An Integrated Framework for Validating ECOSTRESS LST across Diverse Surface and Atmospheric Conditions: Fusing Radiance- and Temperature-Based Approaches to Overcome In-Situ Data Scarcity

High-resolution land surface temperature (LST) is a critical variable for quantifying fine-scale energy and water cycle variations. NASA’s ECOSTRESS mission provides unprecedented high-resolution thermal infrared observations for investigating intricate land-atmosphere interactions. Nevertheless, previous LST validation efforts have been constrained by sparse ground networks with inadequate sample sizes, limited diversity in land cover types and atmospheric conditions, and insufficient high-elevation coverage. To overcome this long-standing challenge, a comprehensive validation framework was established by integrating a feasible radiance-based (R-based) validation method with conventional temperature-based (T-based) validation. The R-based method was first verified at a homogeneous site (BJ station, Mean Bias = 0.21 K) and subsequently extended to evaluate seven previously unvalidated surface types, addressing critical data gaps across challenging surfaces like glaciers and permafrost. A comprehensive uncertainty budget was then systematically quantified for both methodologies, revealing distinct error components and inherent differences between the two approaches. To reconcile these differences and establish a more representative and robust validation reference, results from both approaches were integrated using an Uncertainty-Weighted Averaging (UWA) framework. This integrated framework yielded an overall UWA-based RMSE of 2.12 K for the ECOSTRESS LST product. Notably, retrieval accuracy was significantly degraded over surfaces characterized by high spatiotemporal variability, including alpine meadows, urban environments, and shrubland ecosystems. Furthermore, atmospheric conditions over the Tibetan Plateau (TP) were found to be systematically misclassified by the ECOSTRESS processing chain compared to low-elevation regions, leading to significant emissivity-estimation anomalies. Under these challenging conditions, a split-window algorithm demonstrated superior performance (UWA-based RMSE: 1.71 K) when accurate emissivity information was available. Therefore, rigorous quality screening and consideration of alternative retrieval algorithms are recommended for the current ECOSTRESS LST product over the TP for applications requiring high precision. Collectively, the integrated framework established in this study provides the essential methodology to overcome in-situ data scarcity and enables, for the first time, a comprehensive and systematic validation of the new-generation thermal sensor across diverse surface and atmospheric conditions.