Remote sensing-based estimation of agricultural water : from irrigation to drainage ; Estimation de l'eau agricole basée sur la télédétection : de l'irrigation au drainage

Water used for irrigation represents 70% of global freshwater consumption. Yet very little information on irrigation is available at any spatial scale. Furthermore, the impact of irrigation activities on downstream ecosystems is not often assessed, as drainage from irrigated perimeters is still poorly monitored. In this context, this thesis aimed to develop tools combining remote sensing data and a crop water balance model (SAMIR) to estimate irrigation and drainage of irrigation districts. These tools were calibrated and validated on the 80 km² Algerri-Balaguer (AB) irrigation district in northeastern Spain, where irrigation and drainage data are available on a few experimental plots and at the district scale. First, the sensitivity of SAMIR was analyzed using the Sobol method on 10 sites and 37 agricultural seasons. Of the 12 parameters analyzed, two account for most of SAMIR's sensitivity. A SAMIR sensitivity proxy can predict with good accuracy which is the most sensitive for a given site. Next, a new irrigation simulation approach based on irrigation data assimilation was developed to invert SAMIR's irrigation parameters. The trigger threshold and the irrigation dose were inverted on a monthly basis using the 2019 irrigation data from the AB district. These parameters were then applied over five years (2017 to 2021), resulting in a very good simulation of the observed weekly irrigation (correlation coefficient r of 0.95 ± 0.02). Then, the previous method was adapted to replace irrigation data with high-resolution soil moisture data (15 m) derived from Sentinel-1 as inputs to the assimilation scheme. This approach has shown very encouraging results for 2019 (RMSD of 6.7 mm/week), and better performance than that obtained with a classic irrigation module. Finally, the precision (with calibration using field data) and accuracy (with default calibration) of four parsimonious drainage models (two to four sensitive parameters) were evaluated. All four models showed very good precision (KGE (Q0.5) of 0.77 on ...