尽管文中重点分析了这些城市，但该工作流程原则上可适用于任何城市区域，这体现了地理空间基础模型的核心优势。该方法基于四种主要数据源：（1）经协调的Landsat-Sentinel-2影像，提供30米分辨率多光谱波段，用于通过分裂窗算法反演地表温度（LST）[21, 25]；（2）Impact Observatory提供的10米分辨率土地利用与土地覆盖（LULC）数据，用于识别城市区域与绿地空间[26]；（3）ERA5-Land再分析数据，提供连续的大气背景信息，其中近地表气温数据与HLS影像进行叠加。

While these cities are highlighted, the workflow can in principle be applied to any urban area, which reflects the core advantage of geospatial foundation models. The methodology is based on four primary data sources. (1) Harmonised Landsat Sentinel-2 imagery providing 30 m multispectral bands to derive LST using a split-window algorithm [21, 25], (2) Impact Observatory 10 m LULC data for identifying urban and green spaces [26], (3) ERA5-Land reanalysis providing continuous atmospheric context, with near-surface air temperature stacked onto the HLS image