Residual error modelling for process-based flood frequency analysis

Flood frequency analysis (FFA) is essential for flood risk management and infrastructure design, yet the uncertainty associated with flood quantile estimates is often poorly characterized or disregarded—especially under data-scarce conditions. Existing uncertainty quantification methods are frequently subjective, overly complex, or impractical for routine engineering use. We introduce a simulation-based uncertainty quantification framework—UQ-flood—that integrates a process-based hydrologic model, a stochastic weather generator, and a residual error model (REM). Designed for annual maxima, the REM accounts for model bias and residual variability, enabling the generation of probabilistic streamflow ensembles tailored to extreme event analysis. We apply UQ-flood to three Canadian watersheds with long streamflow records and contrasting hydroclimatic conditions. We compare its performance against traditional statistical FFA using the Generalized Extreme Value distribution and Bayesian inference. UQ-flood yields flood quantile estimates consistent with long-record statistical methods but with substantially narrower uncertainty bounds. Under short-record conditions (e.g., 30 years), UQ-flood maintains statistically consistent estimates, while statistical FFA produces wide, often impractical uncertainty intervals. Additional experiments reveal that omitting the REM introduces systematic bias in flood magnitude estimates. UQ-flood avoids parametric assumptions about flow distributions, circumvents hydrologic model biases, and is adaptable to data-limited conditions. By explicitly propagating uncertainty from hydrologic simulation to flood quantiles, UQ-flood offers a practical alternative for robust flood risk management, including applications in infrastructure design and floodplain mapping. We recommended integrating residual error models into continuous simulation frameworks to improve bias correction and uncertainty quantification in flood risk estimation.