Flood-frequency estimation for very low annual exceedance probabilities using historical, paleoflood, and regional information with consideration of nonstationarity

• DOI: 10.3133/sir20205065
• Publication Date: 2020-08-25T19:49:02Z
• Authors: Karen R. Ryberg; Kelsey A. Kolars; Julie E. Kiang; Meredith L. Carr

First posted August 25, 2020 For additional information, contact: Director, Dakota Water Science CenterU.S. Geological Survey821 East Interstate AvenueBismarck, ND 585031608 Mountain View RoadRapid City, SD 57702 Streamflow estimates for floods with an annual exceedance probability of 0.001 or lower are needed to accurately portray risks critical infrastructure, such as nuclear powerplants and large dams. However, extrapolating flood-frequency curves developed from at-site systematic streamflow records very low probabilities (less than 0.001) results in uncertainties the estimates. Traditionally, methods statistically estimating flood frequency have relied on record, which provides a time series maximum peaks, often including some historical peaks. most peak-flow less 100 years, when trying extrapolate magnitudes events.Other data may be available that extend record beyond dataset. Historical defined outside period but within human records. Examples information include other agencies newspaper accounts can translated magnitude point estimates, interval perception thresholds (such statement 1880 was largest since 1869). Paleoflood data, also dataset, broad range about occurrence sources like sediment deposits tree rings.Several assumptions made analysis, understanding whether conform these is desired. A particularly difficult assumption evaluate analysis underlying stationary—the peak flow varies around constant mean particular values (constant variance). As hydrologic community’s natural systems anthropogenic effects streamflows has evolved, community come understand many surface-water exhibit one more forms nonstationarity, thus stationarity violated degree. there currently (2020) no consensus among hydrologists regarding appropriate flood-frequency-analysis nonstationary systems, this topic remains active area research.A literature review completed summarize state science frequency. The highlights tools detect nonstationarities identifies approaches external inform analysis. To demonstrate initial incorporating paleoflood five sites were selected: Red River North at James Avenue Pumping Station, Winnipeg, Manitoba, Canada; reach, Rapid Creek, South Dakota; Spring Cherry Creek near Melvin, Colorado; Escalante Escalante, Utah. chosen availability published their geographic diversity unique characteristics, highlighted issues autocorrelation, change points, trends, outlier short periods record.An involved examining trends all sites. study used version 7.2 U.S. Survey PeakFQ program. Multiple analyses done each site documenting curve added. When studies available, compared here. comparisons cases simply show effect years whereas distributions fitting those PeakFQ.For North, shows appear necessary reasonably estimate probabilities. reach helped put high context; however, still had extraordinarily confidence bounds. These showed might transferred another, caveat case where we existing test transfer information—this not sites, transferring requires comparability affirmed result earlier generalized Pareto distribution good adding increase uncertainty probabilities, only, peaks much larger record.