In the reservoir models, we have options to mask the reservoir inputs using a specified mask (used for SST runs, currently) but the hybrid inputs had no such option. This PR takes that same mask and also applies it to the readout training by multiplication with the post-encoded spatial gridpoints. It is then saved as a field with the hybrid model which is loaded and applied to any hybrid input data for prediction. This allows us to explore how much of the model behavior is tied to continental atmospheric information vs. the ocean areas where the model should actually experience the coupling.
In the reservoir models, we have options to mask the reservoir inputs using a specified mask (used for SST runs, currently) but the hybrid inputs had no such option. This PR takes that same mask and also applies it to the readout training by multiplication with the post-encoded spatial gridpoints. It is then saved as a field with the hybrid model which is loaded and applied to any hybrid input data for prediction. This allows us to explore how much of the model behavior is tied to continental atmospheric information vs. the ocean areas where the model should actually experience the coupling.
Coverage reports (updated automatically):