Ground calibration of Sentinel-3 SLSTR thermal infrared bands and validation of LST algorithms

Land surface temperature (LST) is an important variable for studying environmental changes and monitoring the Earth climate system. The SLSTR on board Sentinel-3A and 3B satellites is a key sensor that provides LST data on a global scale with daily revisit. The objectives of this paper are: (i) To check the on-board calibration of the SLSTR thermal-infrared bands calibration against ground-measured data for the first time. (ii) To validate the SLSTR LST product and an alternative emissivity-dependent split-window algorithm (SWA) for SLSTR data. To this end, we used the temperature based (t-based) and radiance based (r-based) methods. Ground LSTs measured from narrow-band, directional radiometer measurements from nine homogeneous sites were used for the calibration assessment and the t-based validation. The nine sites represent four different biomes. R-based validation was performed over the same t-based sites plus other five sites of varied biomes where ground LST measurements were not available. The analysis was performed for a dataset corresponding to the period 2020–2022. Results from the calibration assessment showed R-RMSDs (robust root mean square difference) between the simulated and satellite brightness temperatures of 0.7 K and 0.6 K for bands 8 and 9, respectively, for both sensors (3A and 3B), with small median differences. Such R-RMSDs were well within the uncertainty limits of the simulated brightness temperatures. The validation of the operational SLSTR product and the SWA yielded consistent results when comparing Sentinel-3A and 3B. For the t-based method, the SLSTR product showed an overall daytime (nighttime) R-RMSD of 1.5 K (1.7 K), with medians (satellite minus ground) varying between −2.0 K and 1.5 K and thus exceeding the recommended ±1.0 K threshold. For the SWA, R-RMSDs were 1.3 K both for daytime and the nighttime, and medians varied between −0.2 K and −0.9 K for daytime and nighttime. The r-based results confirmed the large medians in the SLSTR product, which ranged from −1.5 K to 1.7 K (from 0.1 K to 0.4 K for SWA). The SLSTR product showed R-RMSDs of 1.0 K (1.0 K) for the overall daytime (nighttime), while they were 0.6 K (0.3 K) for the SWA, which were lower than those for the t-based validation. The results presented here contributed to the global validation of LSTs derived from the SLSTR sensor onboard the Sentinel-3A and 3B satellites.