The files here contain the source coude that has been used to calculate some of the rejection rates presented in the paper "An Ensemble-Based Statistical Methodology to Detect Differences in Weather and Climate Model Executables" from Christian Zeman and Christoph Schär in Geoscientific Model Development (https://doi.org/10.5194/gmd-2021-248).
mwu_gpu_cp_sp_diff.py: Calculate rejection rates using the Mann-Whitney U test on a grid-cell level for ensembles that have been produced with COSMO GPU DP, COSMO CPU DP, COSMO GPU SP, and COSMO GPU DP DIFF. This version runs in parallel on a specified amount of cores (nprocs).
mwu_ks_studt.py: Calculate rejection rates using different local statistical tests on a grid-cell level (Mann-Whitney U test, Kolmogorov-Smirnov test, Student's t-test) for the same cases as above.
mwu_update.py: Calculate rejection rates using the Mann-Whitney U test on a grid-cell level for ensembles that have been produced with COSMO before and after a major system update of the underlying supercomputer.
fdr_test.py: Calculate rejection rates using the Student's t-test where the approach in the paper using subsampling is compared to the FDR approach for determining field significance.
mannwhitneyu.cpp: C++ implementation of the Mann-Whitney U test on a grid-cell level.
kolmogorov-smirnov.cpp: C++ implementation of the Kolmogorov-Smirnov test on a grid-cell level.
The corresponding data is available in Zenodo.
First part: https://doi.org/10.5281/zenodo.6354200
Second part: https://doi.org/10.5281/zenodo.6355647
The latest release can be found here:
