Flood frequency sampling error: insights from regional analysis, stochastic storm transposition, and physics-based modeling

• Published in: Journal of hydrology (Amsterdam) Vol. 662; p. 133802
• Main Authors: Abbasian, Mohammad, Wright, Daniel B., Notaro, Michael, Vavrus, Steve, Vimont, Daniel J.
• Format: Journal Article
• Language: English
• Published: Elsevier B.V 01.12.2025
• Subjects: Flood frequency analysis; Model calibration; Regional analysis; Sampling error; Stochastic storm transposition; WRF-Hydro; Sampling error; WRF-Hydro; Flood frequency analysis; Stochastic storm transposition; Model calibration; Regional analysis
• ISSN: 0022-1694
• DOI: 10.1016/j.jhydrol.2025.133802

•Flood sampling errors can arise from factors beyond small sample sizes.•Even relatively long flood records can be subject to large at-site FFA sampling errors.•Regionally rare and extreme flood events introduce uncertainty to at-site FFA.•Regional and at-site FFA together give better understanding of flood extremes and frequency.•Stochastic Storm Transposition-based FFA can complement statistical FFA. Flood Frequency Analysis (FFA) typically involves fitting a probability distribution to Annual Maximum Peaks (AMPs) to estimate peak flow quantiles. While flood frequency sampling error due to small sample sizes is a well-recognized issue, this study highlights that, as flood data records lengthen, sampling error can remain as a concern. We examine the Kickapoo watershed in the north-central United States, where at-site FFA is susceptible to significant sampling errors despite a relatively long record of flood observations (90 years). We leverage a combination of at-site and regional statistical analyses of AMPs and watershed characteristics, Bulletin 17C guidelines, and a process-driven FFA approach to gain insights into flood extremes, flood frequency, and sampling errors. The process-based FFA integrates stochastic storm transposition with Monte Carlo physics-based hydrologic modeling. It employs the WRF-Hydro hydrologic model and a process-based calibration approach with Fusion, a new high-resolution forcing dataset over the continental United States. We demonstrate that three exceptionally large events, with a combined likelihood of occurrence of less than 2%, significantly affect extreme quantiles and their confidence intervals. Further, we show that while Kickapoo’s physiographic characteristics differ somewhat from neighboring watersheds, these extremes are also shaped by remarkable precipitation variability, contributing to sampling errors. While sample size has traditionally been the focal issue of sampling error, this research underscores additional factors: rare and extreme events and hydroclimate variability. It also highlights how regional analyses and advanced physics-based modeling techniques help improve understanding of flood extremes and frequency.