Evaluating Spatial Representativeness Across Multiple Scales for a Comprehensive Ground Validation Network Using Landsat Land Surface Temperature Data and Random Forest

Land surface temperature (LST) products require rigorous validation before widespread application, and the spatial representativeness of ground validation sites plays a critical role in ensuring the reliability of validation results. Therefore, accurately assessing the spatial representativeness of ground sites is essential for credible validation outcomes. However, existing studies often focus on a small number of sites within confined regional observation networks and generally evaluate representativeness at a single spatial scale. To address these limitations, this study conducts a comprehensive evaluation of 211 sites from five observation networks globally. To estimate representativeness across multiple spatial scales corresponding to typical LST products (i.e., 1, 3, 5, and 10 km), a novel spatial representativeness assessment model is proposed. This model, leveraging long-term Landsat LST data and the random forest (RF) method, quantifies the relationship between spatial representativeness error (SRE) and spatial scale, enabling seamless spatial representativeness evaluations for each site. Based on this framework, 24 sites that demonstrate consistently high representativeness across all scales and seasons are identified as optimal validation sites. Furthermore, this study proposes two site selection strategies: one prioritizing temporal stability, which identifies 44 sites ensuring representativeness across all seasons, and the other emphasizing spatial coverage, which selects 38 sites to guarantee representativeness at different scales. These findings provide valuable guidance and references for future LST product validation efforts.