Abstract
Physically based distributed hydrological models have proven to be effective for simulating hydrological processes. Nevertheless, their applicability is limited by their intricate nature. Recently, deep learning-based models have gained significant attention for their ability to simulate high accuracy of hydrological fluxes. However, these models do not adhere to physical laws and remain difficult to comprehend. What if at the same time a data driven model that can achieve reliable accuracy, could also be clear about physical processes and provide interpretable outputs such as groundwater recharge, evapotranspiration, base flow and soil moisture? In this study, we propose a pipeline for optimizing hydrological model parameters by deep learning, exploiting the potential of earth observation data to train the model. The calibrated model will be used to formulate sub-seasonal drought predictions. The proposed framework will open new possibilities to exploit big data in hydrological modeling.