A supplementary code for the paper.
Data used for the experiments can be found here. It includes NCEP-NSCv2 and NASA-GMAO ensemble members, the ground truth data -- precipitation and 2 meter temperature, observational data: slp, rhum, hgt500; and principal components of SSTs. There are also files with the US mask, an observational climatology, a model climatology and the 33rd and 66th percentile values.
All data was preprocessed: the ensemble members, the ground truth and the climate data are on the same grid. There are two directories:
- train_val -- data for training and validation.
- test -- data for test.
Usually, all variables have the following shape (t, 64, 128), where t = 312 for train and validation, t = 117 for NCEP data and t = 85 for NASA data for test. File utils_data.txt
includes functions to work with the data.
See requirements.txt
file.
- Python 3.3+
- smp package
There are a few Jupyter notebooks to demonstrate how to work with data and train/evaluate models: