Safe Shelter: A Case for Prioritizing Housing Quality in Climate Adaptation Policy by Remotely Sensing Roof Tarps in the San Francisco Bay Area

Abstract

AbstractThe number and intensity of climate‐related hazards are increasing globally, but some communities are much more heavily impacted than others by the same event. One factor among many is housing quality because an intact house is a household's first line of defense against the wrath of nature. By taking the San Francisco Bay Area as a proof‐of‐concept, we develop a deep‐learning algorithm to show that blue roof tarps, used in the area as semipermanent fixtures to reduce roof leakage, can serve as a proxy for poor‐quality housing. Our cascading structure of Convolutional Neural Networks (CNNs) operates in a coarse‐to‐fine manner by first identifying buildings and then classifying them as containing a blue roof tarp or not, achieving a recall of 89.6% and a precision of 34.3% in our study area. Our work suggests that up to 5% of houses are of poor quality in some low‐to‐intermediate income communities in the San Francisco Bay Area. However, the percentage of poor‐quality houses varies by an order of magnitude even in communities with a comparable median income or social vulnerability index, suggesting that prioritizing climate adaptation investments based on these common indicators could be ineffective. Our work emphasizes the value of a community‐centric approach to improving housing quality and reducing the disproportional impacts of not only one but multiple climate hazards.