Training data from DYAMOND data set coarse-grained to R2B5

README concerning the training and test data sets obtained from coarse-graining the DYAMOND data set to R2B5 Date: 20/07/2026 T I T L E O F T H E D A T A S E T train_and_test_sets_cellbased: DYAMOND data set coarse grained to R2B5 cloud_regimes: subset of the test data from train_and_test_sets_cellbased divided into four different cloud regimes O W N E R / P R O D U C E R O F T H E D A T A S E T DYAMOND data set: Stevens et al, and Max-Planck-Institut for Meteorology, Hamburg Coarse-grained to R2B5, extraction of cloud regimes: Lorenzo Pastori and Arthur Grundner and Deutsches Zentrum fuer Luft- und Raumfahrt e. V., Institut für Physik der Atmosphäre, Oberpfaffenhofen D A T A U S A G E L I C E N S E See LICENSE File in this directory. C O N T E N T O F T H E D A T A S E T The archive contains cell-based training and test data from global high-resolution ICON simulations, the DYAMOND project, coarse grained to R2B5, 27levs. It is supplemented to the publication Pastori et al. DOI: 10.1088/3049-4753/ae4981, and the corresponding code repository DOI: 10.5281/zenodo.21504569 It is split into training and test data. Each data point corresponds to a grid cell. File format .npy, can be loaded with numpy.load() into numpy.array (For python and numpy version see requirements.txt file in corresponding repository). Folders: train_and_test_sets_cellbased: Training and testing data sets consisting of 40 simulated days each in summer 2016 (starting 1st of August 2016) and in winter 2020 (starting 20th of January 2020) cloud_regimes: contains a subset of the test data set separated in different cloud regimes, see /DYAMOND_cellbased_implementations/QNN_cloudcover/regression_cloudcover/8_features/cloud_regimes_eval Files: order_of_variable_files.npy includes the order of variables that were extracted * 'hus' (Specific humidity) * 'clw' (specific cloud water content) * 'cli' (cloud ice content ) * 'ta' (air temperatur [K]) * 'pa' (pressure [Pa]) * 'ua' (zonal wind component [ms^(-1)]) * 'va' (meridional wind component [ms^(-1)]) * 'zg' (geometric height at full level [m]) * 'coriolis' (coriololis parameter [s^(-1)], in (Pastori et al 2026) this is tranformed and used a variable for the latitude [rad]) * 'fr_land' (Fraction of land) * 'vert_layer' (which vertical layer) * 'clc' (cloud cover, fraction of cell occupied by clouds) D A T A U S A G E S C E N A R I O S - To train ML-based parameterizations M E T H O D S U S E D F O R D A T A C R E A T I O N See 'Deep Learning Based Cloud Cover Parameterization for ICON' (Grundner et al., 2022) for coarse-graining See "Data-Driven Equation Discovery of a Cloud Cover Parameterization" (Grundner et al. 2024) for coarse-graining and splitting into cloud regimes.