Benefits and pitfalls of irrigation timing and water amounts derived from satellite soil moisture

Despite the key role of irrigation in the Earth system, we lack fundamental information regarding the distribution of irrigated fields, irrigation timing and the amount of water utilized. In the past years, the SMDelta and SMInversion approaches have been independently developed to provide estimates of irrigation timing and water amounts based on satellite soil moisture data. The SMDelta approach retrieves irrigation from variations in soil moisture between an individual pixel and the surrounding rainfed area, while the SMInversion approach estimates the total amount of water entering the soil, then irrigation is derived by subtracting precipitation. In this study, we perform a comprehensive assessment of irrigation estimates from the SMDelta and SMInversion algorithms based on Sentinel-1 surface soil moisture retrievals at 1 km resolution. Our analysis focuses on the Ebro basin, an irrigated region in Spain covering 83000 km2, during the period 2017–2019. We assess the ability of the two methods to discriminate irrigated and rainfed pixels, then we quantify the agreement of irrigation timing and water volumes with reference irrigation data. An inter-comparison between estimates from the SMDelta and SMInversion methods is carried out considering both temporal and spatial features, i. e. , monthly irrigation peaks and spatial irrigation patterns. Finally, we explore two potential applications of satellite-derived irrigation estimates: attributing irrigation water volumes to specific irrigation systems and to individual crops. We observe that both methods erroneously retrieve irrigation over rainfed pixels, and are therefore not suitable to map irrigated and rainfed fields. However, when auxiliary information on irrigated fields is available, we find a satisfactory agreement between district-scale reference data and satellite-retrieved irrigation, using both the SMDelta and SMInversion approaches (Pearson R equal to 0. 67 and 0. 71, bias equal to −4. 99 and −4. 75 mm/15 days, respectively). When aggregated in space or time, the irrigation estimates exhibit coherent temporal dynamics and spatial patterns. For instance, estimates from both SMDelta and SMInversion capture the delayed irrigation that occurred in 2018 due to wetter than usual conditions in spring. However, at the pixel-scale, limited consistency exists between irrigation estimates from the two methods due to different assumptions and parameterizations, e. g. , use of constant vs pixel-specific soil water capacity (in the SMDelta and SMInversion, respectively). Overall, the study demonstrates the reliability of irrigation estimates derived from the SMDelta and SMInversion approaches, especially when shifting from small spatial and short temporal scales (pixel level, sub-weekly) to larger and longer scales (district level, seasonal). Hence, satellite-based irrigation estimates could inform water resources managers and basin authorities, as well as serve the modelling community by providing reliable information on the timing and the amounts of water employed at the basin level.