Abstract
Surface albedo is a key determinant of urban heat islands, which modulates the amount of solar energy absorbed or reflected by urban surfaces, influencing microclimate and thermal comfort. However, high-resolution albedo is usually not available, which makes the understanding of the urban thermal environment at hyperlocal difficult. This study presents the first high-resolution urban albedo maps for 34 major U.S. cities using advanced deep learning models and multisource remote sensing data. By differentiating between impervious and pervious surfaces using a combination of NAIP imagery, roof albedo data, building footprints, land cover classifications, and Sentinel-2 imagery, this work achieves sub-meter resolution in albedo mapping. Employing U-Net for impervious surface classification along with impervious (ISA) and pervious surface albedo (PSA) prediction, these models were validated in selected cities, with ISA showing an R(2) of 0.9028 and MAE of 0.0057, and PSA demonstrating an R(2) of 0.9538 and MAE of 0.0027, highlighting the precision and reliability. The datasets, made publicly available, offer essential insights for urban planning and environmental monitoring.