Predicting daily land surface temperature at high spatial resolution by fusing Sentinel-3 and Sentinel-2 data

Spatially distributed land surface temperature (LST) is important for many environmental studies, in particular applications related to water resources management over agricultural sites. These studies require LST at both high spatial and high temporal resolution. However, the current situation is that there exists a trade-off, such that one can obtain either high spatial or high temporal resolution, thus limiting the use of thermal imagery. To overcome this shortcoming, various sharpening (often termed also downscaling) algorithms have been developed, one of which is the TsHARP method. TsHARP is based on the retrieval of an empirical relationship between visual/near-infrared data (VNIR) and LST to sharpen thermal maps to the native resolution of VNIR images. The aim of our research was to examine the TsHARP algorithm utility for enhancing the spatial resolution of LST acquired by Sentinel-3 by fusing it with Sentinel-2 data over vegetated areas. Two sites were used for examining the suggested analysis, the first one near Campo Grande in South America (488 ha), and the second one in the Central part of the Czech Republic (115 ha). A two-step validation procedure was used: 1) applying TsHARP on a Landsat image that was degraded to a 1000 m resolution, using Landsat derived VNIR data; 2) applying TsHARP on Sentinel-3 LST image (1000 m resolution) using Sentinel-2 derived VNIR data. In both cases, the sharpened images were compared to the original Landsat TM image. Our results indicate that thermal sharpening of Sentinel-3 LST using Sentinel-2 VNIR yielded satisfactorily LST images at a higher than originally acquired spatial resolution. Statistical analysis shows a small difference between the reference Landsat LST and the sharpened Sentinel-3 LST images (RMSE = 1.1 - 1.41 K, MAE = 0.93 -1.05 K, R2 = 0.65 - 0.75). These initial results demonstrate the potential application of TsHARP to different satellite platforms to map temperature variability at higher than currently available spatial resolution, and thus enhance the existing information regarding vegetation stress and water status.