This repository contains the code used in the article "Reduced floating-point precision in regional climate simulations: An ensemble-based statistical verification" (Submitted).
The core of the testing methodology consists in multiple rounds of subsampling, grid-point-level testing using the Kolmogorov-Smirnov test, and averaging.
The most computationally expensive parts are coded using CuPy to work on GPUs, providing a major speedup.
The definitions of the core functions including the statistical tests are found in scripts/utils.py, while other python scripts there are meant to be run with slurm on a HPC cluster to perform the tests (one_ks.py) or other steps of the methodology (concat.py, decisions.py, FDR.py).
All the plots in the article and other side-analyses are done in the main jupyter notebook cosmo-sp.ipynb.
Input parameters for the COSMO runs presented in the article can be found in the subfolder COSMO_INPUTS.