Understanding fine-scale differences between satellite-derived and ground level temperature across urban landscapes - Dataset

Urban areas are emerging as significant hotspots of climatic risk, facing a combination of global temperatures rising, heat waves becoming more intense and severe, and urban heat islands already affecting citizens’ health. Nature-based solutions, such as Urban Forests (UFs), have demonstrated encouraging results in helping urban areas adapt to extreme heat challenges. However, most Urban Temperature studies are based on Land Surface Temperature detected from satellite sensors, which is an imperfect indicator of air temperature perceived by citizens. This study aims to evaluate how strongly Land Surface Temperature (LST) is related to air temperature (AirT), and to assess the strength of such relationship across a distance gradient from forested to built-up areas in an urban setting. We took spatially distributed measurements of AirT over 15 months (June 2023 to September 2024) at a height of 3 meters by deploying 169 AirT sensors across nine urban parks in the city of Milan, Italy. We compared each sensor measurements against LST recorded by Landsat for the same pixel and date and fitted regression models using the distance from forested areas as a co-predictor. We found that LST is not always a reliable predictor of AirT, but its error is highly predictable and dynamic. The bias reverses from an underestimation of AirT under cool conditions to a significant overestimation under hot conditions. The magnitude of this bias intensified with increasing LST and was strongly mediated by local microstructure and vegetation index (NDVI). LST displayed a systematically smaller bias relative to AirT when evaluated against sensors on artificial infrastructure compared to trees, reducing underestimation in cool conditions but intensifying overestimation during heat events.