Evaluation of Hydrologic Predictions Based on Multi-Model and Multi-Precipitation Product Forcing

This study evaluates hydrologic predictions made using multiple distributed hydrologic models driven by multiple high-resolution precipitation forcing products. We use a mix-and-match approach to assess prediction skills affected by the sources of uncertainty due to different combinations of the precipitation forcing products and rainfall-runoff distributed model components. Examples include comparing two different routing schemes keeping everything else exactly the same, or using two different input products for exactly the same hydrologic model configuration. Assessment of the differences based on comparisons with streamflow observations along with the river network-wide examination of water transport dynamics leads to better understanding of the overall prediction system requirements for skillful prediction. We use a variety of forcing products including MRMS, Iowa Flood Center (IFC), and NWS polarimetric QPE with/without HRRR QPF to drive the National Water Model (NWM) and the IFC Hillslope-Link Model (HLM). We perform a multi-year evaluation using statewide stream gauges including about 150 USGS streamflow gauges and 240 IFC stage sensors. We consider all possible combinations of the modeling elements (forcing products and model routing components) to evaluate prediction capability affected by each one of them. The analysis framework investigates differences in multiple aspects of streamflow (e.g., peak flow) among different combinations and the behavior of these differences with changes in the upstream drainage network. We discuss forecast errors that are represented as a function of basin scale, forcing product resolution and uncertainty, and hydrologic model structure and components.