A comparative analysis of Split Window Algorithms to retrieve Land Surface Temperature from ECOSTRESS data

Recent advances in remote sensing have emphasized the need for precise Land Surface Temperature (LST) retrieval, critical for assessing Earth’s energy balance and geophysical processes at global and local scales. This study compares sixteen Split Window Algorithms (SWAs) for LST estimation and evaluates their sensitivity for ECOSTRESS (ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station) data to identify the most reliable methods. SW coefficients were derived from at-sensor signal simulations using the MODerate resolution atmospheric TRANsmittance (MODTRAN) Radiative Transfer Model (RTM) under varying conditions for the ECOSTRESS bands. Among the sixteen SWAs, some are independent of Atmospheric Columnar Water Vapour (ACWV) and View Zenith Angle (VZA), while others depend on one or both parameters. The sixteen SWAs evaluated in this study comprise thirteen previously published algorithms and three modified versions. The terms added in the modified SWAs involve the introduction of quadratic temperature difference terms and explicit VZA correction terms to improve the atmospheric correction. The comparison with simulated dataset, ECOSTRESS LST product, and SURFRAD (Surface Radiation Budget Network) in-situ observations revealed that among the sixteen SWAs, Cheb Du et al. (2015) algorithm and modified François et al. (1996) algorithm consistently outperformed mainly attributed to inclusion of quadratic term, which better account for atmospheric non-linearity. Using simulated data across a range of ACWV and VZA conditions, Cheb Du and modified François algorithms exhibited RMSEs of 0.36–2.00 K and 0.36–1.62 K, respectively. Validation against ECOSTRESS LST yielded a maximum RMSEs of 0.51 K (Cheb Du) and 0.59 K (Modified François), while SURFRAD in-situ observations resulted in RMSEs of 2.69 K, 2.75 K, and 2.71 K for Cheb Du, modified François, and the standard ECOSTRESS product, respectively. These results highlight the suitability of both SWAs for accurate LST retrieval in upcoming missions such as TRISHNA, GISAT-1A, and SBG.