This package provides implementation of several statistical climate downscaling techniques, as well as evaluation tools for downscaling outputs.
See environment.yml. Tensorflow is pinned in this conda environment in the interest of reproducibility.
With conda:
git clone https://github.com/drewpolasky/CCdownscaling
cd CCdownscaling
conda env create -f environment.yml -n ccdown_env
conda activate ccdown_env
export PYTHONPATH=$PWD:$PYTHONPATHAn example use case for downscaling precipitation at Chicago O'Hare airport can be found in the example folder. This example requires some example data, which can be downloaded from: https://zenodo.org/records/7817799
Once that data is in place, the example can be run with:
cd example
python ohare_example.pyAnd runs through several downscaling methods, including SOM, random forest, and quantile mapping. All these methods are then compared on PDF skill score, KS test, RMSE, bias, and autocorrelation, along with the undownscaled values from the NCEP reanalysis.
There are several command line settings that can be adjusted: target variable (max_temp or precip), stationID, and split_type (simple, percentile, seasonal):
python ohare_example.pyThere is also a jupyter notebook with the same example included in the example folder.
If you use CCdownscaling, please cite:
Andrew D. Polasky, Jenni L. Evans, Jose D. Fuentes, CCdownscaling: A Python package for multivariable statistical climate model downscaling, Environmental Modelling & Software, Volume 165, 2023, 105712, ISSN 1364-8152, https://doi.org/10.1016/j.envsoft.2023.105712.