This release incorporates the forwarding code of the high-resolution, differentiable hydrologic model, 𝛿HBV2.0UH, from Song, Yalan, Tadd Bindas, Chaopeng Shen, Haoyu Ji, Wouter Johannes Maria Knoben, Leo Lonzarich, Martyn P. Clark et al. "High-resolution national-scale water modeling is enhanced by multiscale differentiable physics-informed machine learning." Authorea Preprints (2024). https://essopenarchive.org/doi/full/10.22541/essoar.172736277.74497104. This paper is under review at Water Resources Research.
This code is built in a domain-agnostic, PyTorch-based framework for developing trainable differentiable models that merge neural networks with process-based equations. Following as a generalization of HydroDL, 𝛿MG (generic_deltaModel) aims to expand differentiable modeling and learning capabilities to a wide variety of domains where prior equations can bring in benefits. This package is maintained by the MHPI group advised by Dr. Chaopeng Shen.
The 𝛿MG package includes the lumped differentiable rainfall-runoff models, 𝛿HBV1.0, improved 𝛿HBV1.1p, and implicit adjoint-based 𝛿HBV.adj, and the high-resolution, differentiable hydrologic model, 𝛿HBV2.0. This package powers the global- and national-scale water model that provide high-quality seamless hydrologic simulations across US and the world. It also hosts global-scale ecosystem learning and simulations. Many other use cases are being developed concurrently.