Item

Abstract

Distinct atmospheric processes are identified during extreme precipitation (PEx) events over the contiguous United States (CONUS). We train self organizing maps (SOMs) on a pressure-time series of vertical velocity obtained separately from each of the advective forcing terms in the Quasi-geostrophic (QG) omega equation for each grid point experiencing an extreme event. The unsupervised learning of the SOM analysis identifies patterns in vertical velocity associated with precipitation extremes. The SOM patterns include: multiple frontal cyclone weather patterns while grouping primarily convective events into one pattern. Frontal cyclone events include a synoptic pattern consistent with west coast atmospheric river events as well as groups of patterns linked to developing and to mature (“occluded”) frontal cyclones. Extreme events are examined at individual grid points and also aggregated for seven regions containing consistent annual cycles of precipitation. The most common patterns vary seasonally and geographically. During summer: convection is the most common in all regional averages, though frontal cyclones more commonly cause PEx in the northern parts of the Pacific Northwest, Great Plains, and Northeast. During winter, the most common patterns are: atmospheric rivers in the Pacific Northwest, “occlusions” in the Great Plains and parts of the Rockies, convection over the Florida peninsula, and developing frontal cyclones everywhere else. This machine learning approach allows large volumes of climate model data to be quickly processed, making viable a climate model assessment tool based on how well the model captures the mix of processes creating extreme precipitation.