A universal yet easy-to-use data-driven method for angular normalization of directional land surface temperatures acquired from polar orbiters across global cities

Urban thermal anisotropy poses significant challenges for accurately retrieving land surface temperature (LST) in urban environments using wide-swath polar orbiters. Existing physical and kernel-driven models often require detailed urban structural and property information or rely on simultaneous multi-angle LST observations, limiting their applicability for normalizing directional LSTs across diverse urban settings worldwide. Here we propose a UNIversal, easy-To-usE Data-driven (UNITED) method for angular normalization of directional LSTs across global cities, integrating advanced machine learning techniques with multi-source remote sensing and reanalysis data. We applied this method to normalize directional urban LSTs from all available wide-swath polar orbiters (Aqua MODIS, Terra MODIS, Suomi-NPP VIIRS) on Google Earth Engine, leveraging their full archives of multi-angle observations (2003–2024 for MODIS and 2012–2024 for VIIRS). The method's high accuracy in normalizing these three products was rigorously validated using quasi-simultaneous, near-nadir LSTs from various satellite platforms (e.g., Landsat) across tens of millions of urban pixels worldwide under diverse spatial, temporal, and angular conditions. For example, for Aqua MODIS observations with viewing zenith angle exceeding ±55°, angular normalization reduces the root mean square error and bias relative to nadir VIIRS LSTs (used as the reference) from 5.71 °C and −4.92 °C to 2.43 °C and −0.40 °C, respectively, underscoring the effectiveness of the UNITED method in harmonizing directional urban LSTs. Our study holds significant implications for advancing urban thermal remote sensing.