High-resolution simulation and modeling of tropical-cyclone induced power outages in the United States

Abstract

Storm driven power outages impose large and uneven costs on communities, yet operational outage prediction models remain constrained by available datasets. Existing approaches either provide high resolution forecasts for a small number of utilities with internal outage records, or operate across broad regions with insufficient resolution for local planning. We address this gap by first assembling a new sub-kilometer scale multi-utility outage dataset that combines scraped outage records across nine states and ten utility service territories with county-level customers affected data. Based on this new dataset we then develop a topologically informed statistical framework to estimate outage rates and the number of customers affected based on gridded wind, precipitation, and storm surge fields, distribution line exposure, undergrounding, and social vulnerability characteristics. On held-out storms, the framework outperformed existing outage models for the contiguous United States (CONUS). Using the trained model, we reconstruct a decade of tropical cyclone driven outages for 53 events on a uniform high resolution sub-kilometer scale grid. Analyses of this dataset reveal that counties with higher social vulnerability suffer systematically higher outage rates compared to less vulnerable counties exposed to similar wind speeds, partially because counties with the highest levels of social vulnerability are most likely to have high levels of overhead exposure compared to the least vulnerable communities.