Hybrid physically-based and deep learning modeling of a snow dominated mountainous karst watershed

Snow dominated mountainous karst watersheds are a primary water supply in many parts of the world. These watersheds are typically characterized by complex terrain, spatiotemporally varying snow accumulation and melt process, and complex flow and storage dynamics due to a high degree of hydrogeological heterogeneity. As a result, predicting streamflow from meteorological inputs has been challenging. Due to the inability of commonly used hydrologic models to represent these unique characteristics, we propose a hybrid modeling approach that integrates a physically-based snow model with a data-driven karst model. The high resolution snow model captures spatiotemporally varying snowmelt, which is then digested by a deep learning algorithm capable of handling both spatial and temporal connections between snowmelt and streamflow. The hybrid modeling approach is applied to a watershed in northern Utah with seasonal snow cover and variably karstified carbonate bedrock. Results suggest that the hybrid modeling approach simulates streamflow with higher accuracy than a benchmark hydrologic model for the study area.