Estimates of Mobile and Manufactured Home Tornado Exposure using Machine Learning

Tornado deaths occur at disproportionately high rates in mobile and manufactured homes (MMH) compared to other housing types. This increased vulnerability is owed to myriad factors that exacerbate risk and the probability of death or injury for MMH residents during tornadoes. While addressing physical vulnerability to tornadoes is an arduous task, socioeconomic vulnerabilities can be more readily reduced through improved education, forecasting, communication, and other considerations. This work examines automatically identified MMH locations and historical tornado swaths to improve our understanding of MMH exposure to tornadoes. Specifically, we examine two U.S. counties that have contrasting landscapes but share a relatively high tornado frequency and density of MMH: 1) Ford County, KS; and 2) Walker County, AL. Precise MMH locations below the census block level are required because tornadoes have characteristic widths of 1000 yards or less. To address this issue, the locations of MMH are manually identified in a 60-cm resolution aerial image from the National Agriculture Imagery Program (NAIP) over Walker County, AL. These locations are used to generate a machine learning model based on the U-Net architecture to identify contiguous pixel groups associated with MMH. The labeled regions are organized into training and validation datasets to generate and select models with varying hyperparameters, and a testing dataset is used to assess generalization performance using the Intersection over Union (IoU) metric and manual assessments. Finally, the identified locations in Walker County, as well as the predicted locations in Ford County, are cross-referenced with tornado swath data to determine statistics of MMH exposure during select tornado events. These methods could be expanded to other regions and used as operational tools to anticipate forecasted and ongoing exposure of highly vulnerable citizens to hazards associated with tornadoes and other weather hazards.