Agricultural irrigation accounts for nearly 70% of global freshwater withdrawals, making sustainable water management crucial for food security and ecological stability—particularly in arid and semi-arid regions. However, dynamic water-saving thresholds at both inter-annual and intra-annual scales remain insufficiently quantified in current research. To address this gap, this study developed an integrated SWAT-MODFLOW model for the Hetao Irrigation District and quantified dynamic water-saving thresholds by simulating crop yield responses under a range of irrigation scenarios. The model was calibrated (2008–2014) and validated (2014–2016), demonstrating reliable performance (R² = 0.75, NSE = 0.74) in capturing local hydrological processes. Inter-annual scenarios assessed water-saving levels of 5%, 10%, 20%, and 30% under wet, normal, and dry years, while intra-annual scenarios adjusted seasonal irrigation volumes in spring, summer, and autumn with reduction gradients of 33%, 50%, and 100%. Results show that wet and normal years could achieve a water-saving threshold of up to 20%, whereas dry years were limited to 5%. Intra-annually, autumn irrigation offered the greatest saving potential (33–100%), followed by spring (33–50%). Spatially, crop responses varied substantially: the western part of the region proved particularly sensitive, with even the optimal district-wide strategy reducing local crop yields by 10–20%. This study quantifies dynamic water-saving thresholds and incorporates spatial heterogeneity into scenario assessment. The resulting framework is transferable and provides a basis for sustainable water management in water-limited agricultural regions.

Persistent cloud cover during the growing season and mosaic cropping patterns introduce temporal gaps and mixed pixels, undermining the reliability of large-scale crop identification and acreage statistics. To address these issues, we develop a high spatiotemporal-resolution remote-sensing approach tailored to heterogeneous farmlands. First, an improved Spatiotemporal Adaptive Reflectance Fusion Model (STARFM) is used to fuse Landsat, Sentinel-2, and MODIS observations, reconstructing a continuous Normalized Difference Vegetation Index (NDVI) time series at 30 m spatial and 8-day temporal resolution. Second, at the field scale, we derive phenological descriptors from the reconstructed series—key phenophase timing, amplitude, temporal trend, and growth rate—and use a Random Forest (RF) classifier for detailed crop discrimination. We further integrate SHapley Additive exPlanations (SHAP) to quantify each feature’s class-discriminative contribution and signed effect, thereby guiding feature-set optimization and threshold refinement. Finally, we generate a 2024 crop distribution map and conduct comparative evaluations. Relative to baselines without fusion or without phenological variables, the fused series mitigates single-sensor limitations under frequent cloud/rain and irregular acquisitions, enhances NDVI continuity and robustness, and reveals inter-crop temporal phase shifts that, when jointly exploited, reduce early-season confusion and improve identification accuracy. Independent validation yields an overall accuracy (OA) of 90.78% and a Cohen’s kappa(κ) coefficient of 0.882. Coupling dense NDVI reconstruction with phenology-aware constraints and SHAP-based interpretability demonstrably improves the accuracy and reliability of cropping-structure extraction in complex agricultural regions and provides a reusable pathway for regional-scale precision agricultural monitoring.