Check the processes of calculation!

[lovechang1986]

Hi, this is an awesome repo.

I had saw your codes of python and NCL, while I can't check the reason that why the result does not match eachother without compiled NCL.

On my opinine, even the results had some diff, they should not had quite bias. These were some process that may be generate bias:

1. month to season
2. standardized
3. normal EOF
   Your can ouput the results of NCL and compared with Python's.

Anyway, this is great, nice work. Thank for your time.

[raybellwaves]

I've checked the three things you mention.
python and NCL match with normal eof but the calculation drifts in the reof calculation.
The NCL function is fortran and pre-compiled. It's not easy to therefore print out variables in the code to understand it better. Not to mention it is horribly documented.

[ahuang11]

I wasn't able to reproduce CPC's loading patterns completely, but it's kind of close with Andrew's experimental rotation branch https://github.com/ajdawson/eofs/blob/experimental-rotation/lib/eofs/experimental/rotation/kernels.py

Note I had to disable kaisernorm to get non-NaNs and I didn't normalize by STD, only took the anomalies e.g.

da_hgt_anom = (da_hgt.groupby(da_hgt['time'].dt.month) - 
    da_hgt.groupby(da_hgt['time'].dt.month).mean('time'))

NAO attempt

NAO on CPC https://www.cpc.ncep.noaa.gov/products/precip/CWlink/pna/nao_loading.html

PNA attempt

PNA on CPC https://www.cpc.ncep.noaa.gov/products/precip/CWlink/pna/pna_loading.html

Also, interestingly, https://www.ncdc.noaa.gov/teleconnections/pna/ states it uses 0-90N, inconsistent with the 20-90N listed on https://www.cpc.ncep.noaa.gov/products/precip/CWlink/daily_ao_index/history/method.shtml