183 - Coupling Machine Learning with the National Water Model to Account for the Impacts of Water Resources Infrastructure: Improving Streamflow Prediction Skill to Enhance Great Salt Lake Management.

A rigorous evaluation of the National Water Model (NWM) within the Great Salt Lake (GSL) basin using the Research-Oriented Streamflow Evaluation Tool indicated satisfactory prediction skill in headwater catchments but decreased as the three main tributaries meet the GSL, challenging the use of the NWM for GSL environmental management. Since the NWM does not include water resources infrastructure within its codebase (e.g., reservoir, reservoir operations, interbasin transfers, irrigation diversions, water treatment and wastewater treatment plants, etc.), we hypothesized the extensive infrastructure within the basin compromises NWM streamflow estimates. Due to the model architecture of the NWM, the scale of the GSL watershed, and water resource security implications, it is not feasible to explicitly parameterize the operations of each infrastructure component. Addressing the need to improve streamflow prediction skill without explicit infrastructure parameterization, we develop and couple machine learning (ML) algorithms with NWM streamflow estimates to account for the influences of water resources infrastructure. We train the models using the long-term record of nearly 50 USGS monitoring locations distributed throughout the watershed with colocated NHDPlus reach predictions from the NWM v2.1 retrospective and conduct feature engineering and feature selection of catchment characteristics including size, upstream reservoir storage, snow levels, precipitation, temperature, land cover, etc. Preliminary results indicate up to a 50% reduction in mean percentage error, a 20% reduction in percent bias, and a 33% improvement in Kling-Glutz efficiency. We will be coupling the validated NWM-ML into the GSL Integrated Model framework to enhance GSL-level modeling for informed resource management.