Modeling Celerity‐Discharge Behavior and Riverbank Storage for Improving Flood Simulations in Headwater Basins

Abstract Quantitative precipitation estimation (QPE) has been an enduring challenge especially in mountainous regions due to high spatiotemporal variability of precipitation. Because of QPE uncertainty, and fast rainfall‐runoff processes in complex terrain, improvements in flood modeling through calibration of hydrologic model parameters remain elusive. Liao and Barros (2022, https://doi.org/10.1016/j.rse.2022.113107 , 2023), https://doi.org/10.1029/2023wr034456 introduced the Inverse Rainfall Correction (IRC) to calculate QPE corrections. The IRC redistributes runoff simulation errors at the basin outlet by Lagrangian backtracking to the runoff source area along flow pathlines. Whereas the IRC can be implemented using any hydrologic model, the IRC outcome is impacted by hydrological model structure. This work aims to elucidate model structural uncertainty impacts on hydrologic simulations. Two sources of model structural uncertainty were identified: (a) numerical formulation—flood propagation errors attributed to the routing algorithm, and (b) missing physics—representation of riverbank storage impacts on early flood response. Significant advances are achieved by implementing a new flood routing algorithm without calibration of celerity‐discharge relations, and by introducing a riverbank storage parameterization to capture flood response delays tied to the lateral ponds along the streams. Overall, a median Kling‐Gupta Efficiency of 0.83 at 15‐min intervals is achieved. Over 95% of the events have flood timing errors less than 1 hour with the new routing compared to 30% with the classical variable‐parameter Muskingum‐Cunge (MC) routing, and the median error of peak discharge decreases from −7.4% to −1% after applying IRC. This study confirms the IRC as a robust general framework for QPE correction.