A Licenses and Terms of Use 716 ClimateLearn is a software package that can be installed from the Python Package Index as follows

UCLA is the maintainer of ClimateLearn . Table 4 shows the hyperparameters for ResNet in all of our experiments. Table 6: Default hyperparameters of ViT Hyperparameter Meaning V alue p Patch size 2 D Embedding dimension 128 Depth Number of ViT blocks 8 # heads Number of attention heads 4 MLP ratio Determine the hidden dimension of the MLP layer in a ViT block 4 Prediction depth Number of layers of the prediction head 2 Hidden dimension Hidden dimension of the prediction head 128 Drop path For stochastic depth [30] 0 . 1 Dropout Dropout rate 0 . 1 Table 7 summarizes the variables we use for our experiments. Constant represents constant variables, Single represents surface variables, and Atmospheric represents atmospheric properties at the chosen altitudes. Finally, during evaluation time, we use these masks to select subset of data.