Assessing the Spatial Variability of Radar-Rainfall Uncertainty in Streamflow Prediction

Quantitative precipitation estimation (QPE) is critical for testing and operating flood forecasting systems. QPE represents one of the main hydrological models forcings, and its uncertainty may decrease the forecast performance. We have limited information on how QPE uncertainty spatially propagates through hydrological models, possibly biasing their performance. Here, we present a systematic approach to assess the impact of QPE uncertainty in streamflow forecasting, considering its spatial consistency. We ran the Hillslope Link Model (HLM) for Iowa between 2015 and 2020, using the Multi-Radar/Multi-Sensor System (MRMS) and two IFC radar-derived products, i.e., reflectivity-based IFCZ and specific attenuation-based IFCA. We perturb the products with a multiplicative error term. We assessed the model performance at 80 watersheds with nested USGS observations. In each gauge, we computed the KGE performance of the error terms using a moving window with the size of the watershed concentration time. From the moving-window KGE, we derived a time series of the best-performing error term at each time step. Finally, we compared the error term series for the watersheds that belong to the same nested hierarchical system. According to our results, the multiplicative error changes within the same hierarchical system under different conditions. Factors such as the percentage of shared area and the distance influence the correlation between nested watersheds. Our results suggest that QPE uncertainties spatial propagation within hydrological models is complex, and we require further analysis to understand it. The results are a step forward in a direction of separating the model from the input uncertainty.