Evaluating the feasibility of using Sentinel-2 and Sentinel-3 satellites for high-resolution evapotranspiration estimations

Abstract

The Sentinel satellite missions are designed to provide remote-sensing observational capability to many diverse operational applications, including in the field of agriculture and food security. They do this by acquiring frequent observations from a combination of optical, thermal and microwave sensors at various spatial resolutions. However, one currently missing capability, that would enable monitoring of evapotranspiration, crop water stress and water use at field scale, is the lack of high-resolution (tens of meters) thermal sensor. In this study we evaluate a methodology for bridging this data gap by employing a machine learning algorithm to sharpen low-resolution thermal observations from the Sentinel-3 satellites using images acquired by high-resolution optical sensors on the Sentinel-2 satellites. The resulting dataset is then used as input to land-surface energy balance model to estimate evapotranspiration. The methodology is tested using Terra and Landsat satellite observations, due to lack of sufficiently long time-series of Sentinel observations, and benchmarked against fluxes derived with high-resolution thermal observations acquired by the Landsat satellites. We then apply the methodology to Sentinel-2 and Sentinel-3 images to confirm its applicability to this type of data. The results show that the fluxes derived with sharpened thermal data are of acceptable accuracy (relative error lower than 20%) and provide more information at flux-tower footprint scale than the corresponding low-resolution fluxes. They also replicate the spatial and temporal patterns of fluxes derived with high-resolution thermal observations. However, the increase in error of the modelled fluxes compared to using high-resolution thermal observations and the inherent limitations of the sharpening approach point to the need to add high-resolution thermal mission to the Sentinels' constellation.