A Soil Moisture Profile Response-Driven Framework for Estimating Irrigation Water Use Under Spatial Allocation Constraints

Accurate estimation of irrigation water use is essential for agricultural water accounting and water-resource allocation in large irrigated districts, yet existing statistics are usually available only as aggregated administrative totals and cannot adequately characterize seasonal and spatial differences in field-applied water. In this study, a soil moisture profile response-driven framework was developed to estimate spring and summer irrigation water use in the Hetao Irrigation District, a typical large-scale irrigated region in the upper Yellow River Basin. Soil property zones were first delineated using K-means clustering based on field capacity, wilting point, available water capacity, bulk density, porosity, and electrical conductivity, and season-specific soil profile response layers were identified through bootstrap stability tests using 0–100 cm daily soil moisture changes. A net water inflow response was then constructed by integrating soil water storage change, precipitation, and evapotranspiration, and six estimation models were compared, including a soil-water-response conversion model, historical-management baseline models, and recent-management baseline models. Model calibration was conducted for 2016–2021, and independent testing was performed for 2022. Spatial allocation constraints were further introduced to ensure consistency between total estimated irrigation volume and pixel-scale irrigation depth patterns. Results showed that the optimal soil profile stratification was 0–20/20–60/60–100 cm for spring irrigation and 0–20/20–50/50–100 cm for summer irrigation, indicating clear seasonal differences in profile response. For spring irrigation, the recent three-year management baseline model performed best, with a training-period RMSE of 8.8 mm and MAPE of 8.5%, and a testing-period RMSE of 8.4 mm, bias of −2.7%, and R2 of 0.97. For summer irrigation, the historical-median management baseline model was most robust, with a training-period RMSE of 8.1 mm and MAPE of 14.0%, and a testing-period RMSE of 4.4 mm, bias of −9.4%, and R2 of 0.96. Spatially, spring irrigation depths increased from 2019 to 2022 and were higher in WLBH, JFZ, and YJ, whereas summer irrigation depths were generally lower and more concentrated in western and central sub-irrigation districts. The proposed framework provides a practical approach for linking soil moisture profile response, management-based volume constraints, and spatially explicit irrigation mapping in large-scale irrigated regions.