Flood frequency analysis using radar rainfall fields and stochastic storm transposition

Abstract

Flooding is the product of complex interactions among spatially and temporally varying rainfall, heterogeneous land surface properties, and drainage network structure. Conventional approaches to flood frequency analysis rely on assumptions regarding these interactions across a range of scales. The impacts of these assumptions on flood risk estimates are poorly understood. In this study, we present an alternative flood frequency analysis framework based on stochastic storm transposition (SST). We use SST to synthesize long records of rainfall over the Charlotte, North Carolina, USA metropolitan area by “reshuffling” radar rainfall fields, within a probabilistic framework, from a 10 year (2001–2010) high-resolution (15 min, 1 km2) radar data set. We use these resampled fields to drive a physics-based distributed hydrologic model for a heavily urbanized watershed in Charlotte. The approach makes it possible to estimate discharge return periods for all points along the drainage network without the assumptions regarding rainfall structure and its interactions with watershed features that are required using conventional methods. We develop discharge estimates for return periods from 10 to 1000 years for a range of watershed scales up to 110 km2. SST reveals that flood risk in the larger subwatersheds is dominated by tropical storms, while organized thunderstorm systems dominate flood risk in the smaller subwatersheds. We contrast these analyses with examples of potential problems that can arise from conventional frequency analysis approaches. SST provides an approach for examining the spatial extent of flooding and for incorporating nonstationarities in rainfall or land use into flood risk estimates.