A Southeastern United States Warm Season Precipitation Climatology Using Unsupervised Learning

Abstract

Agriculture in the southeastern United States (SEUS) is heavily reliant upon water resources provided by precipitation during the warm season (June–August). The convective and stochastic nature of SEUS warm season precipitation introduces challenges in terms of water availability in the region by creating localized maxima and minima. Clearly, a detailed and updated warm season precipitation climatology for the SEUS is important for end users reliant on these water resources. As such, a nonlinear unsupervised learning method (kernel principal component analysis blended with cluster analysis) was used to develop a NARR-derived SEUS warm season precipitation climatology. Three clusters resulted from the analysis, all of which strongly resembled the mean spatially (r > 0.9) but had widely variable precipitation magnitude, as one cluster denoted a mean pattern, one a dry pattern, and one a wet pattern. The clusters were related back to major SEUS warm season precipitation moderators (tropical cyclone landfall and the El Niño–southern oscillation (ENSO)) and revealed a clearer ENSO relationship when discriminating among the cluster patterns. Ultimately, these updated SEUS precipitation patterns can help end users identify areas of notable sensitivity to different climate phenomena, helping to optimize the economic use of these critical water resources.