !$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
-!$OMP&ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), SHARED(ac_prcp_d, ac_ci_d, &
-!$OMP&ac_cp_d, ac_ct_d, ac_kexc_d, ac_aexc_d, ac_hi_d, ac_hp_d, ac_ht_d&
-!$OMP&, ac_qt_d), PRIVATE(row, col, k, time_step_returns, pn, en, pr, &
-!$OMP&perc, l, prr, prd, qr, qd), PRIVATE(pn_d, en_d, pr_d, perc_d, l_d&
-!$OMP&, prr_d, prd_d, qr_d, qd_d), PRIVATE(temp), SCHEDULE(static)
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_alpha1, ac_alpha2&
+!$OMP&, ac_ct, ac_kexc, ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), SHARED(&
+!$OMP&ac_prcp_d, ac_ci_d, ac_cp_d, ac_alpha1_d, ac_alpha2_d, ac_ct_d, &
+!$OMP&ac_kexc_d, ac_aexc_d, ac_hi_d, ac_hp_d, ac_ht_d, ac_qt_d), PRIVATE&
+!$OMP&(row, col, k, time_step_returns, pn, en, pr, perc, l, prr, prd, qr&
+!$OMP&, qd, split), PRIVATE(pn_d, en_d, pr_d, perc_d, l_d, prr_d, prd_d&
+!$OMP&, qr_d, qd_d, split_d), PRIVATE(temp), SCHEDULE(static)
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -16307,9 +17275,10 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
             CALL GR_INTERCEPTION_D(ac_prcp(k), ac_prcp_d(k), ac_pet(k), &
 &                            ac_ci(k), ac_ci_d(k), ac_hi(k), ac_hi_d(k)&
 &                            , pn, pn_d, en, en_d)
-            CALL GR_PRODUCTION_D(0._sp, 0.0_4, 0._sp, 0.0_4, pn, pn_d, &
-&                          en, en_d, ac_cp(k), ac_cp_d(k), beta, ac_hp(k&
-&                          ), ac_hp_d(k), pr, pr_d, perc, perc_d)
+            CALL GR_RI_PRODUCTION_D(pn, pn_d, en, en_d, ac_cp(k), &
+&                             ac_cp_d(k), beta, ac_alpha1(k), &
+&                             ac_alpha1_d(k), ac_hp(k), ac_hp_d(k), pr, &
+&                             pr_d, perc, perc_d, setup%dt)
             CALL GR_THRESHOLD_EXCHANGE_D(ac_kexc(k), ac_kexc_d(k), &
 &                                  ac_aexc(k), ac_aexc_d(k), ac_ht(k), &
 &                                  ac_ht_d(k), l, l_d)
@@ -16321,10 +17290,13 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
             perc_d = 0.0_4
             pr_d = 0.0_4
           END IF
-          prr_d = 0.9_sp*(pr_d+perc_d) + l_d
-          prr = 0.9_sp*(pr+perc) + l
-          prd_d = 0.1_sp*(pr_d+perc_d)
-          prd = 0.1_sp*(pr+perc)
+          split_d = 0.9_sp*2*TANH(ac_alpha2(k)*pn)*(1.0-TANH(ac_alpha2(k&
+&           )*pn)**2)*(pn*ac_alpha2_d(k)+ac_alpha2(k)*pn_d)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
+          prr_d = (1._sp-split)*(pr_d+perc_d) - (pr+perc)*split_d + l_d
+          prr = (1._sp-split)*(pr+perc) + l
+          prd_d = (pr+perc)*split_d + split*(pr_d+perc_d)
+          prd = split*(pr+perc)
           CALL GR_TRANSFER_D(5._sp, ac_prcp(k), prr, prr_d, ac_ct(k), &
 &                      ac_ct_d(k), ac_ht(k), ac_ht_d(k), qr, qr_d)
           IF (0._sp .LT. prd + l) THEN
@@ -16343,17 +17315,18 @@ SUBROUTINE GR5_TIME_STEP_D(setup, mesh, input_data, options, returns, &
         END IF
       END DO
     END DO
-  END SUBROUTINE GR5_TIME_STEP_D
+  END SUBROUTINE GR5_RI_TIME_STEP_D
 
-!  Differentiation of gr5_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context OpenMP):
-!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
-!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_qt
-!                ac_hi ac_hp ac_ht ac_mlt ac_aexc
-  SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
-&   time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b, ac_ct, &
-&   ac_ct_b, ac_kexc, ac_kexc_b, ac_aexc, ac_aexc_b, ac_hi, ac_hi_b, &
-&   ac_hp, ac_hp_b, ac_ht, ac_ht_b, ac_qt, ac_qt_b)
+!  Differentiation of gr5_ri_time_step in reverse (adjoint) mode (with options fixinterface noISIZE context OpenMP):
+!   gradient     of useful results: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt ac_aexc
+!   with respect to varying inputs: ac_kexc ac_ci ac_cp ac_ct ac_alpha1
+!                ac_alpha2 ac_qt ac_hi ac_hp ac_ht ac_mlt ac_aexc
+  SUBROUTINE GR5_RI_TIME_STEP_B(setup, mesh, input_data, options, &
+&   returns, time_step, ac_mlt, ac_mlt_b, ac_ci, ac_ci_b, ac_cp, ac_cp_b&
+&   , ac_ct, ac_ct_b, ac_alpha1, ac_alpha1_b, ac_alpha2, ac_alpha2_b, &
+&   ac_kexc, ac_kexc_b, ac_aexc, ac_aexc_b, ac_hi, ac_hi_b, ac_hp, &
+&   ac_hp_b, ac_ht, ac_ht_b, ac_qt, ac_qt_b)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -16367,6 +17340,8 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
 &   ac_kexc, ac_aexc
     REAL(sp), DIMENSION(mesh%nac) :: ac_ci_b, ac_cp_b, ac_ct_b, &
 &   ac_kexc_b, ac_aexc_b
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
+    REAL(sp), DIMENSION(mesh%nac) :: ac_alpha1_b, ac_alpha2_b
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi_b, ac_hp_b, &
 &   ac_ht_b
@@ -16375,11 +17350,12 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp_b
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
-    REAL(sp) :: pn_b, en_b, pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    REAL(sp) :: pn_b, en_b, pr_b, perc_b, l_b, prr_b, prd_b, qr_b, qd_b&
+&   , split_b
+    INTRINSIC TANH
     INTRINSIC MAX
-    REAL(sp) :: dummydiff_b
-    REAL(sp) :: dummydiff_b0
+    REAL(sp) :: temp_b
     INTEGER :: branch
     INTEGER :: chunk_start
     INTEGER :: chunk_end
@@ -16391,10 +17367,10 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
 ! Beta percolation parameter is time step dependent
     beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
 !$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
-!$OMP&ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), PRIVATE(row, col, k, &
-!$OMP&time_step_returns, pn, en, pr, perc, l, prr, prd, qr, qd), PRIVATE&
-!$OMP&(chunk_start, chunk_end)
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_alpha1, ac_alpha2&
+!$OMP&, ac_ct, ac_kexc, ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), PRIVATE(&
+!$OMP&row, col, k, time_step_returns, pn, en, pr, perc, l, prr, prd, qr&
+!$OMP&, qd, split), PRIVATE(chunk_start, chunk_end)
     CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
     DO col=chunk_start,chunk_end
       DO row=1,mesh%nrow
@@ -16409,21 +17385,27 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
             CALL PUSHREAL4(ac_hi(k))
             CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
 &                          k), pn, en)
+            CALL PUSHREAL4(perc)
+            CALL PUSHREAL4(pr)
             CALL PUSHREAL4(ac_hp(k))
-            CALL GR_PRODUCTION(0._sp, 0._sp, pn, en, ac_cp(k), beta, &
-&                        ac_hp(k), pr, perc)
+            CALL GR_RI_PRODUCTION(pn, en, ac_cp(k), beta, ac_alpha1(k), &
+&                           ac_hp(k), pr, perc, setup%dt)
             CALL GR_THRESHOLD_EXCHANGE(ac_kexc(k), ac_aexc(k), ac_ht(k)&
 &                                , l)
             CALL PUSHCONTROL1B(1)
           ELSE
-            CALL PUSHCONTROL1B(0)
+            CALL PUSHREAL4(pr)
             pr = 0._sp
+            CALL PUSHREAL4(perc)
             perc = 0._sp
             l = 0._sp
+            CALL PUSHCONTROL1B(0)
           END IF
+          CALL PUSHREAL4(split)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
           CALL PUSHREAL4(prr)
-          prr = 0.9_sp*(pr+perc) + l
-          prd = 0.1_sp*(pr+perc)
+          prr = (1._sp-split)*(pr+perc) + l
+          prd = split*(pr+perc)
           CALL PUSHREAL4(ac_ht(k))
           CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
 &                    qr)
@@ -16436,22 +17418,30 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
         END IF
       END DO
     END DO
+    CALL PUSHREAL4(pr)
+    CALL PUSHREAL4(perc)
     CALL PUSHREAL4(pn)
     CALL PUSHREAL4(prr)
+    CALL PUSHREAL4(split)
     CALL PUSHREAL4(en)
 !$OMP END PARALLEL
     ac_prcp_b = 0.0_4
+    pn_b = 0.0_4
 !$OMP PARALLEL NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
-!$OMP&ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), SHARED(ac_prcp_b, ac_ci_b, &
-!$OMP&ac_cp_b, ac_ct_b, ac_kexc_b, ac_aexc_b, ac_hi_b, ac_hp_b, ac_ht_b&
-!$OMP&, ac_qt_b), PRIVATE(row, col, k, time_step_returns, pn, en, pr, &
-!$OMP&perc, l, prr, prd, qr, qd), PRIVATE(pn_b, en_b, pr_b, perc_b, l_b&
-!$OMP&, prr_b, prd_b, qr_b, qd_b), PRIVATE(branch, chunk_end, &
-!$OMP&chunk_start)
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_alpha1, ac_alpha2&
+!$OMP&, ac_ct, ac_kexc, ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), SHARED(&
+!$OMP&ac_prcp_b, ac_ci_b, ac_cp_b, ac_alpha1_b, ac_alpha2_b, ac_ct_b, &
+!$OMP&ac_kexc_b, ac_aexc_b, ac_hi_b, ac_hp_b, ac_ht_b, ac_qt_b), PRIVATE&
+!$OMP&(row, col, k, time_step_returns, pn, en, pr, perc, l, prr, prd, qr&
+!$OMP&, qd, split), PRIVATE(pn_b, en_b, pr_b, perc_b, l_b, prr_b, prd_b&
+!$OMP&, qr_b, qd_b, split_b), PRIVATE(branch, chunk_end, chunk_start), &
+!$OMP&PRIVATE(temp_b)
     CALL POPREAL4(en)
+    CALL POPREAL4(split)
     CALL POPREAL4(prr)
     CALL POPREAL4(pn)
+    CALL POPREAL4(perc)
+    CALL POPREAL4(pr)
     pn_b = 0.0_4
     en_b = 0.0_4
     pr_b = 0.0_4
@@ -16461,6 +17451,8 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
     prd_b = 0.0_4
     qr_b = 0.0_4
     qd_b = 0.0_4
+    split_b = 0.0_4
+    pn_b = 0.0_4
     CALL GETSTATICSCHEDULE(1, mesh%ncol, 1, chunk_start, chunk_end)
     DO col=chunk_end,chunk_start,-1
       DO row=mesh%nrow,1,-1
@@ -16483,39 +17475,50 @@ SUBROUTINE GR5_TIME_STEP_B(setup, mesh, input_data, options, returns, &
           CALL POPREAL4(ac_ht(k))
           CALL GR_TRANSFER_B(5._sp, ac_prcp(k), prr, prr_b, ac_ct(k), &
 &                      ac_ct_b(k), ac_ht(k), ac_ht_b(k), qr, qr_b)
-          pr_b = 0.1_sp*prd_b + 0.9_sp*prr_b
-          perc_b = 0.1_sp*prd_b + 0.9_sp*prr_b
+          split_b = (pr+perc)*prd_b - (pr+perc)*prr_b
+          pr_b = split*prd_b + (1._sp-split)*prr_b
+          perc_b = split*prd_b + (1._sp-split)*prr_b
           CALL POPREAL4(prr)
           l_b = l_b + prr_b
+          CALL POPREAL4(split)
+          temp_b = (1.0-TANH(ac_alpha2(k)*pn)**2)*2*TANH(ac_alpha2(k)*pn&
+&           )*0.9_sp*split_b
+!$OMP     ATOMIC update
+          ac_alpha2_b(k) = ac_alpha2_b(k) + pn*temp_b
+          pn_b = pn_b + ac_alpha2(k)*temp_b
           CALL POPCONTROL1B(branch)
-          IF (branch .NE. 0) THEN
+          IF (branch .EQ. 0) THEN
+            CALL POPREAL4(perc)
+            CALL POPREAL4(pr)
+          ELSE
             CALL GR_THRESHOLD_EXCHANGE_B(ac_kexc(k), ac_kexc_b(k), &
 &                                  ac_aexc(k), ac_aexc_b(k), ac_ht(k), &
 &                                  ac_ht_b(k), l, l_b)
             CALL POPREAL4(ac_hp(k))
-            pn_b = 0.0_4
-            en_b = 0.0_4
-            CALL GR_PRODUCTION_B(0._sp, dummydiff_b, 0._sp, dummydiff_b0&
-&                          , pn, pn_b, en, en_b, ac_cp(k), ac_cp_b(k), &
-&                          beta, ac_hp(k), ac_hp_b(k), pr, pr_b, perc, &
-&                          perc_b)
+            CALL POPREAL4(pr)
+            CALL POPREAL4(perc)
+            CALL GR_RI_PRODUCTION_B(pn, pn_b, en, en_b, ac_cp(k), &
+&                             ac_cp_b(k), beta, ac_alpha1(k), &
+&                             ac_alpha1_b(k), ac_hp(k), ac_hp_b(k), pr, &
+&                             pr_b, perc, perc_b, setup%dt)
             CALL POPREAL4(ac_hi(k))
             CALL POPREAL4(pn)
             CALL POPREAL4(en)
             CALL GR_INTERCEPTION_B(ac_prcp(k), ac_prcp_b(k), ac_pet(k), &
 &                            ac_ci(k), ac_ci_b(k), ac_hi(k), ac_hi_b(k)&
 &                            , pn, pn_b, en, en_b)
+            pn_b = 0.0_4
           END IF
         END IF
       END DO
     END DO
 !$OMP END PARALLEL
     ac_mlt_b = ac_mlt_b + ac_prcp_b
-  END SUBROUTINE GR5_TIME_STEP_B
+  END SUBROUTINE GR5_RI_TIME_STEP_B
 
-  SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
-&   time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_aexc, ac_hi, &
-&   ac_hp, ac_ht, ac_qt)
+  SUBROUTINE GR5_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&   , time_step, ac_mlt, ac_ci, ac_cp, ac_ct, ac_alpha1, ac_alpha2, &
+&   ac_kexc, ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt)
     IMPLICIT NONE
     TYPE(SETUPDT), INTENT(IN) :: setup
     TYPE(MESHDT), INTENT(IN) :: mesh
@@ -16526,11 +17529,13 @@ SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_mlt
     REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_ci, ac_cp, ac_ct, &
 &   ac_kexc, ac_aexc
+    REAL(sp), DIMENSION(mesh%nac), INTENT(IN) :: ac_alpha1, ac_alpha2
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_hi, ac_hp, ac_ht
     REAL(sp), DIMENSION(mesh%nac), INTENT(INOUT) :: ac_qt
     REAL(sp), DIMENSION(mesh%nac) :: ac_prcp, ac_pet
     INTEGER :: row, col, k, time_step_returns
-    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd
+    REAL(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+    INTRINSIC TANH
     INTRINSIC MAX
     CALL GET_AC_ATMOS_DATA_TIME_STEP(setup, mesh, input_data, time_step&
 &                              , 'prcp', ac_prcp)
@@ -16540,10 +17545,10 @@ SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
 ! Beta percolation parameter is time step dependent
     beta = 9._sp/4._sp*(86400._sp/setup%dt)**0.25_sp
 !$OMP PARALLEL DO NUM_THREADS(options%comm%ncpu), SHARED(setup, mesh, &
-!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_ct, ac_kexc, &
-!$OMP&ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), PRIVATE(row, col, k, &
-!$OMP&time_step_returns, pn, en, pr, perc, l, prr, prd, qr, qd), SCHEDULE&
-!$OMP&                                           (static)
+!$OMP&returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, ac_alpha1, ac_alpha2&
+!$OMP&, ac_ct, ac_kexc, ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt), PRIVATE(&
+!$OMP&row, col, k, time_step_returns, pn, en, pr, perc, l, prr, prd, qr&
+!$OMP&, qd, split), SCHEDULE(static)
     DO col=1,mesh%ncol
       DO row=1,mesh%nrow
         IF (.NOT.(mesh%active_cell(row, col) .EQ. 0 .OR. mesh%&
@@ -16552,8 +17557,8 @@ SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
           IF (ac_prcp(k) .GE. 0._sp .AND. ac_pet(k) .GE. 0._sp) THEN
             CALL GR_INTERCEPTION(ac_prcp(k), ac_pet(k), ac_ci(k), ac_hi(&
 &                          k), pn, en)
-            CALL GR_PRODUCTION(0._sp, 0._sp, pn, en, ac_cp(k), beta, &
-&                        ac_hp(k), pr, perc)
+            CALL GR_RI_PRODUCTION(pn, en, ac_cp(k), beta, ac_alpha1(k), &
+&                           ac_hp(k), pr, perc, setup%dt)
             CALL GR_THRESHOLD_EXCHANGE(ac_kexc(k), ac_aexc(k), ac_ht(k)&
 &                                , l)
           ELSE
@@ -16561,8 +17566,9 @@ SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
             perc = 0._sp
             l = 0._sp
           END IF
-          prr = 0.9_sp*(pr+perc) + l
-          prd = 0.1_sp*(pr+perc)
+          split = 0.9_sp*TANH(ac_alpha2(k)*pn)**2 + 0.1_sp
+          prr = (1._sp-split)*(pr+perc) + l
+          prd = split*(pr+perc)
           CALL GR_TRANSFER(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), &
 &                    qr)
           IF (0._sp .LT. prd + l) THEN
@@ -16577,7 +17583,7 @@ SUBROUTINE GR5_TIME_STEP(setup, mesh, input_data, options, returns, &
         END IF
       END DO
     END DO
-  END SUBROUTINE GR5_TIME_STEP
+  END SUBROUTINE GR5_RI_TIME_STEP
 
 !  Differentiation of gr6_time_step in forward (tangent) mode (with options fixinterface noISIZE context OpenMP):
 !   variations   of useful results: ac_qt ac_he ac_hi ac_hp ac_ht
@@ -21362,6 +22368,59 @@ SUBROUTINE SIMULATION_CHECKPOINT_D(setup, mesh, input_data, parameters&
         checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
         rr_parameters_inc = rr_parameters_inc + 4
         rr_states_inc = rr_states_inc + 3
+      CASE ('gr4_ri') 
+! 'gr4_ri' module
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+1)
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+2)
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+3)
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % hi
+! % hp
+! % ht
+        CALL GR4_RI_TIME_STEP_D(setup, mesh, input_data, options, &
+&                         returns, t, checkpoint_variable%ac_mlt, &
+&                         checkpoint_variable_d%ac_mlt, &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+1), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+1), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+2), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+2), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+3), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+3), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+4), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+4), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+5), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+5), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+6), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+6), h1, &
+&                         h1_d, h2, h2_d, h3, h3_d, checkpoint_variable%&
+&                         ac_qtz(:, setup%nqz), checkpoint_variable_d%&
+&                         ac_qtz(:, setup%nqz))
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+1) = h1_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+2) = h2_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+3) = h3_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        rr_parameters_inc = rr_parameters_inc + 6
+        rr_states_inc = rr_states_inc + 3
       CASE ('gr4_mlp') 
 ! 'gr4_mlp' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -21571,7 +22630,63 @@ SUBROUTINE SIMULATION_CHECKPOINT_D(setup, mesh, input_data, parameters&
         checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
         checkpoint_variable_d%ac_rr_states(:, rr_states_inc+3) = h3_d
         checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
-        rr_parameters_inc = rr_parameters_inc + 5
+        rr_parameters_inc = rr_parameters_inc + 5
+        rr_states_inc = rr_states_inc + 3
+      CASE ('gr5_ri') 
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+1)
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+2)
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3_d = checkpoint_variable_d%ac_rr_states(:, rr_states_inc+3)
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % aexc
+! % hi
+! % hp
+! % ht
+        CALL GR5_RI_TIME_STEP_D(setup, mesh, input_data, options, &
+&                         returns, t, checkpoint_variable%ac_mlt, &
+&                         checkpoint_variable_d%ac_mlt, &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+1), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+1), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+2), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+2), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+3), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+3), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+4), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+4), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+5), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+5), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+6), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+6), &
+&                         checkpoint_variable%ac_rr_parameters(:, &
+&                         rr_parameters_inc+7), checkpoint_variable_d%&
+&                         ac_rr_parameters(:, rr_parameters_inc+7), h1, &
+&                         h1_d, h2, h2_d, h3, h3_d, checkpoint_variable%&
+&                         ac_qtz(:, setup%nqz), checkpoint_variable_d%&
+&                         ac_qtz(:, setup%nqz))
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+1) = h1_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+2) = h2_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable_d%ac_rr_states(:, rr_states_inc+3) = h3_d
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        rr_parameters_inc = rr_parameters_inc + 7
         rr_states_inc = rr_states_inc + 3
       CASE ('gr6') 
 ! 'gr6' module
@@ -21997,6 +23112,49 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 3
         CALL PUSHCONTROL4B(1)
+      CASE ('gr4_ri') 
+! 'gr4_ri' module
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % hi
+! % hp
+! % ht
+        CALL PUSHREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                     SIZE(checkpoint_variable%ac_qtz, 1))
+        CALL PUSHREAL4ARRAY(h3, mesh%nac)
+        CALL PUSHREAL4ARRAY(h2, mesh%nac)
+        CALL PUSHREAL4ARRAY(h1, mesh%nac)
+        CALL GR4_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&                       , t, checkpoint_variable%ac_mlt, &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+1), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+2), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+3), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+4), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+5), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+6), h1, h2&
+&                       , h3, checkpoint_variable%ac_qtz(:, setup%nqz))
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        CALL PUSHINTEGER4(rr_parameters_inc)
+        rr_parameters_inc = rr_parameters_inc + 6
+        CALL PUSHINTEGER4(rr_states_inc)
+        rr_states_inc = rr_states_inc + 3
+        CALL PUSHCONTROL4B(2)
       CASE ('gr4_mlp') 
 ! 'gr4_mlp' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -22039,7 +23197,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 4
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 3
-        CALL PUSHCONTROL4B(2)
+        CALL PUSHCONTROL4B(3)
       CASE ('gr4_ode') 
 ! 'gr4_ode' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -22078,7 +23236,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 4
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 3
-        CALL PUSHCONTROL4B(3)
+        CALL PUSHCONTROL4B(4)
       CASE ('gr4_ode_mlp') 
 ! 'gr4_ode_mlp' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -22122,7 +23280,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 4
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 3
-        CALL PUSHCONTROL4B(4)
+        CALL PUSHCONTROL4B(5)
       CASE ('gr5') 
 ! 'gr5' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -22162,7 +23320,52 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 5
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 3
-        CALL PUSHCONTROL4B(5)
+        CALL PUSHCONTROL4B(6)
+      CASE ('gr5_ri') 
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % aexc
+! % hi
+! % hp
+! % ht
+        CALL PUSHREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                     SIZE(checkpoint_variable%ac_qtz, 1))
+        CALL PUSHREAL4ARRAY(h3, mesh%nac)
+        CALL PUSHREAL4ARRAY(h2, mesh%nac)
+        CALL PUSHREAL4ARRAY(h1, mesh%nac)
+        CALL GR5_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&                       , t, checkpoint_variable%ac_mlt, &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+1), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+2), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+3), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+4), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+5), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+6), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+7), h1, h2, h3, &
+&                       checkpoint_variable%ac_qtz(:, setup%nqz))
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        CALL PUSHINTEGER4(rr_parameters_inc)
+        rr_parameters_inc = rr_parameters_inc + 7
+        CALL PUSHINTEGER4(rr_states_inc)
+        rr_states_inc = rr_states_inc + 3
+        CALL PUSHCONTROL4B(7)
       CASE ('gr6') 
 ! 'gr6' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -22210,7 +23413,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 6
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 4
-        CALL PUSHCONTROL4B(6)
+        CALL PUSHCONTROL4B(8)
       CASE ('grc') 
 ! 'grc' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -22255,7 +23458,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 5
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 4
-        CALL PUSHCONTROL4B(7)
+        CALL PUSHCONTROL4B(9)
       CASE ('grd') 
 ! 'grd' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -22283,7 +23486,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 2
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 2
-        CALL PUSHCONTROL4B(8)
+        CALL PUSHCONTROL4B(10)
       CASE ('loieau') 
 ! 'loieau' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -22314,7 +23517,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 3
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 2
-        CALL PUSHCONTROL4B(9)
+        CALL PUSHCONTROL4B(11)
       CASE ('vic3l') 
 ! 'vic3l' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -22370,7 +23573,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
         rr_parameters_inc = rr_parameters_inc + 9
         CALL PUSHINTEGER4(rr_states_inc)
         rr_states_inc = rr_states_inc + 4
-        CALL PUSHCONTROL4B(10)
+        CALL PUSHCONTROL4B(12)
       CASE DEFAULT
         CALL PUSHCONTROL4B(0)
       END SELECT
@@ -22467,60 +23670,125 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
 &                     ac_qz)
       END IF
       CALL POPCONTROL4B(branch)
-      IF (branch .LT. 5) THEN
-        IF (branch .LT. 2) THEN
+      IF (branch .LT. 6) THEN
+        IF (branch .LT. 3) THEN
           IF (branch .NE. 0) THEN
-            CALL POPINTEGER4(rr_states_inc)
-            CALL POPINTEGER4(rr_parameters_inc)
-            h3_b = 0.0_4
-            h3_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3&
-&             )
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
-&             0.0_4
-            h2_b = 0.0_4
-            h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2&
-&             )
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
-&             0.0_4
-            h1_b = 0.0_4
-            h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1&
-&             )
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
-&             0.0_4
-            CALL POPREAL4ARRAY(h1, mesh%nac)
-            CALL POPREAL4ARRAY(h2, mesh%nac)
-            CALL POPREAL4ARRAY(h3, mesh%nac)
-            CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz)&
-&                        , SIZE(checkpoint_variable%ac_qtz, 1))
-            CALL GR4_TIME_STEP_B(setup, mesh, input_data, options, &
-&                          returns, t, checkpoint_variable%ac_mlt, &
-&                          checkpoint_variable_b%ac_mlt, &
-&                          checkpoint_variable%ac_rr_parameters(:, &
-&                          rr_parameters_inc+1), checkpoint_variable_b%&
-&                          ac_rr_parameters(:, rr_parameters_inc+1), &
-&                          checkpoint_variable%ac_rr_parameters(:, &
-&                          rr_parameters_inc+2), checkpoint_variable_b%&
-&                          ac_rr_parameters(:, rr_parameters_inc+2), &
-&                          checkpoint_variable%ac_rr_parameters(:, &
-&                          rr_parameters_inc+3), checkpoint_variable_b%&
-&                          ac_rr_parameters(:, rr_parameters_inc+3), &
-&                          checkpoint_variable%ac_rr_parameters(:, &
-&                          rr_parameters_inc+4), checkpoint_variable_b%&
-&                          ac_rr_parameters(:, rr_parameters_inc+4), h1&
-&                          , h1_b, h2, h2_b, h3, h3_b, &
-&                          checkpoint_variable%ac_qtz(:, setup%nqz), &
-&                          checkpoint_variable_b%ac_qtz(:, setup%nqz))
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
-&             checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) + &
-&             h3_b
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
-&             checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) + &
-&             h2_b
-            checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
-&             checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
-&             h1_b
+            IF (branch .EQ. 1) THEN
+              CALL POPINTEGER4(rr_states_inc)
+              CALL POPINTEGER4(rr_parameters_inc)
+              h3_b = 0.0_4
+              h3_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +3)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
+&               0.0_4
+              h2_b = 0.0_4
+              h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +2)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&               0.0_4
+              h1_b = 0.0_4
+              h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +1)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&               0.0_4
+              CALL POPREAL4ARRAY(h1, mesh%nac)
+              CALL POPREAL4ARRAY(h2, mesh%nac)
+              CALL POPREAL4ARRAY(h3, mesh%nac)
+              CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz&
+&                          ), SIZE(checkpoint_variable%ac_qtz, 1))
+              CALL GR4_TIME_STEP_B(setup, mesh, input_data, options, &
+&                            returns, t, checkpoint_variable%ac_mlt, &
+&                            checkpoint_variable_b%ac_mlt, &
+&                            checkpoint_variable%ac_rr_parameters(:, &
+&                            rr_parameters_inc+1), checkpoint_variable_b&
+&                            %ac_rr_parameters(:, rr_parameters_inc+1), &
+&                            checkpoint_variable%ac_rr_parameters(:, &
+&                            rr_parameters_inc+2), checkpoint_variable_b&
+&                            %ac_rr_parameters(:, rr_parameters_inc+2), &
+&                            checkpoint_variable%ac_rr_parameters(:, &
+&                            rr_parameters_inc+3), checkpoint_variable_b&
+&                            %ac_rr_parameters(:, rr_parameters_inc+3), &
+&                            checkpoint_variable%ac_rr_parameters(:, &
+&                            rr_parameters_inc+4), checkpoint_variable_b&
+&                            %ac_rr_parameters(:, rr_parameters_inc+4), &
+&                            h1, h1_b, h2, h2_b, h3, h3_b, &
+&                            checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                            checkpoint_variable_b%ac_qtz(:, setup%nqz))
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) +&
+&               h3_b
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) +&
+&               h2_b
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) +&
+&               h1_b
+            ELSE
+              CALL POPINTEGER4(rr_states_inc)
+              CALL POPINTEGER4(rr_parameters_inc)
+              h3_b = 0.0_4
+              h3_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +3)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
+&               0.0_4
+              h2_b = 0.0_4
+              h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +2)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&               0.0_4
+              h1_b = 0.0_4
+              h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc&
+&               +1)
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&               0.0_4
+              CALL POPREAL4ARRAY(h1, mesh%nac)
+              CALL POPREAL4ARRAY(h2, mesh%nac)
+              CALL POPREAL4ARRAY(h3, mesh%nac)
+              CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz&
+&                          ), SIZE(checkpoint_variable%ac_qtz, 1))
+              CALL GR4_RI_TIME_STEP_B(setup, mesh, input_data, options, &
+&                               returns, t, checkpoint_variable%ac_mlt, &
+&                               checkpoint_variable_b%ac_mlt, &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+1), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+1), &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+2), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+2), &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+3), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+3), &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+4), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+4), &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+5), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+5), &
+&                               checkpoint_variable%ac_rr_parameters(:, &
+&                               rr_parameters_inc+6), &
+&                               checkpoint_variable_b%ac_rr_parameters(:&
+&                               , rr_parameters_inc+6), h1, h1_b, h2, &
+&                               h2_b, h3, h3_b, checkpoint_variable%&
+&                               ac_qtz(:, setup%nqz), &
+&                               checkpoint_variable_b%ac_qtz(:, setup%&
+&                               nqz))
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) +&
+&               h3_b
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) +&
+&               h2_b
+              checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&               checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) +&
+&               h1_b
+            END IF
           END IF
-        ELSE IF (branch .EQ. 2) THEN
+        ELSE IF (branch .EQ. 3) THEN
           CALL POPINTEGER4(rr_states_inc)
           CALL POPINTEGER4(rr_parameters_inc)
           h3_b = 0.0_4
@@ -22576,7 +23844,7 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
 &           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
 &           h1_b
-        ELSE IF (branch .EQ. 3) THEN
+        ELSE IF (branch .EQ. 4) THEN
           CALL POPINTEGER4(rr_states_inc)
           CALL POPINTEGER4(rr_parameters_inc)
           h3_b = 0.0_4
@@ -22682,8 +23950,8 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
 &           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
 &           h1_b
         END IF
-      ELSE IF (branch .LT. 8) THEN
-        IF (branch .EQ. 5) THEN
+      ELSE IF (branch .LT. 9) THEN
+        IF (branch .EQ. 6) THEN
           CALL POPINTEGER4(rr_states_inc)
           CALL POPINTEGER4(rr_parameters_inc)
           h3_b = 0.0_4
@@ -22730,7 +23998,60 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
 &           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
 &           h1_b
-        ELSE IF (branch .EQ. 6) THEN
+        ELSE IF (branch .EQ. 7) THEN
+          CALL POPINTEGER4(rr_states_inc)
+          CALL POPINTEGER4(rr_parameters_inc)
+          h3_b = 0.0_4
+          h3_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3)
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = 0.0_4
+          h2_b = 0.0_4
+          h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2)
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = 0.0_4
+          h1_b = 0.0_4
+          h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1)
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = 0.0_4
+          CALL POPREAL4ARRAY(h1, mesh%nac)
+          CALL POPREAL4ARRAY(h2, mesh%nac)
+          CALL POPREAL4ARRAY(h3, mesh%nac)
+          CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                      SIZE(checkpoint_variable%ac_qtz, 1))
+          CALL GR5_RI_TIME_STEP_B(setup, mesh, input_data, options, &
+&                           returns, t, checkpoint_variable%ac_mlt, &
+&                           checkpoint_variable_b%ac_mlt, &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+1), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+1), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+2), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+2), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+3), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+3), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+4), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+4), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+5), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+5), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+6), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+6), &
+&                           checkpoint_variable%ac_rr_parameters(:, &
+&                           rr_parameters_inc+7), checkpoint_variable_b%&
+&                           ac_rr_parameters(:, rr_parameters_inc+7), h1&
+&                           , h1_b, h2, h2_b, h3, h3_b, &
+&                           checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                           checkpoint_variable_b%ac_qtz(:, setup%nqz))
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) = &
+&           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+3) + &
+&           h3_b
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) + &
+&           h2_b
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
+&           h1_b
+        ELSE
           CALL POPINTEGER4(rr_states_inc)
           CALL POPINTEGER4(rr_parameters_inc)
           h4_b = 0.0_4
@@ -22787,7 +24108,9 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
 &           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
 &           h1_b
-        ELSE
+        END IF
+      ELSE IF (branch .LT. 11) THEN
+        IF (branch .EQ. 9) THEN
           CALL POPINTEGER4(rr_states_inc)
           CALL POPINTEGER4(rr_parameters_inc)
           h4_b = 0.0_4
@@ -22840,37 +24163,40 @@ SUBROUTINE SIMULATION_CHECKPOINT_B(setup, mesh, input_data, parameters&
 &           h2_b
           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
 &           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
+&           h1_b
+        ELSE
+          CALL POPINTEGER4(rr_states_inc)
+          CALL POPINTEGER4(rr_parameters_inc)
+          h2_b = 0.0_4
+          h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2)
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = 0.0_4
+          h1_b = 0.0_4
+          h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1)
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = 0.0_4
+          CALL POPREAL4ARRAY(h1, mesh%nac)
+          CALL POPREAL4ARRAY(h2, mesh%nac)
+          CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz), &
+&                      SIZE(checkpoint_variable%ac_qtz, 1))
+          CALL GRD_TIME_STEP_B(setup, mesh, input_data, options, returns&
+&                        , t, checkpoint_variable%ac_mlt, &
+&                        checkpoint_variable_b%ac_mlt, &
+&                        checkpoint_variable%ac_rr_parameters(:, &
+&                        rr_parameters_inc+1), checkpoint_variable_b%&
+&                        ac_rr_parameters(:, rr_parameters_inc+1), &
+&                        checkpoint_variable%ac_rr_parameters(:, &
+&                        rr_parameters_inc+2), checkpoint_variable_b%&
+&                        ac_rr_parameters(:, rr_parameters_inc+2), h1, &
+&                        h1_b, h2, h2_b, checkpoint_variable%ac_qtz(:, &
+&                        setup%nqz), checkpoint_variable_b%ac_qtz(:, &
+&                        setup%nqz))
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
+&           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) + &
+&           h2_b
+          checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
+&           checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + &
 &           h1_b
         END IF
-      ELSE IF (branch .EQ. 8) THEN
-        CALL POPINTEGER4(rr_states_inc)
-        CALL POPINTEGER4(rr_parameters_inc)
-        h2_b = 0.0_4
-        h2_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2)
-        checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = 0.0_4
-        h1_b = 0.0_4
-        h1_b = checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1)
-        checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = 0.0_4
-        CALL POPREAL4ARRAY(h1, mesh%nac)
-        CALL POPREAL4ARRAY(h2, mesh%nac)
-        CALL POPREAL4ARRAY(checkpoint_variable%ac_qtz(:, setup%nqz), &
-&                    SIZE(checkpoint_variable%ac_qtz, 1))
-        CALL GRD_TIME_STEP_B(setup, mesh, input_data, options, returns, &
-&                      t, checkpoint_variable%ac_mlt, &
-&                      checkpoint_variable_b%ac_mlt, checkpoint_variable&
-&                      %ac_rr_parameters(:, rr_parameters_inc+1), &
-&                      checkpoint_variable_b%ac_rr_parameters(:, &
-&                      rr_parameters_inc+1), checkpoint_variable%&
-&                      ac_rr_parameters(:, rr_parameters_inc+2), &
-&                      checkpoint_variable_b%ac_rr_parameters(:, &
-&                      rr_parameters_inc+2), h1, h1_b, h2, h2_b, &
-&                      checkpoint_variable%ac_qtz(:, setup%nqz), &
-&                      checkpoint_variable_b%ac_qtz(:, setup%nqz))
-        checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) = &
-&         checkpoint_variable_b%ac_rr_states(:, rr_states_inc+2) + h2_b
-        checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) = &
-&         checkpoint_variable_b%ac_rr_states(:, rr_states_inc+1) + h1_b
-      ELSE IF (branch .EQ. 9) THEN
+      ELSE IF (branch .EQ. 11) THEN
         CALL POPINTEGER4(rr_states_inc)
         CALL POPINTEGER4(rr_parameters_inc)
         h2_b = 0.0_4
@@ -23072,6 +24398,41 @@ SUBROUTINE SIMULATION_CHECKPOINT(setup, mesh, input_data, parameters, &
         checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
         rr_parameters_inc = rr_parameters_inc + 4
         rr_states_inc = rr_states_inc + 3
+      CASE ('gr4_ri') 
+! 'gr4_ri' module
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % hi
+! % hp
+! % ht
+        CALL GR4_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&                       , t, checkpoint_variable%ac_mlt, &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+1), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+2), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+3), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+4), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+5), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+6), h1, h2&
+&                       , h3, checkpoint_variable%ac_qtz(:, setup%nqz))
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        rr_parameters_inc = rr_parameters_inc + 6
+        rr_states_inc = rr_states_inc + 3
       CASE ('gr4_mlp') 
 ! 'gr4_mlp' module
 ! % To avoid potential aliasing tapenade warning (DF02)
@@ -23206,6 +24567,43 @@ SUBROUTINE SIMULATION_CHECKPOINT(setup, mesh, input_data, parameters, &
         checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
         rr_parameters_inc = rr_parameters_inc + 5
         rr_states_inc = rr_states_inc + 3
+      CASE ('gr5_ri') 
+! % To avoid potential aliasing tapenade warning (DF02)
+! % hi
+        h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc+1)
+! % hp
+        h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc+2)
+! % ht
+        h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc+3)
+! % ci
+! % cp
+! % ct
+! % alpha1
+! % alpha2
+! % kexc
+! % aexc
+! % hi
+! % hp
+! % ht
+        CALL GR5_RI_TIME_STEP(setup, mesh, input_data, options, returns&
+&                       , t, checkpoint_variable%ac_mlt, &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+1), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+2), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+3), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+4), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+5), checkpoint_variable%&
+&                       ac_rr_parameters(:, rr_parameters_inc+6), &
+&                       checkpoint_variable%ac_rr_parameters(:, &
+&                       rr_parameters_inc+7), h1, h2, h3, &
+&                       checkpoint_variable%ac_qtz(:, setup%nqz))
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+1) = h1
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+2) = h2
+        checkpoint_variable%ac_rr_states(:, rr_states_inc+3) = h3
+        rr_parameters_inc = rr_parameters_inc + 7
+        rr_states_inc = rr_states_inc + 3
       CASE ('gr6') 
 ! 'gr6' module
 ! % To avoid potential aliasing tapenade warning (DF02)
diff --git a/smash/fcore/forward/md_simulation.f90 b/smash/fcore/forward/md_simulation.f90
index b5f668f5..30d34905 100644
--- a/smash/fcore/forward/md_simulation.f90
+++ b/smash/fcore/forward/md_simulation.f90
@@ -199,6 +199,40 @@ subroutine simulation_checkpoint(setup, mesh, input_data, parameters, output, op
                 rr_parameters_inc = rr_parameters_inc + 4
                 rr_states_inc = rr_states_inc + 3
 
+                ! 'gr4_ri' module
+            case ("gr4_ri")
+
+                ! % To avoid potential aliasing tapenade warning (DF02)
+                h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc + 1) ! % hi
+                h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc + 2) ! % hp
+                h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc + 3) ! % ht
+
+                call gr4_ri_time_step( &
+                    setup, &
+                    mesh, &
+                    input_data, &
+                    options, &
+                    returns, &
+                    t, &
+                    checkpoint_variable%ac_mlt, &
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 1), & ! % ci
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 2), & ! % cp
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 3), & ! % ct
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 4), & ! % alpha1
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 5), & ! % alpha2
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 6), & ! % kexc
+                    h1, & ! % hi
+                    h2, & ! % hp
+                    h3, & ! % ht
+                    checkpoint_variable%ac_qtz(:, setup%nqz))
+
+                checkpoint_variable%ac_rr_states(:, rr_states_inc + 1) = h1
+                checkpoint_variable%ac_rr_states(:, rr_states_inc + 2) = h2
+                checkpoint_variable%ac_rr_states(:, rr_states_inc + 3) = h3
+
+                rr_parameters_inc = rr_parameters_inc + 6
+                rr_states_inc = rr_states_inc + 3
+
                 ! 'gr4_mlp' module
             case ("gr4_mlp")
 
@@ -340,6 +374,40 @@ subroutine simulation_checkpoint(setup, mesh, input_data, parameters, output, op
                 rr_parameters_inc = rr_parameters_inc + 5
                 rr_states_inc = rr_states_inc + 3
 
+            case ("gr5_ri")
+
+                ! % To avoid potential aliasing tapenade warning (DF02)
+                h1 = checkpoint_variable%ac_rr_states(:, rr_states_inc + 1) ! % hi
+                h2 = checkpoint_variable%ac_rr_states(:, rr_states_inc + 2) ! % hp
+                h3 = checkpoint_variable%ac_rr_states(:, rr_states_inc + 3) ! % ht
+
+                call gr5_ri_time_step( &
+                    setup, &
+                    mesh, &
+                    input_data, &
+                    options, &
+                    returns, &
+                    t, &
+                    checkpoint_variable%ac_mlt, &
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 1), & ! % ci
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 2), & ! % cp
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 3), & ! % ct
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 4), & ! % alpha1
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 5), & ! % alpha2
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 6), & ! % kexc
+                    checkpoint_variable%ac_rr_parameters(:, rr_parameters_inc + 7), & ! % aexc
+                    h1, & ! % hi
+                    h2, & ! % hp
+                    h3, & ! % ht
+                    checkpoint_variable%ac_qtz(:, setup%nqz))
+
+                checkpoint_variable%ac_rr_states(:, rr_states_inc + 1) = h1
+                checkpoint_variable%ac_rr_states(:, rr_states_inc + 2) = h2
+                checkpoint_variable%ac_rr_states(:, rr_states_inc + 3) = h3
+
+                rr_parameters_inc = rr_parameters_inc + 7
+                rr_states_inc = rr_states_inc + 3
+
                 ! 'gr6' module
             case ("gr6")
 
diff --git a/smash/fcore/operator/md_gr_operator.f90 b/smash/fcore/operator/md_gr_operator.f90
index 9060f018..8cb6472e 100644
--- a/smash/fcore/operator/md_gr_operator.f90
+++ b/smash/fcore/operator/md_gr_operator.f90
@@ -91,6 +91,44 @@ subroutine gr_production(fq_ps, fq_es, pn, en, cp, beta, hp, pr, perc)
 
     end subroutine gr_production
 
+    subroutine gr_ri_production(pn, en, cp, beta, alpha1, hp, pr, perc, dt)
+
+        implicit none
+
+        real(sp), intent(in) :: pn, en, cp, beta, alpha1
+        real(sp), intent(in) :: dt
+        real(sp), intent(inout) :: hp
+        real(sp), intent(out) :: pr, perc
+
+        real(sp) :: inv_cp, ps, es, hp_imd
+        real(sp) :: lambda, gam, inv_lambda
+
+        inv_cp = 1._sp/cp
+        pr = 0._sp
+        gam = 1._sp - exp(-pn*alpha1)
+        lambda = sqrt(1._sp - gam)
+        inv_lambda = 1._sp/lambda
+
+        ps = cp*inv_lambda*tanh(lambda*pn*inv_cp)*(1._sp - (lambda*hp)**2) &
+        & /(1._sp + lambda*hp*tanh(lambda*pn*inv_cp)) - gam*dt
+
+        es = (hp*cp)*(2._sp - hp)*tanh(en*inv_cp)/ &
+        & (1._sp + (1._sp - hp)*tanh(en*inv_cp))
+
+        hp_imd = hp + (ps - es)*inv_cp
+
+        if (pn .gt. 0) then
+
+            pr = pn - (hp_imd - hp)*cp
+
+        end if
+
+        perc = (hp_imd*cp)*(1._sp - (1._sp + (hp_imd/beta)**4)**(-0.25_sp))
+
+        hp = hp_imd - perc*inv_cp
+
+    end subroutine gr_ri_production
+
     subroutine gr_exchange(fq_l, kexc, ht, l)
 
         implicit none
@@ -381,6 +419,104 @@ subroutine gr4_time_step(setup, mesh, input_data, options, returns, time_step, a
 #endif
     end subroutine gr4_time_step
 
+    subroutine gr4_ri_time_step(setup, mesh, input_data, options, returns, time_step, ac_mlt, ac_ci, ac_cp, ac_ct, &
+    & ac_alpha1, ac_alpha2, ac_kexc, ac_hi, ac_hp, ac_ht, ac_qt)
+
+        implicit none
+
+        type(SetupDT), intent(in) :: setup
+        type(MeshDT), intent(in) :: mesh
+        type(Input_DataDT), intent(in) :: input_data
+        type(OptionsDT), intent(in) :: options
+        type(ReturnsDT), intent(inout) :: returns
+        integer, intent(in) :: time_step
+        real(sp), dimension(mesh%nac), intent(in) :: ac_mlt
+        real(sp), dimension(mesh%nac), intent(in) :: ac_ci, ac_cp, ac_ct, ac_kexc
+        real(sp), dimension(mesh%nac), intent(in) :: ac_alpha1, ac_alpha2
+        real(sp), dimension(mesh%nac), intent(inout) :: ac_hi, ac_hp, ac_ht
+        real(sp), dimension(mesh%nac), intent(inout) :: ac_qt
+
+        real(sp), dimension(mesh%nac) :: ac_prcp, ac_pet
+        integer :: row, col, k, time_step_returns
+        real(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+
+        call get_ac_atmos_data_time_step(setup, mesh, input_data, time_step, "prcp", ac_prcp)
+        call get_ac_atmos_data_time_step(setup, mesh, input_data, time_step, "pet", ac_pet)
+
+        ac_prcp = ac_prcp + ac_mlt
+
+        ! Beta percolation parameter is time step dependent
+        beta = (9._sp/4._sp)*(86400._sp/setup%dt)**0.25_sp
+#ifdef _OPENMP
+        !$OMP parallel do schedule(static) num_threads(options%comm%ncpu) &
+        !$OMP& shared(setup, mesh, returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, &
+        !$OMP&  ac_ct, ac_kexc, ac_hi, ac_hp, ac_ht, ac_qt) &
+        !$OMP& private(row, col, k, time_step_returns, pn, en, pr, perc, l, prr, prd, qr, qd, split)
+#endif
+        do col = 1, mesh%ncol
+            do row = 1, mesh%nrow
+
+                if (mesh%active_cell(row, col) .eq. 0 .or. mesh%local_active_cell(row, col) .eq. 0) cycle
+
+                k = mesh%rowcol_to_ind_ac(row, col)
+
+                if (ac_prcp(k) .ge. 0._sp .and. ac_pet(k) .ge. 0._sp) then
+
+                    call gr_interception(ac_prcp(k), ac_pet(k), ac_ci(k), &
+                    & ac_hi(k), pn, en)
+
+                    call gr_ri_production(pn, en, ac_cp(k), beta, ac_alpha1(k), ac_hp(k), pr, perc, setup%dt)
+
+                    call gr_exchange(0._sp, ac_kexc(k), ac_ht(k), l)
+
+                else
+
+                    pr = 0._sp
+                    perc = 0._sp
+                    l = 0._sp
+
+                end if
+                split = 0.9_sp*tanh(ac_alpha2(k)*pn)**2 + 0.1_sp
+
+                prr = (1._sp - split)*(pr + perc) + l
+                prd = split*(pr + perc)
+
+                call gr_transfer(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), qr)
+
+                qd = max(0._sp, prd + l)
+
+                ac_qt(k) = qr + qd
+
+                ! Transform from mm/dt to m3/s
+                ac_qt(k) = ac_qt(k)*1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)/setup%dt
+
+                !$AD start-exclude
+                !internal fluxes
+                if (returns%internal_fluxes_flag) then
+                    if (allocated(returns%mask_time_step)) then
+                        if (returns%mask_time_step(time_step)) then
+                            time_step_returns = returns%time_step_to_returns_time_step(time_step)
+                            ! the fluxes of the snow module are the first ones inside internal fluxes
+                            ! due to the building of the modules so n_snow_fluxes
+                            ! moves the index of the array
+                            returns%internal_fluxes( &
+                                row, &
+                                col, &
+                                time_step_returns, &
+                                setup%n_snow_fluxes + 1:setup%n_snow_fluxes + setup%n_hydro_fluxes &
+                                ) = (/pn, en, pr, perc, l, prr, prd, qr, qd, ac_qt(k)/)
+                        end if
+                    end if
+                end if
+                !$AD end-exclude
+
+            end do
+        end do
+#ifdef _OPENMP
+        !$OMP end parallel do
+#endif
+    end subroutine gr4_ri_time_step
+
     subroutine gr4_mlp_time_step(setup, mesh, input_data, options, returns, time_step, weight_1, bias_1, &
     & weight_2, bias_2, weight_3, bias_3, ac_mlt, ac_ci, ac_cp, ac_ct, ac_kexc, ac_hi, ac_hp, ac_ht, ac_qt)
 
@@ -858,6 +994,105 @@ subroutine gr5_time_step(setup, mesh, input_data, options, returns, time_step, a
 #endif
     end subroutine gr5_time_step
 
+    subroutine gr5_ri_time_step(setup, mesh, input_data, options, returns, time_step, ac_mlt, ac_ci, ac_cp, ac_ct, &
+    & ac_alpha1, ac_alpha2, ac_kexc, ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt)
+
+        implicit none
+
+        type(SetupDT), intent(in) :: setup
+        type(MeshDT), intent(in) :: mesh
+        type(Input_DataDT), intent(in) :: input_data
+        type(OptionsDT), intent(in) :: options
+        type(ReturnsDT), intent(inout) :: returns
+        integer, intent(in) :: time_step
+        real(sp), dimension(mesh%nac), intent(in) :: ac_mlt
+        real(sp), dimension(mesh%nac), intent(in) :: ac_ci, ac_cp, ac_ct, ac_kexc, ac_aexc
+        real(sp), dimension(mesh%nac), intent(in) :: ac_alpha1, ac_alpha2
+        real(sp), dimension(mesh%nac), intent(inout) :: ac_hi, ac_hp, ac_ht
+        real(sp), dimension(mesh%nac), intent(inout) :: ac_qt
+
+        real(sp), dimension(mesh%nac) :: ac_prcp, ac_pet
+        integer :: row, col, k, time_step_returns
+        real(sp) :: beta, pn, en, pr, perc, l, prr, prd, qr, qd, split
+
+        call get_ac_atmos_data_time_step(setup, mesh, input_data, time_step, "prcp", ac_prcp)
+        call get_ac_atmos_data_time_step(setup, mesh, input_data, time_step, "pet", ac_pet)
+
+        ac_prcp = ac_prcp + ac_mlt
+
+        ! Beta percolation parameter is time step dependent
+        beta = (9._sp/4._sp)*(86400._sp/setup%dt)**0.25_sp
+
+#ifdef _OPENMP
+        !$OMP parallel do schedule(static) num_threads(options%comm%ncpu) &
+        !$OMP& shared(setup, mesh, returns, ac_prcp, ac_pet, ac_ci, ac_cp, beta, &
+        !$OMP& ac_alpha1, ac_alpha2, ac_ct, ac_kexc, ac_aexc, ac_hi, ac_hp, ac_ht, ac_qt) &
+        !$OMP& private(row, col, k, time_step_returns, pn, en, pr, perc, l, prr, prd, qr, qd, split)
+#endif
+        do col = 1, mesh%ncol
+            do row = 1, mesh%nrow
+
+                if (mesh%active_cell(row, col) .eq. 0 .or. mesh%local_active_cell(row, col) .eq. 0) cycle
+
+                k = mesh%rowcol_to_ind_ac(row, col)
+
+                if (ac_prcp(k) .ge. 0._sp .and. ac_pet(k) .ge. 0._sp) then
+
+                    call gr_interception(ac_prcp(k), ac_pet(k), ac_ci(k), &
+                    & ac_hi(k), pn, en)
+
+                    call gr_ri_production(pn, en, ac_cp(k), beta, ac_alpha1(k), ac_hp(k), pr, perc, setup%dt)
+
+                    call gr_threshold_exchange(ac_kexc(k), ac_aexc(k), ac_ht(k), l)
+
+                else
+
+                    pr = 0._sp
+                    perc = 0._sp
+                    l = 0._sp
+
+                end if
+
+                split = 0.9_sp*tanh(ac_alpha2(k)*pn)**2 + 0.1_sp
+
+                prr = (1._sp - split)*(pr + perc) + l
+                prd = split*(pr + perc)
+
+                call gr_transfer(5._sp, ac_prcp(k), prr, ac_ct(k), ac_ht(k), qr)
+
+                qd = max(0._sp, prd + l)
+
+                ac_qt(k) = qr + qd
+
+                ! Transform from mm/dt to m3/s
+                ac_qt(k) = ac_qt(k)*1e-3_sp*mesh%dx(row, col)*mesh%dy(row, col)/setup%dt
+
+                !$AD start-exclude
+                !internal fluxes
+                if (returns%internal_fluxes_flag) then
+                    if (allocated(returns%mask_time_step)) then
+                        if (returns%mask_time_step(time_step)) then
+                            time_step_returns = returns%time_step_to_returns_time_step(time_step)
+                            ! the fluxes of the snow module are the first ones inside internal fluxes
+                            ! due to the building of the modules so n_snow_fluxes
+                            ! moves the index of the array
+                            returns%internal_fluxes( &
+                                row, &
+                                col, &
+                                time_step_returns, &
+                                setup%n_snow_fluxes + 1:setup%n_snow_fluxes + setup%n_hydro_fluxes &
+                                ) = (/pn, en, pr, perc, l, prr, prd, qr, qd, ac_qt(k)/)
+                        end if
+                    end if
+                end if
+                !$AD end-exclude
+            end do
+        end do
+#ifdef _OPENMP
+        !$OMP end parallel do
+#endif
+    end subroutine gr5_ri_time_step
+
     subroutine gr6_time_step(setup, mesh, input_data, options, returns, time_step, ac_mlt, ac_ci, ac_cp, ac_ct, &
     & ac_be, ac_kexc, ac_aexc, ac_hi, ac_hp, ac_ht, ac_he, ac_qt)
 
diff --git a/smash/tests/baseline.hdf5 b/smash/tests/baseline.hdf5
index 5af61754..60aaeb3e 100644
Binary files a/smash/tests/baseline.hdf5 and b/smash/tests/baseline.hdf5 differ
diff --git a/smash/tests/diff_baseline.csv b/smash/tests/diff_baseline.csv
index 866535ef..eb6a43d0 100644
--- a/smash/tests/diff_baseline.csv
+++ b/smash/tests/diff_baseline.csv
@@ -1,12 +1,8 @@
-commit 10b2e6d630dd602f840176b838e2f7885e45e761
-Author: Ngo Nghi Truyen Huynh <129378719+nghi-truyen@users.noreply.github.com>
-Date:   Thu Sep 12 22:35:22 2024 +0200
+commit 9026756332322c49a2c4ee3c5f7359b030906938
+Author: Apolline Elbaz <aelbaz@localhost.localdomain>
+Date:   Mon Sep 16 12:38:24 2024 +0200
 
-    MAINT: change default network for optimization with ann mapping
-    
-    - Change default net ann optimize
-    - Enhance net checking
-    - Correct/update lez regionalization for ann mapping
+    change constant for fast tests
 
 TEST NAME                                                      |STATUS
 bbox_mesh.active_cell                                          |NON MODIFIED
@@ -47,10 +43,10 @@ custom_optimize.zero-gr4-lr.custom_set_12.iter_control         |NON MODIFIED
 custom_optimize.zero-gr4-lr.custom_set_12.iter_cost            |NON MODIFIED
 custom_optimize.zero-gr4-lr.custom_set_12.iter_projg           |NON MODIFIED
 custom_optimize.zero-gr4-lr.custom_set_12.sim_q                |NON MODIFIED
-custom_optimize.zero-gr4-lr.custom_set_13.iter_control         |MODIFIED
-custom_optimize.zero-gr4-lr.custom_set_13.iter_cost            |MODIFIED
-custom_optimize.zero-gr4-lr.custom_set_13.iter_projg           |MODIFIED
-custom_optimize.zero-gr4-lr.custom_set_13.sim_q                |MODIFIED
+custom_optimize.zero-gr4-lr.custom_set_13.iter_control         |NON MODIFIED
+custom_optimize.zero-gr4-lr.custom_set_13.iter_cost            |NON MODIFIED
+custom_optimize.zero-gr4-lr.custom_set_13.iter_projg           |NON MODIFIED
+custom_optimize.zero-gr4-lr.custom_set_13.sim_q                |NON MODIFIED
 custom_optimize.zero-gr4-lr.custom_set_2.iter_control          |NON MODIFIED
 custom_optimize.zero-gr4-lr.custom_set_2.iter_cost             |NON MODIFIED
 custom_optimize.zero-gr4-lr.custom_set_2.sim_q                 |NON MODIFIED
@@ -173,6 +169,28 @@ forward_run.zero-gr4_ode_mlp-lr.rr_states.hlr                  |NON MODIFIED
 forward_run.zero-gr4_ode_mlp-lr.rr_states.hp                   |NON MODIFIED
 forward_run.zero-gr4_ode_mlp-lr.rr_states.ht                   |NON MODIFIED
 forward_run.zero-gr4_ode_mlp-lr.sim_q                          |NON MODIFIED
+forward_run.zero-gr4_ri-kw.cost                                |ADDED
+forward_run.zero-gr4_ri-kw.jobs                                |ADDED
+forward_run.zero-gr4_ri-kw.q_domain                            |ADDED
+forward_run.zero-gr4_ri-kw.rr_states.hi                        |ADDED
+forward_run.zero-gr4_ri-kw.rr_states.hp                        |ADDED
+forward_run.zero-gr4_ri-kw.rr_states.ht                        |ADDED
+forward_run.zero-gr4_ri-kw.sim_q                               |ADDED
+forward_run.zero-gr4_ri-lag0.cost                              |ADDED
+forward_run.zero-gr4_ri-lag0.jobs                              |ADDED
+forward_run.zero-gr4_ri-lag0.q_domain                          |ADDED
+forward_run.zero-gr4_ri-lag0.rr_states.hi                      |ADDED
+forward_run.zero-gr4_ri-lag0.rr_states.hp                      |ADDED
+forward_run.zero-gr4_ri-lag0.rr_states.ht                      |ADDED
+forward_run.zero-gr4_ri-lag0.sim_q                             |ADDED
+forward_run.zero-gr4_ri-lr.cost                                |ADDED
+forward_run.zero-gr4_ri-lr.jobs                                |ADDED
+forward_run.zero-gr4_ri-lr.q_domain                            |ADDED
+forward_run.zero-gr4_ri-lr.rr_states.hi                        |ADDED
+forward_run.zero-gr4_ri-lr.rr_states.hlr                       |ADDED
+forward_run.zero-gr4_ri-lr.rr_states.hp                        |ADDED
+forward_run.zero-gr4_ri-lr.rr_states.ht                        |ADDED
+forward_run.zero-gr4_ri-lr.sim_q                               |ADDED
 forward_run.zero-gr5-kw.cost                                   |NON MODIFIED
 forward_run.zero-gr5-kw.jobs                                   |NON MODIFIED
 forward_run.zero-gr5-kw.q_domain                               |NON MODIFIED
@@ -195,6 +213,28 @@ forward_run.zero-gr5-lr.rr_states.hlr                          |NON MODIFIED
 forward_run.zero-gr5-lr.rr_states.hp                           |NON MODIFIED
 forward_run.zero-gr5-lr.rr_states.ht                           |NON MODIFIED
 forward_run.zero-gr5-lr.sim_q                                  |NON MODIFIED
+forward_run.zero-gr5_ri-kw.cost                                |ADDED
+forward_run.zero-gr5_ri-kw.jobs                                |ADDED
+forward_run.zero-gr5_ri-kw.q_domain                            |ADDED
+forward_run.zero-gr5_ri-kw.rr_states.hi                        |ADDED
+forward_run.zero-gr5_ri-kw.rr_states.hp                        |ADDED
+forward_run.zero-gr5_ri-kw.rr_states.ht                        |ADDED
+forward_run.zero-gr5_ri-kw.sim_q                               |ADDED
+forward_run.zero-gr5_ri-lag0.cost                              |ADDED
+forward_run.zero-gr5_ri-lag0.jobs                              |ADDED
+forward_run.zero-gr5_ri-lag0.q_domain                          |ADDED
+forward_run.zero-gr5_ri-lag0.rr_states.hi                      |ADDED
+forward_run.zero-gr5_ri-lag0.rr_states.hp                      |ADDED
+forward_run.zero-gr5_ri-lag0.rr_states.ht                      |ADDED
+forward_run.zero-gr5_ri-lag0.sim_q                             |ADDED
+forward_run.zero-gr5_ri-lr.cost                                |ADDED
+forward_run.zero-gr5_ri-lr.jobs                                |ADDED
+forward_run.zero-gr5_ri-lr.q_domain                            |ADDED
+forward_run.zero-gr5_ri-lr.rr_states.hi                        |ADDED
+forward_run.zero-gr5_ri-lr.rr_states.hlr                       |ADDED
+forward_run.zero-gr5_ri-lr.rr_states.hp                        |ADDED
+forward_run.zero-gr5_ri-lr.rr_states.ht                        |ADDED
+forward_run.zero-gr5_ri-lr.sim_q                               |ADDED
 forward_run.zero-gr6-kw.cost                                   |NON MODIFIED
 forward_run.zero-gr6-kw.jobs                                   |NON MODIFIED
 forward_run.zero-gr6-kw.q_domain                               |NON MODIFIED
@@ -408,6 +448,39 @@ internal_fluxes.zero-gr4_ode_mlp-lr.lexc                       |NON MODIFIED
 internal_fluxes.zero-gr4_ode_mlp-lr.pn                         |NON MODIFIED
 internal_fluxes.zero-gr4_ode_mlp-lr.qt                         |NON MODIFIED
 internal_fluxes.zero-gr4_ode_mlp-lr.qup                        |NON MODIFIED
+internal_fluxes.zero-gr4_ri-kw.en                              |ADDED
+internal_fluxes.zero-gr4_ri-kw.lexc                            |ADDED
+internal_fluxes.zero-gr4_ri-kw.perc                            |ADDED
+internal_fluxes.zero-gr4_ri-kw.pn                              |ADDED
+internal_fluxes.zero-gr4_ri-kw.pr                              |ADDED
+internal_fluxes.zero-gr4_ri-kw.prd                             |ADDED
+internal_fluxes.zero-gr4_ri-kw.prr                             |ADDED
+internal_fluxes.zero-gr4_ri-kw.qd                              |ADDED
+internal_fluxes.zero-gr4_ri-kw.qim1j                           |ADDED
+internal_fluxes.zero-gr4_ri-kw.qr                              |ADDED
+internal_fluxes.zero-gr4_ri-kw.qt                              |ADDED
+internal_fluxes.zero-gr4_ri-lag0.en                            |ADDED
+internal_fluxes.zero-gr4_ri-lag0.lexc                          |ADDED
+internal_fluxes.zero-gr4_ri-lag0.perc                          |ADDED
+internal_fluxes.zero-gr4_ri-lag0.pn                            |ADDED
+internal_fluxes.zero-gr4_ri-lag0.pr                            |ADDED
+internal_fluxes.zero-gr4_ri-lag0.prd                           |ADDED
+internal_fluxes.zero-gr4_ri-lag0.prr                           |ADDED
+internal_fluxes.zero-gr4_ri-lag0.qd                            |ADDED
+internal_fluxes.zero-gr4_ri-lag0.qr                            |ADDED
+internal_fluxes.zero-gr4_ri-lag0.qt                            |ADDED
+internal_fluxes.zero-gr4_ri-lag0.qup                           |ADDED
+internal_fluxes.zero-gr4_ri-lr.en                              |ADDED
+internal_fluxes.zero-gr4_ri-lr.lexc                            |ADDED
+internal_fluxes.zero-gr4_ri-lr.perc                            |ADDED
+internal_fluxes.zero-gr4_ri-lr.pn                              |ADDED
+internal_fluxes.zero-gr4_ri-lr.pr                              |ADDED
+internal_fluxes.zero-gr4_ri-lr.prd                             |ADDED
+internal_fluxes.zero-gr4_ri-lr.prr                             |ADDED
+internal_fluxes.zero-gr4_ri-lr.qd                              |ADDED
+internal_fluxes.zero-gr4_ri-lr.qr                              |ADDED
+internal_fluxes.zero-gr4_ri-lr.qt                              |ADDED
+internal_fluxes.zero-gr4_ri-lr.qup                             |ADDED
 internal_fluxes.zero-gr5-kw.en                                 |NON MODIFIED
 internal_fluxes.zero-gr5-kw.lexc                               |NON MODIFIED
 internal_fluxes.zero-gr5-kw.perc                               |NON MODIFIED
@@ -441,6 +514,39 @@ internal_fluxes.zero-gr5-lr.qd                                 |NON MODIFIED
 internal_fluxes.zero-gr5-lr.qr                                 |NON MODIFIED
 internal_fluxes.zero-gr5-lr.qt                                 |NON MODIFIED
 internal_fluxes.zero-gr5-lr.qup                                |NON MODIFIED
+internal_fluxes.zero-gr5_ri-kw.en                              |ADDED
+internal_fluxes.zero-gr5_ri-kw.lexc                            |ADDED
+internal_fluxes.zero-gr5_ri-kw.perc                            |ADDED
+internal_fluxes.zero-gr5_ri-kw.pn                              |ADDED
+internal_fluxes.zero-gr5_ri-kw.pr                              |ADDED
+internal_fluxes.zero-gr5_ri-kw.prd                             |ADDED
+internal_fluxes.zero-gr5_ri-kw.prr                             |ADDED
+internal_fluxes.zero-gr5_ri-kw.qd                              |ADDED
+internal_fluxes.zero-gr5_ri-kw.qim1j                           |ADDED
+internal_fluxes.zero-gr5_ri-kw.qr                              |ADDED
+internal_fluxes.zero-gr5_ri-kw.qt                              |ADDED
+internal_fluxes.zero-gr5_ri-lag0.en                            |ADDED
+internal_fluxes.zero-gr5_ri-lag0.lexc                          |ADDED
+internal_fluxes.zero-gr5_ri-lag0.perc                          |ADDED
+internal_fluxes.zero-gr5_ri-lag0.pn                            |ADDED
+internal_fluxes.zero-gr5_ri-lag0.pr                            |ADDED
+internal_fluxes.zero-gr5_ri-lag0.prd                           |ADDED
+internal_fluxes.zero-gr5_ri-lag0.prr                           |ADDED
+internal_fluxes.zero-gr5_ri-lag0.qd                            |ADDED
+internal_fluxes.zero-gr5_ri-lag0.qr                            |ADDED
+internal_fluxes.zero-gr5_ri-lag0.qt                            |ADDED
+internal_fluxes.zero-gr5_ri-lag0.qup                           |ADDED
+internal_fluxes.zero-gr5_ri-lr.en                              |ADDED
+internal_fluxes.zero-gr5_ri-lr.lexc                            |ADDED
+internal_fluxes.zero-gr5_ri-lr.perc                            |ADDED
+internal_fluxes.zero-gr5_ri-lr.pn                              |ADDED
+internal_fluxes.zero-gr5_ri-lr.pr                              |ADDED
+internal_fluxes.zero-gr5_ri-lr.prd                             |ADDED
+internal_fluxes.zero-gr5_ri-lr.prr                             |ADDED
+internal_fluxes.zero-gr5_ri-lr.qd                              |ADDED
+internal_fluxes.zero-gr5_ri-lr.qr                              |ADDED
+internal_fluxes.zero-gr5_ri-lr.qt                              |ADDED
+internal_fluxes.zero-gr5_ri-lr.qup                             |ADDED
 internal_fluxes.zero-gr6-kw.en                                 |NON MODIFIED
 internal_fluxes.zero-gr6-kw.lexc                               |NON MODIFIED
 internal_fluxes.zero-gr6-kw.perc                               |NON MODIFIED
@@ -691,8 +797,8 @@ net_init.weight_layer_1                                        |NON MODIFIED
 net_init.weight_layer_2                                        |NON MODIFIED
 net_init.weight_layer_3                                        |NON MODIFIED
 net_init.weight_layer_4                                        |NON MODIFIED
-optimize.zero-gr4-kw.ann.control_vector                        |MODIFIED
-optimize.zero-gr4-kw.ann.sim_q                                 |MODIFIED
+optimize.zero-gr4-kw.ann.control_vector                        |NON MODIFIED
+optimize.zero-gr4-kw.ann.sim_q                                 |NON MODIFIED
 optimize.zero-gr4-kw.distributed.control_vector                |NON MODIFIED
 optimize.zero-gr4-kw.distributed.sim_q                         |NON MODIFIED
 optimize.zero-gr4-kw.multi-linear.control_vector               |NON MODIFIED
@@ -701,8 +807,8 @@ optimize.zero-gr4-kw.multi-polynomial.control_vector           |NON MODIFIED
 optimize.zero-gr4-kw.multi-polynomial.sim_q                    |NON MODIFIED
 optimize.zero-gr4-kw.uniform.control_vector                    |NON MODIFIED
 optimize.zero-gr4-kw.uniform.sim_q                             |NON MODIFIED
-optimize.zero-gr4-lag0.ann.control_vector                      |MODIFIED
-optimize.zero-gr4-lag0.ann.sim_q                               |MODIFIED
+optimize.zero-gr4-lag0.ann.control_vector                      |NON MODIFIED
+optimize.zero-gr4-lag0.ann.sim_q                               |NON MODIFIED
 optimize.zero-gr4-lag0.distributed.control_vector              |NON MODIFIED
 optimize.zero-gr4-lag0.distributed.sim_q                       |NON MODIFIED
 optimize.zero-gr4-lag0.multi-linear.control_vector             |NON MODIFIED
@@ -711,8 +817,8 @@ optimize.zero-gr4-lag0.multi-polynomial.control_vector         |NON MODIFIED
 optimize.zero-gr4-lag0.multi-polynomial.sim_q                  |NON MODIFIED
 optimize.zero-gr4-lag0.uniform.control_vector                  |NON MODIFIED
 optimize.zero-gr4-lag0.uniform.sim_q                           |NON MODIFIED
-optimize.zero-gr4-lr.ann.control_vector                        |MODIFIED
-optimize.zero-gr4-lr.ann.sim_q                                 |MODIFIED
+optimize.zero-gr4-lr.ann.control_vector                        |NON MODIFIED
+optimize.zero-gr4-lr.ann.sim_q                                 |NON MODIFIED
 optimize.zero-gr4-lr.distributed.control_vector                |NON MODIFIED
 optimize.zero-gr4-lr.distributed.sim_q                         |NON MODIFIED
 optimize.zero-gr4-lr.multi-linear.control_vector               |NON MODIFIED
@@ -721,8 +827,8 @@ optimize.zero-gr4-lr.multi-polynomial.control_vector           |NON MODIFIED
 optimize.zero-gr4-lr.multi-polynomial.sim_q                    |NON MODIFIED
 optimize.zero-gr4-lr.uniform.control_vector                    |NON MODIFIED
 optimize.zero-gr4-lr.uniform.sim_q                             |NON MODIFIED
-optimize.zero-gr4_mlp-kw.ann.control_vector                    |MODIFIED
-optimize.zero-gr4_mlp-kw.ann.sim_q                             |MODIFIED
+optimize.zero-gr4_mlp-kw.ann.control_vector                    |NON MODIFIED
+optimize.zero-gr4_mlp-kw.ann.sim_q                             |NON MODIFIED
 optimize.zero-gr4_mlp-kw.distributed.control_vector            |NON MODIFIED
 optimize.zero-gr4_mlp-kw.distributed.sim_q                     |NON MODIFIED
 optimize.zero-gr4_mlp-kw.multi-linear.control_vector           |NON MODIFIED
@@ -731,8 +837,8 @@ optimize.zero-gr4_mlp-kw.multi-polynomial.control_vector       |NON MODIFIED
 optimize.zero-gr4_mlp-kw.multi-polynomial.sim_q                |NON MODIFIED
 optimize.zero-gr4_mlp-kw.uniform.control_vector                |NON MODIFIED
 optimize.zero-gr4_mlp-kw.uniform.sim_q                         |NON MODIFIED
-optimize.zero-gr4_mlp-lag0.ann.control_vector                  |MODIFIED
-optimize.zero-gr4_mlp-lag0.ann.sim_q                           |MODIFIED
+optimize.zero-gr4_mlp-lag0.ann.control_vector                  |NON MODIFIED
+optimize.zero-gr4_mlp-lag0.ann.sim_q                           |NON MODIFIED
 optimize.zero-gr4_mlp-lag0.distributed.control_vector          |NON MODIFIED
 optimize.zero-gr4_mlp-lag0.distributed.sim_q                   |NON MODIFIED
 optimize.zero-gr4_mlp-lag0.multi-linear.control_vector         |NON MODIFIED
@@ -741,8 +847,8 @@ optimize.zero-gr4_mlp-lag0.multi-polynomial.control_vector     |NON MODIFIED
 optimize.zero-gr4_mlp-lag0.multi-polynomial.sim_q              |NON MODIFIED
 optimize.zero-gr4_mlp-lag0.uniform.control_vector              |NON MODIFIED
 optimize.zero-gr4_mlp-lag0.uniform.sim_q                       |NON MODIFIED
-optimize.zero-gr4_mlp-lr.ann.control_vector                    |MODIFIED
-optimize.zero-gr4_mlp-lr.ann.sim_q                             |MODIFIED
+optimize.zero-gr4_mlp-lr.ann.control_vector                    |NON MODIFIED
+optimize.zero-gr4_mlp-lr.ann.sim_q                             |NON MODIFIED
 optimize.zero-gr4_mlp-lr.distributed.control_vector            |NON MODIFIED
 optimize.zero-gr4_mlp-lr.distributed.sim_q                     |NON MODIFIED
 optimize.zero-gr4_mlp-lr.multi-linear.control_vector           |NON MODIFIED
@@ -751,8 +857,8 @@ optimize.zero-gr4_mlp-lr.multi-polynomial.control_vector       |NON MODIFIED
 optimize.zero-gr4_mlp-lr.multi-polynomial.sim_q                |NON MODIFIED
 optimize.zero-gr4_mlp-lr.uniform.control_vector                |NON MODIFIED
 optimize.zero-gr4_mlp-lr.uniform.sim_q                         |NON MODIFIED
-optimize.zero-gr4_ode-kw.ann.control_vector                    |MODIFIED
-optimize.zero-gr4_ode-kw.ann.sim_q                             |MODIFIED
+optimize.zero-gr4_ode-kw.ann.control_vector                    |NON MODIFIED
+optimize.zero-gr4_ode-kw.ann.sim_q                             |NON MODIFIED
 optimize.zero-gr4_ode-kw.distributed.control_vector            |NON MODIFIED
 optimize.zero-gr4_ode-kw.distributed.sim_q                     |NON MODIFIED
 optimize.zero-gr4_ode-kw.multi-linear.control_vector           |NON MODIFIED
@@ -761,8 +867,8 @@ optimize.zero-gr4_ode-kw.multi-polynomial.control_vector       |NON MODIFIED
 optimize.zero-gr4_ode-kw.multi-polynomial.sim_q                |NON MODIFIED
 optimize.zero-gr4_ode-kw.uniform.control_vector                |NON MODIFIED
 optimize.zero-gr4_ode-kw.uniform.sim_q                         |NON MODIFIED
-optimize.zero-gr4_ode-lag0.ann.control_vector                  |MODIFIED
-optimize.zero-gr4_ode-lag0.ann.sim_q                           |MODIFIED
+optimize.zero-gr4_ode-lag0.ann.control_vector                  |NON MODIFIED
+optimize.zero-gr4_ode-lag0.ann.sim_q                           |NON MODIFIED
 optimize.zero-gr4_ode-lag0.distributed.control_vector          |NON MODIFIED
 optimize.zero-gr4_ode-lag0.distributed.sim_q                   |NON MODIFIED
 optimize.zero-gr4_ode-lag0.multi-linear.control_vector         |NON MODIFIED
@@ -771,8 +877,8 @@ optimize.zero-gr4_ode-lag0.multi-polynomial.control_vector     |NON MODIFIED
 optimize.zero-gr4_ode-lag0.multi-polynomial.sim_q              |NON MODIFIED
 optimize.zero-gr4_ode-lag0.uniform.control_vector              |NON MODIFIED
 optimize.zero-gr4_ode-lag0.uniform.sim_q                       |NON MODIFIED
-optimize.zero-gr4_ode-lr.ann.control_vector                    |MODIFIED
-optimize.zero-gr4_ode-lr.ann.sim_q                             |MODIFIED
+optimize.zero-gr4_ode-lr.ann.control_vector                    |NON MODIFIED
+optimize.zero-gr4_ode-lr.ann.sim_q                             |NON MODIFIED
 optimize.zero-gr4_ode-lr.distributed.control_vector            |NON MODIFIED
 optimize.zero-gr4_ode-lr.distributed.sim_q                     |NON MODIFIED
 optimize.zero-gr4_ode-lr.multi-linear.control_vector           |NON MODIFIED
@@ -781,8 +887,8 @@ optimize.zero-gr4_ode-lr.multi-polynomial.control_vector       |NON MODIFIED
 optimize.zero-gr4_ode-lr.multi-polynomial.sim_q                |NON MODIFIED
 optimize.zero-gr4_ode-lr.uniform.control_vector                |NON MODIFIED
 optimize.zero-gr4_ode-lr.uniform.sim_q                         |NON MODIFIED
-optimize.zero-gr4_ode_mlp-kw.ann.control_vector                |MODIFIED
-optimize.zero-gr4_ode_mlp-kw.ann.sim_q                         |MODIFIED
+optimize.zero-gr4_ode_mlp-kw.ann.control_vector                |NON MODIFIED
+optimize.zero-gr4_ode_mlp-kw.ann.sim_q                         |NON MODIFIED
 optimize.zero-gr4_ode_mlp-kw.distributed.control_vector        |NON MODIFIED
 optimize.zero-gr4_ode_mlp-kw.distributed.sim_q                 |NON MODIFIED
 optimize.zero-gr4_ode_mlp-kw.multi-linear.control_vector       |NON MODIFIED
@@ -791,8 +897,8 @@ optimize.zero-gr4_ode_mlp-kw.multi-polynomial.control_vector   |NON MODIFIED
 optimize.zero-gr4_ode_mlp-kw.multi-polynomial.sim_q            |NON MODIFIED
 optimize.zero-gr4_ode_mlp-kw.uniform.control_vector            |NON MODIFIED
 optimize.zero-gr4_ode_mlp-kw.uniform.sim_q                     |NON MODIFIED
-optimize.zero-gr4_ode_mlp-lag0.ann.control_vector              |MODIFIED
-optimize.zero-gr4_ode_mlp-lag0.ann.sim_q                       |MODIFIED
+optimize.zero-gr4_ode_mlp-lag0.ann.control_vector              |NON MODIFIED
+optimize.zero-gr4_ode_mlp-lag0.ann.sim_q                       |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lag0.distributed.control_vector      |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lag0.distributed.sim_q               |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lag0.multi-linear.control_vector     |NON MODIFIED
@@ -801,8 +907,8 @@ optimize.zero-gr4_ode_mlp-lag0.multi-polynomial.control_vector |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lag0.multi-polynomial.sim_q          |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lag0.uniform.control_vector          |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lag0.uniform.sim_q                   |NON MODIFIED
-optimize.zero-gr4_ode_mlp-lr.ann.control_vector                |MODIFIED
-optimize.zero-gr4_ode_mlp-lr.ann.sim_q                         |MODIFIED
+optimize.zero-gr4_ode_mlp-lr.ann.control_vector                |NON MODIFIED
+optimize.zero-gr4_ode_mlp-lr.ann.sim_q                         |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lr.distributed.control_vector        |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lr.distributed.sim_q                 |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lr.multi-linear.control_vector       |NON MODIFIED
@@ -811,8 +917,38 @@ optimize.zero-gr4_ode_mlp-lr.multi-polynomial.control_vector   |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lr.multi-polynomial.sim_q            |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lr.uniform.control_vector            |NON MODIFIED
 optimize.zero-gr4_ode_mlp-lr.uniform.sim_q                     |NON MODIFIED
-optimize.zero-gr5-kw.ann.control_vector                        |MODIFIED
-optimize.zero-gr5-kw.ann.sim_q                                 |MODIFIED
+optimize.zero-gr4_ri-kw.ann.control_vector                     |ADDED
+optimize.zero-gr4_ri-kw.ann.sim_q                              |ADDED
+optimize.zero-gr4_ri-kw.distributed.control_vector             |ADDED
+optimize.zero-gr4_ri-kw.distributed.sim_q                      |ADDED
+optimize.zero-gr4_ri-kw.multi-linear.control_vector            |ADDED
+optimize.zero-gr4_ri-kw.multi-linear.sim_q                     |ADDED
+optimize.zero-gr4_ri-kw.multi-polynomial.control_vector        |ADDED
+optimize.zero-gr4_ri-kw.multi-polynomial.sim_q                 |ADDED
+optimize.zero-gr4_ri-kw.uniform.control_vector                 |ADDED
+optimize.zero-gr4_ri-kw.uniform.sim_q                          |ADDED
+optimize.zero-gr4_ri-lag0.ann.control_vector                   |ADDED
+optimize.zero-gr4_ri-lag0.ann.sim_q                            |ADDED
+optimize.zero-gr4_ri-lag0.distributed.control_vector           |ADDED
+optimize.zero-gr4_ri-lag0.distributed.sim_q                    |ADDED
+optimize.zero-gr4_ri-lag0.multi-linear.control_vector          |ADDED
+optimize.zero-gr4_ri-lag0.multi-linear.sim_q                   |ADDED
+optimize.zero-gr4_ri-lag0.multi-polynomial.control_vector      |ADDED
+optimize.zero-gr4_ri-lag0.multi-polynomial.sim_q               |ADDED
+optimize.zero-gr4_ri-lag0.uniform.control_vector               |ADDED
+optimize.zero-gr4_ri-lag0.uniform.sim_q                        |ADDED
+optimize.zero-gr4_ri-lr.ann.control_vector                     |ADDED
+optimize.zero-gr4_ri-lr.ann.sim_q                              |ADDED
+optimize.zero-gr4_ri-lr.distributed.control_vector             |ADDED
+optimize.zero-gr4_ri-lr.distributed.sim_q                      |ADDED
+optimize.zero-gr4_ri-lr.multi-linear.control_vector            |ADDED
+optimize.zero-gr4_ri-lr.multi-linear.sim_q                     |ADDED
+optimize.zero-gr4_ri-lr.multi-polynomial.control_vector        |ADDED
+optimize.zero-gr4_ri-lr.multi-polynomial.sim_q                 |ADDED
+optimize.zero-gr4_ri-lr.uniform.control_vector                 |ADDED
+optimize.zero-gr4_ri-lr.uniform.sim_q                          |ADDED
+optimize.zero-gr5-kw.ann.control_vector                        |NON MODIFIED
+optimize.zero-gr5-kw.ann.sim_q                                 |NON MODIFIED
 optimize.zero-gr5-kw.distributed.control_vector                |NON MODIFIED
 optimize.zero-gr5-kw.distributed.sim_q                         |NON MODIFIED
 optimize.zero-gr5-kw.multi-linear.control_vector               |NON MODIFIED
@@ -821,8 +957,8 @@ optimize.zero-gr5-kw.multi-polynomial.control_vector           |NON MODIFIED
 optimize.zero-gr5-kw.multi-polynomial.sim_q                    |NON MODIFIED
 optimize.zero-gr5-kw.uniform.control_vector                    |NON MODIFIED
 optimize.zero-gr5-kw.uniform.sim_q                             |NON MODIFIED
-optimize.zero-gr5-lag0.ann.control_vector                      |MODIFIED
-optimize.zero-gr5-lag0.ann.sim_q                               |MODIFIED
+optimize.zero-gr5-lag0.ann.control_vector                      |NON MODIFIED
+optimize.zero-gr5-lag0.ann.sim_q                               |NON MODIFIED
 optimize.zero-gr5-lag0.distributed.control_vector              |NON MODIFIED
 optimize.zero-gr5-lag0.distributed.sim_q                       |NON MODIFIED
 optimize.zero-gr5-lag0.multi-linear.control_vector             |NON MODIFIED
@@ -831,8 +967,8 @@ optimize.zero-gr5-lag0.multi-polynomial.control_vector         |NON MODIFIED
 optimize.zero-gr5-lag0.multi-polynomial.sim_q                  |NON MODIFIED
 optimize.zero-gr5-lag0.uniform.control_vector                  |NON MODIFIED
 optimize.zero-gr5-lag0.uniform.sim_q                           |NON MODIFIED
-optimize.zero-gr5-lr.ann.control_vector                        |MODIFIED
-optimize.zero-gr5-lr.ann.sim_q                                 |MODIFIED
+optimize.zero-gr5-lr.ann.control_vector                        |NON MODIFIED
+optimize.zero-gr5-lr.ann.sim_q                                 |NON MODIFIED
 optimize.zero-gr5-lr.distributed.control_vector                |NON MODIFIED
 optimize.zero-gr5-lr.distributed.sim_q                         |NON MODIFIED
 optimize.zero-gr5-lr.multi-linear.control_vector               |NON MODIFIED
@@ -841,8 +977,38 @@ optimize.zero-gr5-lr.multi-polynomial.control_vector           |NON MODIFIED
 optimize.zero-gr5-lr.multi-polynomial.sim_q                    |NON MODIFIED
 optimize.zero-gr5-lr.uniform.control_vector                    |NON MODIFIED
 optimize.zero-gr5-lr.uniform.sim_q                             |NON MODIFIED
-optimize.zero-gr6-kw.ann.control_vector                        |MODIFIED
-optimize.zero-gr6-kw.ann.sim_q                                 |MODIFIED
+optimize.zero-gr5_ri-kw.ann.control_vector                     |ADDED
+optimize.zero-gr5_ri-kw.ann.sim_q                              |ADDED
+optimize.zero-gr5_ri-kw.distributed.control_vector             |ADDED
+optimize.zero-gr5_ri-kw.distributed.sim_q                      |ADDED
+optimize.zero-gr5_ri-kw.multi-linear.control_vector            |ADDED
+optimize.zero-gr5_ri-kw.multi-linear.sim_q                     |ADDED
+optimize.zero-gr5_ri-kw.multi-polynomial.control_vector        |ADDED
+optimize.zero-gr5_ri-kw.multi-polynomial.sim_q                 |ADDED
+optimize.zero-gr5_ri-kw.uniform.control_vector                 |ADDED
+optimize.zero-gr5_ri-kw.uniform.sim_q                          |ADDED
+optimize.zero-gr5_ri-lag0.ann.control_vector                   |ADDED
+optimize.zero-gr5_ri-lag0.ann.sim_q                            |ADDED
+optimize.zero-gr5_ri-lag0.distributed.control_vector           |ADDED
+optimize.zero-gr5_ri-lag0.distributed.sim_q                    |ADDED
+optimize.zero-gr5_ri-lag0.multi-linear.control_vector          |ADDED
+optimize.zero-gr5_ri-lag0.multi-linear.sim_q                   |ADDED
+optimize.zero-gr5_ri-lag0.multi-polynomial.control_vector      |ADDED
+optimize.zero-gr5_ri-lag0.multi-polynomial.sim_q               |ADDED
+optimize.zero-gr5_ri-lag0.uniform.control_vector               |ADDED
+optimize.zero-gr5_ri-lag0.uniform.sim_q                        |ADDED
+optimize.zero-gr5_ri-lr.ann.control_vector                     |ADDED
+optimize.zero-gr5_ri-lr.ann.sim_q                              |ADDED
+optimize.zero-gr5_ri-lr.distributed.control_vector             |ADDED
+optimize.zero-gr5_ri-lr.distributed.sim_q                      |ADDED
+optimize.zero-gr5_ri-lr.multi-linear.control_vector            |ADDED
+optimize.zero-gr5_ri-lr.multi-linear.sim_q                     |ADDED
+optimize.zero-gr5_ri-lr.multi-polynomial.control_vector        |ADDED
+optimize.zero-gr5_ri-lr.multi-polynomial.sim_q                 |ADDED
+optimize.zero-gr5_ri-lr.uniform.control_vector                 |ADDED
+optimize.zero-gr5_ri-lr.uniform.sim_q                          |ADDED
+optimize.zero-gr6-kw.ann.control_vector                        |NON MODIFIED
+optimize.zero-gr6-kw.ann.sim_q                                 |NON MODIFIED
 optimize.zero-gr6-kw.distributed.control_vector                |NON MODIFIED
 optimize.zero-gr6-kw.distributed.sim_q                         |NON MODIFIED
 optimize.zero-gr6-kw.multi-linear.control_vector               |NON MODIFIED
@@ -851,8 +1017,8 @@ optimize.zero-gr6-kw.multi-polynomial.control_vector           |NON MODIFIED
 optimize.zero-gr6-kw.multi-polynomial.sim_q                    |NON MODIFIED
 optimize.zero-gr6-kw.uniform.control_vector                    |NON MODIFIED
 optimize.zero-gr6-kw.uniform.sim_q                             |NON MODIFIED
-optimize.zero-gr6-lag0.ann.control_vector                      |MODIFIED
-optimize.zero-gr6-lag0.ann.sim_q                               |MODIFIED
+optimize.zero-gr6-lag0.ann.control_vector                      |NON MODIFIED
+optimize.zero-gr6-lag0.ann.sim_q                               |NON MODIFIED
 optimize.zero-gr6-lag0.distributed.control_vector              |NON MODIFIED
 optimize.zero-gr6-lag0.distributed.sim_q                       |NON MODIFIED
 optimize.zero-gr6-lag0.multi-linear.control_vector             |NON MODIFIED
@@ -861,8 +1027,8 @@ optimize.zero-gr6-lag0.multi-polynomial.control_vector         |NON MODIFIED
 optimize.zero-gr6-lag0.multi-polynomial.sim_q                  |NON MODIFIED
 optimize.zero-gr6-lag0.uniform.control_vector                  |NON MODIFIED
 optimize.zero-gr6-lag0.uniform.sim_q                           |NON MODIFIED
-optimize.zero-gr6-lr.ann.control_vector                        |MODIFIED
-optimize.zero-gr6-lr.ann.sim_q                                 |MODIFIED
+optimize.zero-gr6-lr.ann.control_vector                        |NON MODIFIED
+optimize.zero-gr6-lr.ann.sim_q                                 |NON MODIFIED
 optimize.zero-gr6-lr.distributed.control_vector                |NON MODIFIED
 optimize.zero-gr6-lr.distributed.sim_q                         |NON MODIFIED
 optimize.zero-gr6-lr.multi-linear.control_vector               |NON MODIFIED
@@ -871,8 +1037,8 @@ optimize.zero-gr6-lr.multi-polynomial.control_vector           |NON MODIFIED
 optimize.zero-gr6-lr.multi-polynomial.sim_q                    |NON MODIFIED
 optimize.zero-gr6-lr.uniform.control_vector                    |NON MODIFIED
 optimize.zero-gr6-lr.uniform.sim_q                             |NON MODIFIED
-optimize.zero-grc-kw.ann.control_vector                        |MODIFIED
-optimize.zero-grc-kw.ann.sim_q                                 |MODIFIED
+optimize.zero-grc-kw.ann.control_vector                        |NON MODIFIED
+optimize.zero-grc-kw.ann.sim_q                                 |NON MODIFIED
 optimize.zero-grc-kw.distributed.control_vector                |NON MODIFIED
 optimize.zero-grc-kw.distributed.sim_q                         |NON MODIFIED
 optimize.zero-grc-kw.multi-linear.control_vector               |NON MODIFIED
@@ -881,8 +1047,8 @@ optimize.zero-grc-kw.multi-polynomial.control_vector           |NON MODIFIED
 optimize.zero-grc-kw.multi-polynomial.sim_q                    |NON MODIFIED
 optimize.zero-grc-kw.uniform.control_vector                    |NON MODIFIED
 optimize.zero-grc-kw.uniform.sim_q                             |NON MODIFIED
-optimize.zero-grc-lag0.ann.control_vector                      |MODIFIED
-optimize.zero-grc-lag0.ann.sim_q                               |MODIFIED
+optimize.zero-grc-lag0.ann.control_vector                      |NON MODIFIED
+optimize.zero-grc-lag0.ann.sim_q                               |NON MODIFIED
 optimize.zero-grc-lag0.distributed.control_vector              |NON MODIFIED
 optimize.zero-grc-lag0.distributed.sim_q                       |NON MODIFIED
 optimize.zero-grc-lag0.multi-linear.control_vector             |NON MODIFIED
@@ -891,8 +1057,8 @@ optimize.zero-grc-lag0.multi-polynomial.control_vector         |NON MODIFIED
 optimize.zero-grc-lag0.multi-polynomial.sim_q                  |NON MODIFIED
 optimize.zero-grc-lag0.uniform.control_vector                  |NON MODIFIED
 optimize.zero-grc-lag0.uniform.sim_q                           |NON MODIFIED
-optimize.zero-grc-lr.ann.control_vector                        |MODIFIED
-optimize.zero-grc-lr.ann.sim_q                                 |MODIFIED
+optimize.zero-grc-lr.ann.control_vector                        |NON MODIFIED
+optimize.zero-grc-lr.ann.sim_q                                 |NON MODIFIED
 optimize.zero-grc-lr.distributed.control_vector                |NON MODIFIED
 optimize.zero-grc-lr.distributed.sim_q                         |NON MODIFIED
 optimize.zero-grc-lr.multi-linear.control_vector               |NON MODIFIED
@@ -901,8 +1067,8 @@ optimize.zero-grc-lr.multi-polynomial.control_vector           |NON MODIFIED
 optimize.zero-grc-lr.multi-polynomial.sim_q                    |NON MODIFIED
 optimize.zero-grc-lr.uniform.control_vector                    |NON MODIFIED
 optimize.zero-grc-lr.uniform.sim_q                             |NON MODIFIED
-optimize.zero-grd-kw.ann.control_vector                        |MODIFIED
-optimize.zero-grd-kw.ann.sim_q                                 |MODIFIED
+optimize.zero-grd-kw.ann.control_vector                        |NON MODIFIED
+optimize.zero-grd-kw.ann.sim_q                                 |NON MODIFIED
 optimize.zero-grd-kw.distributed.control_vector                |NON MODIFIED
 optimize.zero-grd-kw.distributed.sim_q                         |NON MODIFIED
 optimize.zero-grd-kw.multi-linear.control_vector               |NON MODIFIED
@@ -911,8 +1077,8 @@ optimize.zero-grd-kw.multi-polynomial.control_vector           |NON MODIFIED
 optimize.zero-grd-kw.multi-polynomial.sim_q                    |NON MODIFIED
 optimize.zero-grd-kw.uniform.control_vector                    |NON MODIFIED
 optimize.zero-grd-kw.uniform.sim_q                             |NON MODIFIED
-optimize.zero-grd-lag0.ann.control_vector                      |MODIFIED
-optimize.zero-grd-lag0.ann.sim_q                               |MODIFIED
+optimize.zero-grd-lag0.ann.control_vector                      |NON MODIFIED
+optimize.zero-grd-lag0.ann.sim_q                               |NON MODIFIED
 optimize.zero-grd-lag0.distributed.control_vector              |NON MODIFIED
 optimize.zero-grd-lag0.distributed.sim_q                       |NON MODIFIED
 optimize.zero-grd-lag0.multi-linear.control_vector             |NON MODIFIED
@@ -921,8 +1087,8 @@ optimize.zero-grd-lag0.multi-polynomial.control_vector         |NON MODIFIED
 optimize.zero-grd-lag0.multi-polynomial.sim_q                  |NON MODIFIED
 optimize.zero-grd-lag0.uniform.control_vector                  |NON MODIFIED
 optimize.zero-grd-lag0.uniform.sim_q                           |NON MODIFIED
-optimize.zero-grd-lr.ann.control_vector                        |MODIFIED
-optimize.zero-grd-lr.ann.sim_q                                 |MODIFIED
+optimize.zero-grd-lr.ann.control_vector                        |NON MODIFIED
+optimize.zero-grd-lr.ann.sim_q                                 |NON MODIFIED
 optimize.zero-grd-lr.distributed.control_vector                |NON MODIFIED
 optimize.zero-grd-lr.distributed.sim_q                         |NON MODIFIED
 optimize.zero-grd-lr.multi-linear.control_vector               |NON MODIFIED
@@ -931,8 +1097,8 @@ optimize.zero-grd-lr.multi-polynomial.control_vector           |NON MODIFIED
 optimize.zero-grd-lr.multi-polynomial.sim_q                    |NON MODIFIED
 optimize.zero-grd-lr.uniform.control_vector                    |NON MODIFIED
 optimize.zero-grd-lr.uniform.sim_q                             |NON MODIFIED
-optimize.zero-loieau-kw.ann.control_vector                     |MODIFIED
-optimize.zero-loieau-kw.ann.sim_q                              |MODIFIED
+optimize.zero-loieau-kw.ann.control_vector                     |NON MODIFIED
+optimize.zero-loieau-kw.ann.sim_q                              |NON MODIFIED
 optimize.zero-loieau-kw.distributed.control_vector             |NON MODIFIED
 optimize.zero-loieau-kw.distributed.sim_q                      |NON MODIFIED
 optimize.zero-loieau-kw.multi-linear.control_vector            |NON MODIFIED
@@ -941,8 +1107,8 @@ optimize.zero-loieau-kw.multi-polynomial.control_vector        |NON MODIFIED
 optimize.zero-loieau-kw.multi-polynomial.sim_q                 |NON MODIFIED
 optimize.zero-loieau-kw.uniform.control_vector                 |NON MODIFIED
 optimize.zero-loieau-kw.uniform.sim_q                          |NON MODIFIED
-optimize.zero-loieau-lag0.ann.control_vector                   |MODIFIED
-optimize.zero-loieau-lag0.ann.sim_q                            |MODIFIED
+optimize.zero-loieau-lag0.ann.control_vector                   |NON MODIFIED
+optimize.zero-loieau-lag0.ann.sim_q                            |NON MODIFIED
 optimize.zero-loieau-lag0.distributed.control_vector           |NON MODIFIED
 optimize.zero-loieau-lag0.distributed.sim_q                    |NON MODIFIED
 optimize.zero-loieau-lag0.multi-linear.control_vector          |NON MODIFIED
@@ -951,8 +1117,8 @@ optimize.zero-loieau-lag0.multi-polynomial.control_vector      |NON MODIFIED
 optimize.zero-loieau-lag0.multi-polynomial.sim_q               |NON MODIFIED
 optimize.zero-loieau-lag0.uniform.control_vector               |NON MODIFIED
 optimize.zero-loieau-lag0.uniform.sim_q                        |NON MODIFIED
-optimize.zero-loieau-lr.ann.control_vector                     |MODIFIED
-optimize.zero-loieau-lr.ann.sim_q                              |MODIFIED
+optimize.zero-loieau-lr.ann.control_vector                     |NON MODIFIED
+optimize.zero-loieau-lr.ann.sim_q                              |NON MODIFIED
 optimize.zero-loieau-lr.distributed.control_vector             |NON MODIFIED
 optimize.zero-loieau-lr.distributed.sim_q                      |NON MODIFIED
 optimize.zero-loieau-lr.multi-linear.control_vector            |NON MODIFIED
@@ -961,8 +1127,8 @@ optimize.zero-loieau-lr.multi-polynomial.control_vector        |NON MODIFIED
 optimize.zero-loieau-lr.multi-polynomial.sim_q                 |NON MODIFIED
 optimize.zero-loieau-lr.uniform.control_vector                 |NON MODIFIED
 optimize.zero-loieau-lr.uniform.sim_q                          |NON MODIFIED
-optimize.zero-vic3l-kw.ann.control_vector                      |MODIFIED
-optimize.zero-vic3l-kw.ann.sim_q                               |MODIFIED
+optimize.zero-vic3l-kw.ann.control_vector                      |NON MODIFIED
+optimize.zero-vic3l-kw.ann.sim_q                               |NON MODIFIED
 optimize.zero-vic3l-kw.distributed.control_vector              |NON MODIFIED
 optimize.zero-vic3l-kw.distributed.sim_q                       |NON MODIFIED
 optimize.zero-vic3l-kw.multi-linear.control_vector             |NON MODIFIED
@@ -971,8 +1137,8 @@ optimize.zero-vic3l-kw.multi-polynomial.control_vector         |NON MODIFIED
 optimize.zero-vic3l-kw.multi-polynomial.sim_q                  |NON MODIFIED
 optimize.zero-vic3l-kw.uniform.control_vector                  |NON MODIFIED
 optimize.zero-vic3l-kw.uniform.sim_q                           |NON MODIFIED
-optimize.zero-vic3l-lag0.ann.control_vector                    |MODIFIED
-optimize.zero-vic3l-lag0.ann.sim_q                             |MODIFIED
+optimize.zero-vic3l-lag0.ann.control_vector                    |NON MODIFIED
+optimize.zero-vic3l-lag0.ann.sim_q                             |NON MODIFIED
 optimize.zero-vic3l-lag0.distributed.control_vector            |NON MODIFIED
 optimize.zero-vic3l-lag0.distributed.sim_q                     |NON MODIFIED
 optimize.zero-vic3l-lag0.multi-linear.control_vector           |NON MODIFIED
@@ -981,8 +1147,8 @@ optimize.zero-vic3l-lag0.multi-polynomial.control_vector       |NON MODIFIED
 optimize.zero-vic3l-lag0.multi-polynomial.sim_q                |NON MODIFIED
 optimize.zero-vic3l-lag0.uniform.control_vector                |NON MODIFIED
 optimize.zero-vic3l-lag0.uniform.sim_q                         |NON MODIFIED
-optimize.zero-vic3l-lr.ann.control_vector                      |MODIFIED
-optimize.zero-vic3l-lr.ann.sim_q                               |MODIFIED
+optimize.zero-vic3l-lr.ann.control_vector                      |NON MODIFIED
+optimize.zero-vic3l-lr.ann.sim_q                               |NON MODIFIED
 optimize.zero-vic3l-lr.distributed.control_vector              |NON MODIFIED
 optimize.zero-vic3l-lr.distributed.sim_q                       |NON MODIFIED
 optimize.zero-vic3l-lr.multi-linear.control_vector             |NON MODIFIED
diff --git a/smash/tests/test_constant.py b/smash/tests/test_constant.py
index 9d50ae98..f2d9cad5 100644
--- a/smash/tests/test_constant.py
+++ b/smash/tests/test_constant.py
@@ -51,10 +51,12 @@ def test_module_name():
     # % Check hydrological module
     assert HYDROLOGICAL_MODULE == [
         "gr4",
+        "gr4_ri",
         "gr4_mlp",
         "gr4_ode",
         "gr4_ode_mlp",
         "gr5",
+        "gr5_ri",
         "gr6",
         "grc",
         "grd",
@@ -88,10 +90,12 @@ def test_module_parameters():
     # % Check hydrological module rr parameters
     assert list(HYDROLOGICAL_MODULE_RR_PARAMETERS.values()) == [
         ["ci", "cp", "ct", "kexc"],  # % gr4
+        ["ci", "cp", "ct", "alpha1", "alpha2", "kexc"],  # % gr4_ri
         ["ci", "cp", "ct", "kexc"],  # % gr4_mlp
         ["ci", "cp", "ct", "kexc"],  # % gr4_ode
         ["ci", "cp", "ct", "kexc"],  # % gr4_ode_mlp
         ["ci", "cp", "ct", "kexc", "aexc"],  # % gr5
+        ["ci", "cp", "ct", "alpha1", "alpha2", "kexc", "aexc"],  # % gr5_ri
         ["ci", "cp", "ct", "be", "kexc", "aexc"],  # % gr6
         ["ci", "cp", "ct", "cl", "kexc"],  # % grc
         ["cp", "ct"],  # % grd
@@ -102,10 +106,12 @@ def test_module_parameters():
     # % Check hydrological module rr states
     assert list(HYDROLOGICAL_MODULE_RR_STATES.values()) == [
         ["hi", "hp", "ht"],  # % gr4
+        ["hi", "hp", "ht"],  # % gr4_ri
         ["hi", "hp", "ht"],  # % gr4_mlp
         ["hi", "hp", "ht"],  # % gr4_ode
         ["hi", "hp", "ht"],  # % gr4_ode_mlp
         ["hi", "hp", "ht"],  # % gr5
+        ["hi", "hp", "ht"],  # % gr5_ri
         ["hi", "hp", "ht", "he"],  # % gr6
         ["hi", "hp", "ht", "hl"],  # % grc
         ["hp", "ht"],  # % grd
@@ -116,10 +122,12 @@ def test_module_parameters():
     # % Check hydrological module rr internal fluxes
     assert list(HYDROLOGICAL_MODULE_RR_INTERNAL_FLUXES.values()) == [
         ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "qr", "qd", "qt"],  # % gr4
+        ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "qr", "qd", "qt"],  # % gr4-ri
         ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "qr", "qd", "qt"],  # % gr4_mlp
         ["pn", "en", "lexc", "qt"],  # % gr4_ode
         ["pn", "en", "lexc", "qt"],  # % gr4_ode_mlp
         ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "qr", "qd", "qt"],  # % gr5
+        ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "qr", "qd", "qt"],  # % gr5-ri
         ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "pre", "qr", "qd", "qe", "qt"],  # % gr6
         ["pn", "en", "pr", "perc", "lexc", "prr", "prd", "prl", "qr", "qd", "ql", "qt"],  # % grc
         ["ei", "pn", "en", "pr", "perc", "prr", "qr", "qt"],  # % grd
@@ -146,6 +154,8 @@ def test_parameters():
         "ci",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc)
         "cp",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc, grd)
         "ct",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc, grd)
+        "alpha1",  # % (gr4_ri, gr5_ri)
+        "alpha2",  # % (gr4_ri, gr5_ri)
         "cl",  # % grc
         "be",  # % gr6
         "kexc",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6)
@@ -170,9 +180,9 @@ def test_parameters():
     # % Check rainfall-runoff states
     assert RR_STATES == [
         "hs",  # % ssn
-        "hi",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc)
-        "hp",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc, grd)
-        "ht",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr6, grc, grd)
+        "hi",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr5_ri, gr6, grc)
+        "hp",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr5_ri, gr6, grc, grd)
+        "ht",  # % (gr4, gr4_mlp, gr4_ode, gr4_ode_mlp, gr5, gr5_ri, gr6, grc, grd)
         "hl",  # % grc
         "he",  # % gr6
         "ha",  # % loieau
@@ -232,6 +242,8 @@ def test_feasible_domain():
         (0, np.inf),  # % ci
         (0, np.inf),  # % cp
         (0, np.inf),  # % ct
+        (0, np.inf),  # % alpha1
+        (0, np.inf),  # % alpha2
         (0, np.inf),  # % cl
         (0, np.inf),  # % be
         (-np.inf, np.inf),  # % kexc
@@ -291,6 +303,8 @@ def test_default_parameters():
         1e-6,  # % ci
         200,  # % cp
         500,  # % ct
+        3e-4,  # % alpha1
+        1e-3,  # % alpha2
         500,  # % cl
         10,  # % be
         0,  # % kexc
@@ -350,6 +364,8 @@ def test_default_bounds_parameters():
         (1e-6, 1e2),  # % ci
         (1e-6, 1e3),  # % cp
         (1e-6, 1e3),  # % ct
+        (1e-6, 5e-4),  # % alpha1
+        (1e-5, 1.0),  # % alpha2
         (1e-6, 1e3),  # % cl
         (1e-3, 20),  # % be
         (-50, 50),  # % kexc