Spatially explicit estimation of high-resolution irrigation water use across China using earth observation data and deep learning

Accurately estimating irrigation water use (IWU) at high spatial resolution is essential for addressing water scarcity and promoting sustainable agricultural water management. This study develops a physically guided framework that integrates Earth observation data, deep learning, and water balance modeling to generate a 500-m resolution IWU dataset across China during 2004–2019. Through multi-source data fusion and advanced downscaling strategies, we produce high quality hydrological variable products including evapotranspiration (ET), precipitation, and surface soil moisture. Irrigation period is detected using an integration of remotely sensed land surface temperature and reanalysis-based skin temperature by exploiting irrigation-induced surface cooling effect. We employ a deep learning strategy that transfers the water balance-based paradigm derived from non-irrigation periods to estimate non-ET water loss components during irrigation periods, demonstrating predictive performance with correlation coefficients exceeding 0.9 in test set. Validation against reported records from Chinese Ministry of Water Resources and cropland stations confirms the robustness of our IWU estimates, yielding root mean square error of 25.3 mm/yr and 3.9 km³/yr at station and national scales, respectively. The resulting IWU dataset reveals a peak in national IWU around 2013 (with a positive trend of 2.53 km³/yr during 2004–2013), followed by a reversal to a declining trend of −3.32 km³/yr during 2014–2019. This shift aligns with policy-driven improvements in irrigation conveyance efficiency and the implementation of China’s most stringent water resource management regulations. By enhancing the quality of remote sensing forcing datasets and the physical realism of model framework, our work could offer a transferable solution for irrigation monitoring and sustainable water resource management at both national and global scales. © 2026 International Society for Photogrammetry and Remote Sensing, Inc. (ISPRS). Published by Elsevier B.V. All rights are reserved, including those for text and data mining, AI training, and similar technologies.