Flood nonstationarity across the US: Detection, attribution, and adjustment

As a statistical method, flood-frequency analysis has fundamental underlying assumptions, including an assumption that floods are generated by stationary processes (constant mean within a window of variance). As our understanding of nature, our effect on nature, and statistical principles has improved, standard flood-frequency analysis methods have become increasingly questionable for some sites or time periods. Yet, flood-frequency analysis remains critical for the appropriate sizing and construction of culverts, bridges, and other flood-control infrastructure and for informing decisions related to the safety of homes and businesses and to ecosystem management. Our goals, to date, for a multi-year project funded by the U.S. Federal Highway Administration have been to document trends and change points (nonstationarities that are violations of the assumptions of flood-frequency analysis) in annual peak streamflow across the conterminous United States and attribute these changes, where possible, to anthropogenic and environmental factors for which there are data. Once the anthropogenic or environmental changes causing these nonstationarities are better understood, analysts can then begin to make choices about the best methods for adjusting flood-frequency analyses. Our current goal is to take what we have learned about nonstationarities and their attributions and test potential methods for adjusting flood-frequency estimates. This poster demonstrates a framework for synchronizing efforts to detect, attribute, and adjust for changes in flood regimes, utilizing knowledge from experts in hydrologically diverse regions in the conterminous United States.