我国西北地区水资源短缺，同时生态极其脆弱。西北干旱内陆盆地特有的生态系统与地下水系统之间存在密切的相互联系。随着人类活动的增加以及全球气候变化，这种紧密联系使得两者的安全日趋严峻。为了确保流域可持续发展，亟需揭示地下水循环的规律及其对于湿地植被的影响，同时定量评估未来地下水循环及植被生态的变化趋势。本研究选取苏干湖盆地作为研究对象，首先采用图解法、氢氧稳定同位素分析以及多元统计分析等方法，阐述了盆地地下水循环的过程以及变化的规律；随后，使用地下水-地表水耦合模型模拟了盆地地下水位、河流径流量和地表水-地下水交互量的动态变化；再通过Mann-Kendall趋势检验、Moran’s I指数和时空地理加权回归模型（GTWR）等方法，详细分析了盆地湿地植被覆盖度（FVC）的时空分布以及其变化特征，深入探讨了FVC分布模式与水文、气候、地形和土壤含盐量等环境因子的复杂时空关系；然后，利用支持向量机（SVM）和随机森林算法（RF），构建了FVC动态响应模型。最后，综合应用气候变化模型（GCM）、融雪径流模型、地下水-地表水耦合模型以及FVC动态响应模型，对未来环境变化条件下地下水循环与FVC分布格局的变化趋势进行了预测分析。研究的主要结论如下：

（1）水文地球化学综合研究表明盆地地下水系统存在2个不同补给来源，而且主要排泄途径是向地表水排泄和蒸发。东-西主流向的地下水起源于盆地东部山区，流动过程中经历多次地下水-地表水相互转化，地下水流速较快。中、上游水化学类型主要受碳酸盐矿物风化控制，而下游湿地区域水化学特征主要受蒸发作用与混合作用影响，河道附近的浅层地下水受入渗过程影响强烈，地下水中盐分和重同位素不能有效累积；盆地西部的山麓地区地下水接受冰川融水补给，沿地形坡度向下流动，该区域水化学类型表现为强水岩作用特征，表明地下水补给条件差，流速缓慢。盆地中部Na–SO₄型地下水局部聚集，推断在努呼图湿地西南侧存在隐伏地下隆起构造。

（2）构建了基于MODFLOW-SFR/LAK的地下水-地表水耦合模型并利用地下水水位和遥感湖泊水位等数据进行了识别验证，通过敏感性分析发现含水层渗透系数与河床导水系数较为敏感。在模拟期内（2012~2023年），冲洪积扇与戈壁平原地下水水位变化幅度高于湿地；地下水、河流、湖泊系统在模拟期内整体储量呈增长状态（+0.69亿m³·a^-1）；地下水-地表水交互关系和交互量呈现了明显的空间差异性，在距离出山口约20 km处地表水的补给作用达到峰值，而在苏干湖湿地东侧溢出带地下水的排泄作用达到峰值，模拟期内最大值分别为6×10⁵ m³·d^-1、3×10⁵ m³ d^-1。

（3）利用遥感解译的FVC影像揭示了1986~2023年湿地植被变化规律和主要驱动因素，发现2016年为区内湿地FVC时空分布变化的突变点。1986~2016年平均FVC与分布格局均处于动态平衡状态，而2017~2023年湿地的平均FVC均呈上升趋势，极低FVC区域面积显著减小，苏干湖湿地植被核心区域向东迁移。GTWR回归系数结果表明在所有环境解释变量中，水文变量与地形变量对FVC时空分布变化的影响程度较大。地下水埋深在2个湿地与FVC均呈强负相关性，而地表径流在高FVC区域以外呈正相关性，苏干湖湿地东部植被变化可能由地表径流变化引起。地形变量对土壤水分含量以及土壤积温具有控制作用，使得湿地FVC与地形变量之间具有较强的空间异质性。气候变量的影响程度最低，降雨在高表层土壤盐分区域与FVC之间呈负相关性。

（4）构建了基于SVM和RF算法的苏干湖湿地和努呼图湿地FVC动态响应模型，发现RF算法的FVC动态响应模型模拟效果更好，且在考虑GTWR先验信息后模型的准确性提升。G-RF模型在苏干湖与努呼图湿地中模拟结果的最小均方差误差分别为0.083、0.047（2019~2022），苏干湖湿地高FVC模拟分布向东迁移，中FVC区域面积增加明显，努呼图湿地中部分极低FVC区域逐渐转化为低FVC，模拟的FVC时空分布变化与实际观测过程相符，表明模型具有较高的可靠性。特征变量重要性得分结果指出坡度特征变量对预测结果的影响最大，其次为地下水埋深变量。

（5）建立了地下水循环、气候变化与FVC动态响应的综合模拟模型，预测了未来地下水循环的动态变化和FVC分布格局的演化趋势。GCM的气温数据在研究区具有较高精度，而降水数据在雨季存在低估，因此校正处理后的气温与降水数据在研究区相关系数分别0.994以及0.790。在SSP1-2.6、SSP2-45与SSP3-7.0远期情景下，综合模拟模型模拟发现盆地上游出山口径流量呈上升趋势，径流量分别增加了25.9%、57.8%与138.8%；盆地地下水水位抬升速度为4.08、5.18与7.02 cm·a^-1；苏干湖湿地被淹没面积分别为25、33与240 km²。湿地植被平均FVC随着排放标准的增大而上升，苏干湖与努呼图湿地平均FVC在SSP3-7.0远期分别达到最大值0.415、0.313。苏干湖湿地的东部将会发展成为植被生态的核心区域，将以中或中高FVC植被为主。高FVC植被分布面积减少，在努呼图湿地中增加趋势不明显，在苏干湖湿地中则呈下降趋势。在实施调水计划后，地下水水位与湿地平均FVC上升幅度减缓。

综上，本研究立足于观测数据缺乏的苏干湖盆地生态水文研究，定量化评价了变化环境下地下水与植被的动态响应机制，有助于更好地平衡当地绿洲经济和生态安全，为流域综合管理、调水工程等政策提供科学理论依据。

Northwestern China faces a shortage of water resources and extremely fragile ecology. There exists a close interconnection between the unique ecosystems of the northwest arid inland basins and the groundwater systems. With increasing human activities and global climate change, this close connection poses increasingly severe systemic risks to both. To ensure sustainable development in the basin, and to maintain local water resources security and ecological balance, it is urgently necessary to uncover the laws governing groundwater circulation and its impact on surface vegetation ecosystems, and quantitatively assess the changing trends in groundwater circulation and the evolution of ecological vegetation in the future. This study selected the Sugan Lake Basin as the research object. Firstly, methods such as mathematical statistics, graphical analysis, stable hydrogen and oxygen isotope analysis, and multivariate statistical analysis were used to elucidate the process of groundwater circulation and its changing patterns in the basin. Subsequently, the groundwater-surface water coupling model was used to simulate the dynamic changes in groundwater level, river runoff, and groundwater-surface water interaction in the basin. Advanced methods such as the Mann-Kendall trend test, Moran’s I index, and Geographically and Temporally Weighted Regression (GTWR) model were then used to analyze the spatiotemporal distribution of regional wetland vegetation cover (FVC) and its changing characteristics. The complex temporal and spatial relationships between FVC distribution pattern and environmental factors such as hydrology, climate, topography and soil salinity were discussed. Next, machine learning models such as Support Vector Machine and Random Forest algorithms were utilized to construct a dynamic response model for FVC changes. Finally, by integrating Global Climate Model (GCM), Snowmelt Runoff Model (SRM), the groundwater-surface water coupling model, and the vegetation dynamic response model, the changing trends of groundwater circulation and FVC distribution pattern under future environmental conditions were predicted. The main conclusions of the study are as follows.

 (1) Hydrogeochemical comprehensive research indicates that the groundwater system in the basin has two distinct sources of recharge, with the main discharge pathways being discharge to surface water and evaporation. The east-west main groundwater flow originates from the eastern mountainous areas of the basin, groundwater-surface water undergoes multiple interactions during its flow, with relatively fast groundwater flow rates. The hydrochemical types of groundwater in the middle and upper reaches are mainly controlled by carbonate minerals, while the hydrochemical characteristics of groundwater in the downstream wetland areas are mainly influenced by evaporation and mixing processes. Shallow groundwater near the river channels is strongly influenced by infiltration processes, and salts and heavy isotopes in groundwater cannot effectively accumulate. Secondly, groundwater in the foothill areas of the western basin originates from glacial meltwater and flows downhill along the terrain gradient. The hydrochemical types of groundwater in this area exhibit strong water-rock interaction characteristics, indicating poor groundwater recharge conditions and slow groundwater flow rates. Halite dissolution occurs in groundwater in the foothill area, and the minerals should originate from wetland areas under the action of wind. Na-SO₄ type groundwater accumulates in the central basin, inferring the existence of hidden underground uplift structures on the southwest side of Nuhutu Wetland.

(2) A MODFLOW-SFR/LAK groundwater-surface water coupled model was constructed, and validation was conducted using groundwater levels and remotely sensed lake water levels. Sensitivity analysis revealed that the aquifer hydraulic conductivity and riverbed conductance were the most sensitive parameters. During the simulation period (2012-2023), the groundwater level changes in alluvial fans and Gobi plains are higher than those in wetlands. The overall storage of groundwater, rivers, and lakes increased during the simulation period (+0.69 billion m³·yr^-1). There are significant spatial differences in the interaction and exchange of groundwater-surface water. The recharge of surface water reaching its peak approximately 20 km from the mountain outlet, and the discharge of groundwater in the overflow zone east of Lake Sugan Wetland reaching its peak, with maximum values during the simulation period were 6×10⁵ m³·day^-1 and 3×10⁵ m³·day^-1, respectively.

(3) It was found that 2016 marked a turning point in the vegetation change patterns and primary driving factors of the Sugan Lake Wetland and Nuhutu Wetland from 1986 to 2023 were revealed using remotely sensed FVC images. Spatiotemporal distribution of wetland FVC in the region. From 1986 to 2016, the average FVC and its distribution pattern remained in a dynamic equilibrium state. However, from 2017 to 2023, the average FVC of the wetlands showed an upward trend, with a significant decrease in the area of extremely low FVC regions. The core vegetation area of the Sugan Lake Wetland shifted eastward. Results from the Geographically and Temporally Weighted Regression (GTWR) coefficients show that among all environmental explanatory variables, hydrological and topographical variables have a significant impact on the spatial-temporal distribution changes of FVC. Groundwater depth is strongly negatively correlated with FVC in both wetlands, while surface runoff is positively correlated outside high FVC areas. The vegetation changes in the eastern part of Lake Sugan Wetland may be influenced by changes in surface runoff. Topographical variables exert control over soil moisture content and accumulated temperature, leading to strong spatial heterogeneity between wetland FVC and topographical variables. Climate variables have the least impact, with rainfall exhibiting a negative correlation with FVC in areas with high superficial soil salinity.

(4) A FVC (Fraction of Vegetation Cover) dynamic response model was constructed for the Sugan Lake Wetland and Nuhutu Wetland based on SVM (Support Vector Machine) and RF (Random Forest) algorithms. It was found that compared to SVM, the FVC dynamic response model based on RF algorithm exhibited better simulation performance, and the accuracy of the model was further improved when considering the prior information of GTWR (Geographically and Temporally Weighted Regression). The minimum mean square error of the G-RF model simulation results in Sugan Lake and Nuhutu wetlands are 0.083 and 0.047, respectively. In Lake Sugan Wetland, the simulated distribution of high FVC shifts eastward, with a significant increase of medium FVC. In Nuhutu Wetland, partial areas with extremely low FVC gradually transition to low FVC vegetation. The temporal and spatial changes in simulated FVC distribution correspond well with the actual observation process, indicating the model's high reliability. The results of feature variable importance scores indicate that slope features have the greatest impact on prediction results, followed by groundwater depth variables.

(5) A comprehensive simulation model integrating groundwater circulation, climate change, and FVC dynamic response was established to predict the dynamic changes in groundwater circulation and the evolutionary trends of FVC distribution. The GCM temperature data exhibited high accuracy in the study area, while precipitation data tended to underestimate during the rainy season. After calibration, the correlation coefficients of temperature and precipitation data in the study area were 0.994 and 0.790 respectively. Under the SSP1-2.6, SSP2-45, and SSP3-7.0 long-term scenarios, the integrated simulation model revealed an increasing trend in upstream runoff at the basin, with runoff volumes increasing by 25.9%, 57.8%, and 138.8% respectively; groundwater levels rising at rates of 4.08 cm·a^-1, 5.18 cm·a^-1, and 7.02 cm·a^-1 in the whole basin; the submerged area of the Sugan Lake Wetland was 25 km², 33 km², and 240 km², respectively. The average FVC of wetland vegetation increased with increasing emission standards, reaching maximum values of 0.415 and 0.313 for Sugan Lake and Nuhutu wetlands, respectively, in the long term under SSP3-7.0. Vegetation in extremely low FVC grades within the region showed the most significant response to climate change, exhibiting a significant decreasing trend. The eastern part of the Sugan Lake wetland will develop into a core area for vegetation ecology, with predominantly medium or medium-high FVC vegetation in the future. However, the future distribution area of high FVC vegetation is limited, with an unclear increasing trend in Nuhutu Wetland and a decreasing trend in Sugan Lake Wetland. After implementing the water diversion plan, the rate of increase in groundwater level and average FVC of wetlands slowed down.

In summary, this study is based on the ecological hydrology research of the Sugan Lake Basin with limited observational data. It quantitatively evaluates the dynamic response mechanism of groundwater and vegetation under changing environments, which helps to better balance local oasis economy and ecological security. It provides scientific theoretical basis for basin comprehensive management, water diversion projects, and other policies.