Bias in Peak Flood Discharges: Are Our Bridges and Culverts Undersized?

Abstract

Reliable methods for peak discharge predictions at ungaged locations are required for infrastructure design and floodplain management. For decades, a standard practice in the United States has been to utilize US Geological Survey regional regression equations (StreamStats) as a singular method. However, implementation of multiple methods, such as streamgage-based transfers, rainfall-runoff modeling, and downscaled climate modeling, provides a more rigorous approach. The Flood Potential Portal (https://floodpotential.erams.com/) Watershed Analysis module provides three streamgage-based methods for predictions, facilitating comparative evaluations of prediction bias across the United States. Our findings indicate that regional regression equations, overall, significantly underpredict 100-year (1% chance of exceedance; Q 100) peak discharges by −26.6%. In contrast, index flood frequency (Q 100) and expected flood potential discharge (Q efp) show minimal bias of +4.4% and 0%, respectively. Within the 18 analyzed watersheds, regional regression equations underpredict in 15 (−55.1% to −1.8%, average: −27.8%), predict best in two, and overpredict in one. The other methods yield far less relative bias. Underprediction of −55.1% results in road-stream crossings having less than half the needed capacity, and an average bias of −27.8% indicates that crossings may be commonly undersized by more than a quarter—use of only regional regression equations for peak discharge predictions for bridges and culverts may result in infrastructure having insufficient flood resilience. A case study for a watershed impacted by the remnants of Hurricane Helene reveals underprediction of up to −53.2%; peak discharge bias could result in undersized structures. Resilient recovery requires the use of multiple predictive methods.