406 - On the Systematic Underestimation of Frequent Floods in the U.S.

In the U.S., most hydrologists consistently perform flood frequency analyses (FFA) with annual maxima (AM) series, regardless of their study’s goals. AM series enjoy wider spatio-temporal availability, as compared to the continuous flow records necessary for the alternative FFA approach, based on peaks over a threshold (POT, also known as partial duration series). However, AM-FFA is intrinsically limited when it comes to predicting frequent floods (FFs - with return periods below 5 years), as it cannot account for events at the sub-annual scale. National guidelines in other countries (e.g., Australia and the U.K.) clearly recognize this situation, recommending POT-FFA for the correct prediction of low-return-period quantiles. Reliable estimates of FFs are required in a variety of both engineering and scientific river-related disciplines, such as channel restoration, stream ecology, and fluvial geomorphology. Our work sheds light on the fact that most current FF estimates in the U.S. are likely affected by a sizeable negative bias, because of the improper use of AM-FFA when predicting low-return-period peak flow events. The degree of underestimation is strongly dependent on the geographical location, reaching maxima of up to 50-60% in arid and semi-arid regions, for the 1.5-year flood. We investigate the drivers of the spatial variability in underestimation and propose a theoretical framework to correct biased FF estimates from AM-FFA. This allows practitioners to adjust AM-derived FF quantiles without the need to perform the more cumbersome POT-FFA. However, the correction approach is inaccurate in the case of basins that display mixed flood populations.