Refined Modeling of Arctic Circumpolar Building Stock Increases Estimated Mid‐Century Permafrost Degradation Damages

Abstract Near‐surface permafrost degradation has the potential to damage infrastructure needed to sustain Arctic communities, but risk assessments remain highly uncertain due to a limited representation of the Arctic circumpolar building stock. With deep learning building footprint detection and building occupancy classification models, we expand the two‐dimensional building footprint area and disaggregate the building stock based on occupancy type. Based on story counts derived from a circumpolar digital surface model, we estimate the total floor area of the residential building stock to be 75% greater than the two‐dimensional footprint area. Monte Carlo‐estimated mid‐century circumpolar building damages amount to 76 B USD (2.88–259.48) under SSP2‐4.5 and 261 B USD (17.01–379.11) under SSP5‐8.5 (median and 5th–95th percentile range), with uncertainty primarily driven by climatic variability, engineering practices, and deep learning model errors. By refining permafrost‐affected building damage estimates, this study supports the development of data‐driven decision making in the Arctic.