Reclamation Information Sharing Environment (RISE)

Reclamation partnered with scientists at NCAR to assess sensitivity and sources of uncertainty in rainfall runoff modeling to support flood frequency analyses. Pieces of the modeling chain, including initial conditions, model parameters, precipitation inputs, and model structure were examined across a range of return intervals from 2-100,000 years. Two example watersheds representing different hydrologies in the 17 western states were used for this study. A stochastic hydrologic modeling workflow was developed using NCAR’s FUSE modeling framework. Results indicated that modeling chain pieces have variable uncertainty contributions across return periods. Results found that precipitation inputs are most important for rare events while initial conditions are important for frequent events; however, uncertainties from model structure and structure-parameter interactions still play a proportionally important role in results. This highlights the importance of understanding flood generation processes and selecting appropriate models based on that understanding. In addition to these findings, a review of calibration metrics indicated that the KGE is a more robust metric than NSE for calibration of models for extreme events.