The contribution of remote sensing data assimilation to simulate daily evapotranspiration of irrigated and non-irrigated crops in semi-arid context

Remote sensing data provide valuable information on the spatial distribution of land surface conditions and properties, such as soil moisture, soil and vegetation water status. However, the frequency and resolution of remotely sensed data vary depending on the satellite and sensor. The frequency of observation of thermal infrared that allows an estimation of evapotranspiration is carried out daily by the satellites AQUA and TERRA (res. 1km), every 2 days by Sentinel-3 (res. 1km), 8 days by LANDSAT-8 and 9 (res. 60m) and will be 3 times per period of 8 days by the satellite TRISHNA (res. 60m). In addition, there is no data on days with heavy cloud cover. In order to obtain a daily evaluation of ET, we propose to correct the trajectory of a surface model based on the water balance with the assimilation of ET data from remote sensing. The question is what are the advantages of assimilation compared to open-loop or interpolation of observation. We present our work on modelling evapotranspiration and irrigation at the field scale with the SAMIR (Satellite monitoring of irrigation) model. This is a crop water balance model forced by weather data, soil and crop parameters to simulate the daily components of the water balance. A particle filter method is implemented to assimilate evapotranspiration from remote sensing. This evaluation is performed on several types of crops (wheat, barley and olive), irrigated or not, and in a semi-arid Mediterranean context (Tunisia and Morocco). Compared to open loop simulations, data assimilation allows to quickly reduce the simulation uncertainty. On the other hand, the higher the revisit frequency, the more the simulation uncertainty depends on the observation uncertainty and the model uncertainty is reduced.