高效用水和土壤盐渍化防治是干旱区灌区可持续发展的核心。灌区应用广泛的秋浇、冬灌（简称秋冬灌溉）具有储水和洗盐等作用。然而，秋冬灌溉定额大，一方面可能造成水资源的浪费，另一方面会使灌区地下水位过高，增加土壤次生盐渍化的风险。因此，秋冬灌溉制度的研究关系到灌区节水控盐的总目标能否实现。论文以中国干旱区最大的灌区——内蒙古河套灌区为代表，通过融合多源遥感数据，建立了三个递进模型以反演秋冬灌溉范围、时间和水量，包括像元尺度秋冬灌溉时空演进的遥感反演模型（阈值法）、亚像元尺度秋冬灌溉时空演进的遥感反演模型（随机森林与长短期记忆网络模型相结合）和农田秋冬灌溉水量的遥感反演模型（水量平衡分析与随机森林模型相结合）。基于反演结果，揭示了以上要素的空间分布、年际变化及其驱动因素；进一步提出了像元尺度的灌溉模式划分方法，分析了不同模式的变化规律。使用MODFLOW-SURFACT构建了适用于秋冬灌溉情景的区域地下水数值模拟模型，揭示了秋冬灌溉对灌区地下水动态的影响机制。将以上模型应用于河套灌区，反演得到的2010至2020年秋浇范围和时间的总体精度分别达到90%和76.4%，水量估算的合理性也得到了有关研究的验证。研究发现，秋浇范围的空间分布在不同年份间差异较小，主要集中在解放闸、永济、乌兰布和南部、义长东部及乌拉特北部，非灌溉区则主要位于草地、荒地和向日葵种植区。秋浇一般从10月15日开始大面积展开，到11月15日基本完成，但其空间分布存在明显的年际差异。秋浇面积总体呈下降趋势，年均减少53.76 km2，灌溉时间趋向延后，这些变化主要受作物种植模式、气候变化及水费政策等因素影响。根据像元尺度上灌溉面积和时间的不同，识别出了灌区八种秋浇模式，其中集中模式占主导，且短时间灌溉地区逐步增加，反映了土地政策促进的灌溉集约化。灌水深度在年际间相对稳定，集中在10–30 cm。各灌域的平均灌水深度从上游向下游递增。整个灌区的秋浇用水量年均减少约1900万m3，主要原因是秋浇面积的减少。秋浇期间，解放闸灌域的地下水位平均抬升1.33 m。灌溉入渗和渠系渗漏是地下水系统的主要补给来源，而沟道排水是主要排泄途径。进一步揭示了渠道和灌溉地地下水水平交换（内排水）特征的差异。以上结果可为灌区节水及水盐合理调控提供参考依据。综上所述，利用遥感技术获取的秋冬灌溉时空分布信息精度较高，有助于灌区优化灌溉制度、合理调控水盐平衡，同时为农业水文过程的研究提供支持。
High-efficient use of irrigation water and prevention of soil salinization are the core for sustainable development of arid irrigation districts. Irrigation during late autumn to early winter (referred to as autumn/winter irrigation) is widely adopted in these areas, aiming at water storage and salt leaching. However, the large quota for autumn/winter irrigation might lead to the wastage of water resources, and also could result in excessively high groundwater levels in the irrigation districts that increases the risk of secondary soil salinization. Therefore, the study of the autumn/winter irrigation scheduling is crucial to achieving the goal of water-saving and salinization control in the arid irrigation districts.Taking the largest irrigation district in the arid regions of China, the Hetao Irrigation District in Inner Mongolia, as an example, multi-source remote sensing data are integrated to establish three progressive models for the extent, timing, and volume inversions of autumn/winter irrigation. The first is a pixel-scale remote sensing inversion model that tracks the temporal and spatial evolution of irrigation using the threshold method. The second model operates at a sub-pixel scale, combining Random Forest and Long Short-Term Memory network models to refine the analysis of temporal and spatial evolution of irrigation. The third model focuses on estimating autumn/winter irrigation volume in identified irrigated farmland, employing a combination of water balance analysis and the Random Forest model. Based on the model inversion results, spatial distribution, interannual variations, and their driving factors of these elements were revealed. A pixel-scale irrigation pattern classification method is further proposed to analyze the variation pattern of irrigation modes in the district. Moreover, a regional groundwater numerical simulation model suitable for autumn/winter irrigation scenarios was developed using the MODFLOW-SURFACT software to reveal the mechanism by which autumn/winter irrigation affects the groundwater dynamics.The models are applied to the Hetao Irrigation District from 2010 to 2020, achieving an overall accuracy of 90% for irrigation extent and 76.4% for timing, with the rationality of irrigation water use estimates validated by other studies. It was found that the spatial distribution of autumn irrigation extent shows minor interannual variations, primarily concentrated in areas like Jiefangzha, Yongji, southern Wulanbuhe, eastern Yichang, and northern Wulate sub-irrigation districts. Non-irrigated areas are mainly found in grasslands, wastelands, and sunflower planting areas. Generally, extensive irrigation starts on October 15 and ends by November 15, but there are noticeable yearly differences in the timing of autumn irrigation. The total area of autumn irrigation in the district shows a declining trend with an average annual decrease of 53.76 km2, and the timing of irrigation is delayed, which are mainly influenced by cropping patterns, climate change, and irrigation fee policies. Based on the characteristics in irrigation area and timing at the pixel scale, eight autumn irrigation patterns were identified. The concentrated irrigation pattern dominates within the district, and the increase in short-duration irrigation areas reflects the intensification of agricultural irrigation promoted by land policy. The irrigation depth remains relatively stable year-to-year focusing in the range of 10–30 cm, while the average irrigation depth of sub-districts increases from upstream to downstream. The annual water usage for autumn irrigation across the whole district decreases by 19 million m3, driven primarily by the reduction in the area of autumn irrigation. During the autumn irrigation period, the groundwater level in the Jiefangzha sub-irrigation district rises by an average of 1.33 m. Irrigation infiltration and canal leakage are the main sources of replenishment for the groundwater system, while drainage to drainage ditches is the primary discharge pathway. Distributed drainage results further reveal different interior drainage characteristics between canals and irrigation lands. These results provide a basis for reference in water-saving irrigation and rational water-salt management in the district.In summary, the spatial-temporal distribution information of autumn/winter irrigation obtained through remote sensing technology is highly accurate, which facilitates the optimization of irrigation regimes in irrigation districts, the regulation of water and salt balance, and provides support for research on agro-hydrological processes.
