Nonstationarity in Extreme Precipitation Return Values along the U.S. Gulf and Southeastern Coasts

Abstract

Abstract This study estimates extreme rainfall trends across the Gulf Coast and southeastern coast of the United States while applying methods for extending the temporal record and aggregating across spatial trend variations. Nonstationary generalized extreme value (GEV) models are applied to historical annual daily maximum precipitation data (1890–2019) while using CMIP5 global mean surface temperature (GMST) as the covariate. County composites and multicounty regions are used for local data record extension and pooling. Unlike most previous studies, return periods as long as 100 years are analyzed. The local trend estimates themselves are found to be too noisy to be reliable as estimates of climate-driven trends. However, application of a Gaussian process model to the spatial distribution of observed trends yields overall trend detection at the 95% significance level. The overall historical increase due to nonstationarity across the study region, with associated 95% confidence intervals, is 9% (3%, 15%) for the 2-yr return period and 16% (4%, 26%) for the 100-yr return period. A trend is also detectable in the Gulf Coast subregion, but not in the smaller southeast subregion. Recent weather events and nonstationarity have caused the official return value estimates for parts of North and South Carolina to be much lower than the return values estimated here. Significance Statement Protection of people and infrastructure from flooding relies on accurate estimates of potential extreme rainfall intensity. Some official estimates of extreme rainfall near the Gulf Coast and southeastern coast of the United States are over 20 years old. We show that, across this region, there is a clear trend in daily rainfall so extreme that it only has a 1% chance of happening in any given year (the so-called 100-yr rainfall). This trend means that many existing estimates of extreme rainfall are too low, both now and in the future, so flooding risks based on those estimates would be underestimated as well.