Abstract
Remote sensing data play an important role in understanding the spatio-temporal variations in hydro-meteorological conditions at different spatial scales. In particular, one of the key processes of hydrological cycle for monitoring water loss from space is evapotranspiration (ET). In contrast to sparsely distributed in-situ measurements, development of two surface energy balance (TSEB) models forced by satellite observations has made a significant contribution to estimate ET with global coverage. In this regard, in the framework of ESA’s 4DMED-Hydrology project, we combine Copernicus data from Sentinel-2 (S2) Multispectral Instrument (MSI) and Sentinel-3 (S3) Land Surface Temperature Radiometer (SLSTR) with ERA5 climate reanalysis dataset derived within the period 2017-2021 for daily ET retrieval at high (100 m) spatial resolution. In this work, an open-source implementation of TSEB developed in the framework of the ESA’s Sen-ET project has been applied over wide areas represented by four Mediterranean basins in Italy, Spain, France, and Tunisia (Po, Ebro, Hérault and Medjerda). Considering large volume of satellite data and high computational requirements of the Sen-ET, all processes have been optimized to be run in the automatic manner by combining multiple steps into one processing workflow utilized in cloud computing platforms offered by EODC and ESA HPC of CloudFerro. First, due to incomplete time-series of S2 Level-2A, we pre-process Sentinel-2 data for further retrieval of 100-m reflectance and biophysical parameters needed for the ET estimation afterwards. Next, we downscale S3 land surface temperature (LST) product by exploiting relationships between 1-km Sentinel-3 and time-coincident 100-m S2 reflectances using decision trees (DT) algorithm. Apart from biophysical properties (e.g., leaf area index and fractional vegetation cover) and sharpened LST data, meteorological forcings and solar radiation from ERA5 have been generated for estimating instantaneous energy fluxes and daily evapotranspiration. Based on preliminary results over Po basin, DT algorithm allowed predicting 100-m LST with the average root mean square error (RMSE) of 3.2°C when compared to ground-derived skin temperature from two eddy covariance (EC) towers. Meanwhile, turbulent fluxes driven by downscaled LST resulted in RMSE equal to 52 Wm-2 and 108 Wm-2 for sensible and latent heat fluxes, respectively. Despite some limitations mainly related to the EC locations in complex mountain areas, ET estimates forced by satellite observations have potential for providing energy fluxes at wider scale.