A Physically Informed Generalized Least Squares Fusion Framework for ECOSTRESS LST Validation Over Heterogeneous Landscapes

Read full text Summarise with VIRAT
Authors: Yuting Qi, Lei Zhong, Yaoming Ma, Zixin Wang, Jie Liu
Cited by: 0
References: 50

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

Validating new-generation Land Surface Temperature (LST) products, such as ECOSTRESS, is constrained by sparse in-situ measurements that are insufficient to capture diverse atmospheric and surface heterogeneity. While Radiance-based methods offer extensive spatial coverage complementary to traditional Temperature-based approaches, establishing a unified validation benchmark remains challenging due to the lack of a rigorous framework to reconcile their inconsistent error structures. To address this limitation, a novel validation framework based on a Physically Informed Generalized Least Squares (PI-GLS) approach is proposed. Specifically, an error covariance matrix is analytically formulated based on radiative transfer theory to explicitly quantify the uncertainties inherent in each validation source and emissivity-induced correlations. By integrating this physical error structure into the GLS estimator, a statistically optimal validation metric is generated, enabling a unified and comprehensive accuracy assessment of ECOSTRESS LST products. Results reveal substantial performance variability across environmental gradients, with mean RMSE ranging from 0.70 K to 4.48 K. This variability underscores the sensitivity of validation metrics to sampling bias; spatially sparse sampling tends to bias error assessments, thereby necessitating the broadcoverage approach proposed herein. Applying the PI-GLS framework yields a validation RMSE of 2.33 K for ECOSTRESS. This metric provides a spatially representative assessment of sensor performance, effectively correcting the systematic underestimation observed in traditional equal-weighting (1.78 K) or weighted fusion (1.58 K) schemes that lack rigorous treatment of error correlations and sampling heterogeneity. Crucially, expanding validation coverage to previously under-sampled environments exposes systematic biases masked in limited local validations. Notably, diagnostic analysis points to inaccuracies in the atmospheric profiles used by ECOSTRESS as a likely source of the emissivity anomalies observed over the Tibetan Plateau. Beyond ECOSTRESS, this study establishes a scalable methodology to overcome in-situ data scarcity, providing a rigorous validation standard for future global TIR missions.

Fields of Study & Subjects

Calibration and Measurement Techniques Soil Moisture and Remote Sensing Remote Sensing in Agriculture

Publication details

Journal / source
IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
Volume
19
Pages
11713–11729
Publication date
1 January 2026
Language
EN