Diagnostics and postprocessing of the National Hydrological Model product
Abstract
Aiming to provide unified national scale hydrological forecasts and water balance estimates, continental scale hydrological models are under development. Advancing these national scale models requires a standardized approach to model benchmarking, diagnostics and error attribution. HydroBench is an open-source suite of model benchmarking tools that fills this need. This study applies HydroBench to the National Hydrological Model (NHM) application of the Precipitation-Runoff Modeling System (PRMS) and performs error attribution using a Recurrent Neural Network (RNN). The benchmarking results show that the model performs better than an annual average streamflow estimate in close to 1,000 watersheds covering the conterminous US. Across these watersheds, performance deteriorates with an increase in infiltration excess runoff fraction and aridity index. In the error attribution, the RNN-based postprocessing primarily associated poor model performance to the inefficiency of the model products to extract the information content of the observed hydrometeorological data while suggesting that information contained in the catchment physical characteristics can improve model performance incrementally. Thus, the results demonstrate that strategies to improve the NHM-PRMS performance may start from the extraction of existing observed hydrometeorological and catchment physical characteristics information prior to incorporating new data.
- Publication:
-
AGU Fall Meeting Abstracts
- Pub Date:
- December 2022
- Bibcode:
- 2022AGUFM.H12M0847M