An inter-comparison of approaches and frameworks to quantify irrigation from satellite data

Abstract. This study provides the first inter-comparison of different state-of-the-art approaches and frameworks that share a commonality in their utilization of satellite remote-sensing data to quantify irrigation at a regional scale. The compared approaches vary in their reliance on either soil moisture or evapotranspiration data or their joint utilization of both. The two compared frameworks either extract irrigation information from residuals between satellite observations and rainfed hydrological models in a baseline framework or use soil water balance modeling in a soil-moisture-based inversion framework. The inter-comparison is conducted over the lower Ebro catchment in Spain where observed irrigation amounts are available for benchmarking. Our results showed that within the baseline framework, the joint approach using both soil moisture and evapotranspiration (ET) remote-sensing data only differed by +37 mm from the irrigation benchmark (922 mm) during the main irrigation season over 2 years and by +47 and −208 mm for approaches relying solely on soil moisture and ET, respectively. A comparison of the different frameworks showed that the main advantage of the more complex baseline framework was the consistency between soil moisture and ET components within the hydrological model, which made it unlikely that either one ended up representing all irrigation water use. However, the simplicity of the soil-moisture-based inversion framework, coupled with its direct conversion of soil moisture changes into actual water volumes, effectively addresses the key challenges inherent in the baseline framework, which are associated with uncertainties related to an unknown remote-sensing observation depth and the static depth of the soil layers in a conceptual model. The performance of the baseline framework came closest to the irrigation benchmark and was able to account for the precipitation input, which resulted in more plausible temporal distributions of irrigation than what was expected from the benchmark observations.