Abstract
Guidelines to estimate at-site design flood estimates (e.g., 100-year flood) have historically relied on annual maximum (AM) flow series. While widely used, the AM approach discards additional large floods occurring within the same year and may include annual maxima that are not true flood events. The peaks-over-threshold (POT) approach addresses these limitations by retaining all independent floods exceeding a threshold, increasing the sample size of observed flood events. Despite these theoretical advantages, previous studies have not found reduced uncertainty in POT-based design flood estimates. This study investigates underlying flood characteristics of AM and POT data from 482 basins under different hydroclimatic regimes across the coterminous United States. POT thresholds are systematically varied over their full feasible range to evaluate the tradeoff between sample size and assumption validity. Diagnostic analyses assess model assumptions, while Monte Carlo simulation and bootstrapping quantify uncertainty in 10- and 100-year flood estimates. Results show that the benefits of POT depend strongly on hydrologic regime. In humid basins, particularly eastern US rivers with generalized extreme value shape parameters between 0.1 and 0.3, POT samples satisfy model assumptions and reduce uncertainty in 100-year flood estimates relative to AM. In arid and semiarid basins, high thresholds exclude near-zero annual maxima that violate AM assumptions, improving estimate accuracy without substantially increasing uncertainty. In snowmelt-dominated basins, POT samples violate the assumptions before enough independent peaks accumulate, making AM generally preferable over POT. These findings reconcile theoretical and empirical evaluations of POT and provide regime-specific guidance for design flood estimation.