Deep learning-based building and road detection reveals higher permafrost thaw-related damage costs than previously estimated for Alaska

Abstract Arctic infrastructure is challenged by ice-rich permafrost thaw that causes differential ground subsidence. Economic impact estimates of permafrost thaw damages require accurate infrastructure inventories. We developed a deep learning-based mapping pipeline, HABITAT (High-resolution Arctic Built Infrastructure and Terrain Analysis Tool), to automatically detect infrastructure from Maxar satellite imagery in 285 Alaskan communities. Combining HABITAT with OpenStreetMap (OSM), we mapped a building footprint of 53M m 2 and a road network of 50, 477 km across Alaska. HABITAT adds 17M m 2 to the statewide building footprint not accounted for by OSM and 6M m 2 within discontinuous and continuous permafrost. We identified at-risk infrastructure on discontinuous and continuous permafrost with bearing capacity loss and ground subsidence between the decades 2015–2024 and 2055–2064 projected by a permafrost geotechnical model. Projected permafrost thaw damages to buildings and roads could cost Alaska 51. 4B. This is 21. 7 to 43. 2B more than estimates from previous literature due to the additional building footprint mapped by HABITAT. Several communities bearing the highest of this cost are identified in a government report as already experiencing the effects of permafrost thaw. Additionally, communities with high social vulnerability may experience higher economic losses that further challenge mitigation and disaster recovery efforts.