Downscaling MODIS land surface temperature to Sentinel-2 spatial resolution in the Barrax test site

The traditional limitation in the lower spatial resolution of Thermal Infrared (TIR) versus Visible Near Infrared (VNIR) satellite data can be faced by applying recent disaggregation techniques. These techniques are based on the VNIR-TIR variable regressions at coarse spatial resolution, and the assumption that the relationship between spectral bands is independent of the spatial resolution. A comprehensive analysis of different disaggregation methods in the literature using MODIS and Landsat images was already addressed by [1] in a previous publication. The aim of this work is to evaluate the performance of the downscaling method that showed the best results, when applied now to the MODIS/Sentinel-2 tandem for the estimation of daily land surface temperature (LST) at 10 m spatial resolution. An experiment was carried out in an agricultural area located in the Barrax test site, Spain (39o 03’ 35’’ N, 2o 06’ W), for the summer of 2018. Ground measurements of LST transects centered in the MODIS overpasses, and covering a variety of crops and surface conditions, were used for a robust local validation of the disaggregation approach. An additional set of Landsat-7/ETM+ images were also used for a more extended assessment of the LST product generated. Data from 6 different dates were available for this study, covering 10 different crop fields. Despite the large range of temperatures registered (300-325 K), differences within ±4.0K are obtained, with an average estimation error of ±2.2K and a systematic deviation of 0.6K for the full dataset. A similar error was obtained for the extended assessment of the high resolution LST products, based on the pixel-to-pixel comparison between Landsat and disaggregated Sentinel-2 LST products.