合理的水盐调控是灌区可持续发展的基础。在干旱区灌区，大量引水灌溉往往导致地下水位升高，并引发盐渍化等问题。本文以内蒙古河套灌区作为干旱区典型灌区，基于遥感蒸散发计算结果，分析灌区景观单元和主要作物的耗水规律，并建立干旱区灌区水文模型和盐分迁移模型，模拟典型干旱区灌区的排水、排盐和内排水、旱排等水盐运移过程，为干旱区灌区水盐合理调控提供支撑。为识别灌区主要作物的多年空间分布，提出了前置约束的机器学习方法，对2012至2016年河套灌区主要作物（玉米、向日葵和小麦）的空间分布进行识别。结果表明，该方法识别精度较高，并且克服了机器学习时间泛化性差的问题，在河套灌区应用效果良好。结果表明研究时段内灌区小麦种植面积逐渐减少，玉米和向日葵种植面积逐渐增加，其中向日葵从五原南部延伸至整个河套中东部。利用混合双源遥感蒸散发模型（HTEM）计算了河套灌区2008至2016年蒸散发的时空分布。进一步结合土地利用和主要作物识别结果，对不同景观单元、主要作物的蒸散发规律进行了对比研究。结果表明，5-9月的农田蒸散发仅次于湿地水体，年均475 mm。在每年5月，由于向日葵播种前的大量灌水压盐，其蒸散发显著高于玉米和小麦，全生育期平均耗水492 mm。基于遥感蒸散发结果，定义渠系和沟道控制单元等灌区基本单元，耦合排水模型、长周期地下水波动模型、冻结期地下水位经验模型以及地下水横向交换模块，建立了新的灌区水文模型和盐分迁移模型。除了蒸散发、潜水蒸发、沟道排水等灌区水文过程及其伴生的盐分迁移过程，模型中重点考虑了干旱区地下水浅埋灌区特有的内排水和旱排过程。结果表明，模拟沟道排水和排盐的纳什效率系数（NSE）和确定性系数（r2）分别高于0.74和0.87，模拟地下水埋深和矿化度的NSE和r2分别高于0.93和0.84。模型能够准确描述河套灌区主要的水文过程及盐分迁移过程。在以上结果的基础上，对河套灌区水、盐的迁移、平衡和演变的趋势和规律进行分析。结果表明，灌区内排水占灌溉的量12.87%-14.98%，较沟道排水多34.9%，是灌溉地最主要的水分排泄途径。内排水作用在乌兰布和、义长中南部效果明显，在乌拉特北部效果最差。当前的引水灌溉对地下水位调控有积极作用。解放闸和永济灌域盐分积累较多，需要减少引水，完善排水系统，改善种植结构，提高旱排和沟道排盐效率。研究结果可为干旱区灌区水盐合理调控提供技术支撑。
Appropriate water and salt regulation is essential for sustainable development of irrigation districts. In arid irrigation district, agricultural production relies heavily on irrigation due to intense evapotranspiration and sparse precipitation. However, the large amount of irrigation usually causes groundwater table rising and results in waterlogging and salinity problems, which severely threating the food production and ecological environment. In current paper, Hetao Irrigation District (HID) of Inner Mogolia, the typical and largest arid irrigation district in China, is selected as the study area. The remote sensing-based crop identification and evapotranspiration models are developed to analyze water consumpution of different landscapes and dominated crops in the HID. Based on the remote sensing-based evapotranspiration results, the distributed models for water flow and salt transport are developed to simulate typical agro-hydrological processes in aird irrigation district, such as drainage, interior drainage, and dry drainage. The current study is significant for exploring the efficient water utilization and regulation measures of waterlogging and salinity problems.In order to map the multi-year spatial distribution of dominated crops (maize, sunflower, and wheat) in the HID, a pre-constrained machine learning method is proposed and applied in the HID from 2012 to 2016. Results indicate that the method shows good performance in the HID and can enhance the temporal generalization of traditional machine learning method. Affected by economic benefit, the planting areas gradually decreased for wheats, and increased for maizes and sunflowers. In addition, the sunflowers increased rapidly from southern Wuyuan to the whole midlle and eastern parts of the HID.A dual-source remote sensing evapotranspiration model (HTEM) is used to calculate spatial and temporal distribution of evapotranspiration in the HID from 2008 to 2016. The evapotranspiration of different landscapes and dominated crops is compared and analyzed by combining the land use map and crop mapping results above. Results show that the average evapotranspiration of farmland is 475 mm from May to September, and higher than other landscapes except for wetland and water body. In addition, due to the the large irrigation for soil leaching for sunflower areas in May, the evapotranspiration of sunflower (492 mm) is significantly higher than maize and wheat from May to September.Based on the remote sensing-based evapotranspiration results above, new monthly distributed agro-hydrological models for water flow and salt transport are developed by defining basic agro-hydrological units, such as canal command areas and sub-draiange command areas. Modules including distributed drainage model, groundwater table flucatuation method in long term prediction, empirical model for groundwater table flucatuation estimation in freezing and thawing period, and horizontal groundwater exchange (interior drainage) model are coupled in current models. Besides general agro-hydrological processes including evapotranspiration, drainage, and groundwater evaporation, the models focus on typical interior drainage and dry drainage process between irrigated areas and non-irrigated areas in aird irrigation districts with shallow groundwater. Results show the Nash-Sutcliffe efficiency (NSE) and coefficient of determination (r2) for drainage and salinity simulation are above 0.74 and 0.87, respectively. NSE and r2 for groundwater table depth and salinity simulation are above 0.93 and 0.84, respectively. Therefore, the developed models are applicable to simulate main agro-hydrological process of the HID.The water flow and salt transport processes are further analyzed based on developed models above. Results indicate that the interior drainage is 34.9% higher than drainage into ditches, and accounts for 12.87%-14.98% of total irrigation amount. The interior drainage is higher in Wulanbuhe and Yichang sub-irrigation districts rather than Wulate sub-irrigation districts. Current water diversion strategy is positive to control the groundwater table in the HID. However, in order to improve the efficiency of dry drainage and salt discharge of ditches, sub-irrigation districts including Jiefangzha and Yongji should decrease water diversion, improve ditch system, and change planting structures. These results can provide technical support for high efficiency utilization of irrigation and appropriate regulation of salt in arid irrigation district.
