Global evaluation of high-resolution ECOSTRESS land surface temperature and emissivity products: Collection 1 versus Collection 2
Auteurs
Zhang H., Mahmood A.N., Hu T., Mallick K., Didry Y., Hitzelberger P., Szantoi Z., Pérez-Planells L., Göttsche F.M., Hulley G.C., Hook S.J.
Référence
Remote Sensing of Environment, vol. 326, art. no. 114799, 2025
Description
The ECOsystem Spaceborne Thermal Radiometer Experiment on Space Station (ECOSTRESS) mission, launched to the International Space Station on June 29, 2018, currently provides high spatial resolution thermal observations in five bands with a revisit time of 1–5 days. The ECO2LSTE product, which provides the land surface temperature (LST) and emissivity (LSE) retrieved using the temperature and emissivity separation (TES) algorithm, serves as an essential ECOSTRESS product for generating the other higher-level products such as evapotranspiration and water use efficiency. Considering the radiometric calibration issues identified in the ECOSTRESS Collection 1 (C1) radiance data, recently the Collection 2 (C2) products have been released to address the cold bias in LST estimates below 300 K. Additionally, following the restoration of the two bands centered at 8.29 and 9.20 μm, the ECOSTRESS TES algorithm was shifted from 3-band to 5-band after May 2023. Consequently, a comprehensive evaluation is necessary to have a better understanding how LST&E estimates from different collections and algorithms perform. In this study, the impact of radiometric recalibration on LST&E estimates was first investigated for both the 3-band and 5-band TES algorithms based on a representative simulation dataset. Subsequently, we evaluated the accuracy of ECOSTRESS LST&E estimates in both collections using globally distributed ground measurements, and a spatial comparison between the C1 and C2 LST estimates was conducted over three selected regions to investigate their discrepancies of temperature spatial distribution. The simulation analysis revealed that the C2 LST was marginally lower than the C1 LST above 320 K but became higher below 320 K, with significant overestimation observed below 270 K. For LSE, the C2 estimates were generally lower than those in C1, particularly in the first three bands. Compared to the 3-band TES, larger differences between the C1 and C2 LST&E estimates were found for the 5-band TES. Based on the site-scale evaluation of LST, the cold bias between 270 and 300 K was effectively corrected in C2, with the bias in this range significantly reduced from approximately −2 K to 0. However, an overestimation exceeding 2 K was found for the C2 LST below 270 K, which may be caused by the overcorrection at low temperatures. LST estimates from both collections performed similarly above 300 K, with a systematically overestimation about 1 K. The spatial evaluation of LST reaffirms the findings from the simulation analysis and site-scale evaluation regarding the differences between C1 and C2 LST estimates. In the site-scale evaluation of LSE, a minor improvement was achieved in C2 over non-gray bodies, but the accuracy decreased over gray bodies. The LSE estimates using the 3-band and 5-band TES algorithms were comparable. Overall, this study provides updated information on the accuracies and differences of ECOSTRESS LST&E estimates from both collections, promoting the informed use of ECO2LSTE products in various downstream research applications with a clearer understanding of their performance.
