水是人类不可或缺的宝贵资源。我国山地较多，在偏远山区以及寒冷的高原中分布有大量的水资源。然而，这些地区存在着复杂的地形地貌，给山区水体的提取和监测造成了巨大的困难。通过遥感影像及时获取山区水体的动态变化，进而采取有效措施预防和化解自然灾害，对于守护绿水青山和建设水生态文明具有重要意义。针对山区水体因地形复杂、河道狭窄、易受山体阴影干扰等因素导致的提取难、精度低的问题，本文以Sentinel-2A卫星影像为主要研究数据，以水土流失严重的洮河下游所在的黄土高原地区为研究区，对山区水体提取方法开展研究，探讨Sentinel-2A数据在山区水体提取中的潜力，对比分析水体指数法和离散粒子群算法提取山区水体的有效性，主要研究内容及结论如下：

（1）适用于山区的水体指数构建。针对研究区水体提取精度易受山体阴影等影响的问题，通过分析各典型地物的平均反射率并考虑水体提取中易受影响的低反射率地表信息，利用蕴含不同水质水体信息的多个波段，构建了一种新的水体指数Water Index(WI2021)。选用黄河流域、长江流域等不同环境的4个测试点验证，用Sentinel-2A影像数据实验结果表明，与已有的7种水体指数相比，WI2021以平均总体精度、Kappa系数、错分和漏提误差分别为96.78%、0.9336、3.61%和4.11%，总体上较其他水体指数具有一定的优势，能够稳定地提取地表水且对山体阴影有较好的抑制作用。

（2）基于离散粒子群算法的山区水体提取。通过详细介绍光谱匹配耦合离散粒子群算法（SMDPSO）和最大熵模型耦合离散粒子群算法（MEDPSO）两种算法模型的构建和山区水体提取过程，对比两种离散粒子群算法模型在研究区的提取效果，分析两种方法对山区水体提取的优势和不足。实验发现，SMDPSO分类方法简单，所需参数较少，不需要人工干预，但运行时间较长；MEDPSO能够更完整的提取山区细小支流，但是人工干预性较大，不同的环境约束条件可能产生效果不同的水体提取结果。将Sentinel-2A影像数据应用于SMDPSO和MEDPSO两种算法模型的构建，验证了离散粒子群算法在Sentinel-2A数据中的可行性，进一步推广了离散粒子群算法在遥感领域提取地物的应用。

（3）山区水体提取结果及精度评价。通过对比分析WI2021水体指数法和两种离散粒子群算法在研究区中对山区水体的提取效果并进行精度评价发现，WI2021、SMDPSO和MEDPSO对研究区水体均有较好的提取效果，总体精度均达到95%以上，为黄土高原中的山区水体提取方法的选择提供了参考。为验证三种方法对不同特征山区水体的适应性和不同空间分辨率影像对山区水体提取精度的影响，分别选取4种不同山区环境的测试区和采用Landsat 8 OLI影像数据进行实验。结果表明，WI2021能稳定地剔除山体阴影的干扰；MEDPSO的平均Kappa系数和总体精度分别为0.88和94.49%，高于其他2种方法，对各种复杂环境的水体提取表现出更强的适应性和稳定性；SMDPSO不适用于含有大量冰雪的高原山区，在没有冰雪干扰的地区对山区水体具有更好的提取效果。在混淆矩阵得到的各类精度指标中，Sentinel-2A均高于Landsat 8更适合于开展山区水体提取方法的研究。

Water is an indispensable and precious resource for human beings. There are many mountains in my country, and there are a lot of water resources distributed in remote mountainous areas and cold plateaus. However, these areas are often inaccessible and have complex topography. In addition, the characteristics of mountain rivers, plateau lakes and other water bodies have created huge difficulties for the extraction and monitoring of mountain water bodies. Identifying and monitoring mountain water bodies through high spatial resolution image data, obtaining dynamic changes in mountain water bodies in a timely manner, and then taking effective measures to prevent and resolve natural disasters is of great significance for protecting lucid waters and lush mountains and building a water ecological civilization. Sentinel-2A satellite image data has significant advantages such as high spatial resolution and rich band information, and is an important image data suitable for monitoring mountain water bodies. Aiming at the problems of difficult extraction and low accuracy of mountain water bodies due to complex terrain, narrow river channels, and easy interference from mountain shadows, this paper takes the high-resolution images of Sentinel-2A satellite as the main research data, and takes the Loess Plateau region where the lower reaches of the Tao River with serious soil erosion are located as the study area. The research on the extraction methods of mountain water bodies was carried out, and the potential of Sentinel-2A data in the extraction of mountain water bodies was discussed. The main research contents and conclusions are as follows:

(1) Construction of water body index suitable for mountainous areas. Aiming at the problem that the extraction accuracy of water body in the study area is easily affected by suspended sediment and mountain shadows, the Yellow River Basin and Liujiaxia Reservoir near Lanzhou City are used as experimental areas to analyze the average reflectivity of each typical feature and consider the vulnerable low reflectivity information of water body extraction. A new water index, Water Index (WI2021), was constructed by using multiple bands containing information of different water quality. Four test points in the Yellow River Basin, the Yangtze River Basin, the urban water body and the estuary area are selected. The experimental results of Sentinel-2A image data show that, compared with the existing 7 water body indices, the average overall accuracy, Kappa coefficient, misclassification and omission errors of WI2021 are 96.78%, 0.9336, 3.61% and 4.11%, respectively. Overall, WI2021 has certain advantages over other water body indices. It can stably extract surface water and has a good inhibitory effect on and mountain shadows.

(2) Mountain water body extraction based on discrete particle swarm optimization. By introducing in detail the construction process of the two algorithm models, the spectral matching coupled discrete particle swarm optimization (SMDPSO) and the maximum entropy model coupled discrete particle swarm optimization (MEDPSO), and the extraction process of mountain water bodies, the extraction effects of the two discrete particle swarm optimization models were compared in the study area. The advantages and disadvantages of the two algorithm models in the extraction of mountainous water bodies were analyzed. The experiment found that the SMDPSO classification method is simple, requires less parameters, does not require manual intervention, but has a long running time; MEDPSO can more completely extract small tributaries in mountainous areas, but it requires more manual intervention. Different environmental constraints may produce different water extraction results. The Sentinel-2A image data was applied to the construction of two algorithm models of SMDPSO and MEDPSO, which verified the feasibility of discrete particle swarm optimization in Sentinel-2A data, and further promoted the application of discrete particle swarm optimization in the field of remote sensing extraction of ground objects.

(3) Extraction results and precision evaluation of mountain water bodies.By comparing and analyzing the extraction effects of WI2021 water body index method and two discrete particle swarm optimization algorithms on mountain water bodies in the study area and evaluating the accuracy, it is found that the constructed WI2021 water body index method and the two discrete particle swarm algorithms have better performance on mountain water bodies. The overall accuracy reached more than 95%, which provided a reference for the selection of water extraction methods in mountainous areas in the Loess Plateau. In order to verify the adaptability of the three methods to mountain water bodies with different characteristics and the impact of different spatial resolution images on the extraction accuracy of mountain water bodies, four test areas with different mountain environments and Landsat 8 oli image data were selected for experiments. The experimental results show that WI2021 can stably remove the interference of hill shadows. The average Kappa coefficient and overall accuracy of MEDPSO are 0.88 and 94.49%, respectively, which are higher than the other two methods, showing stronger adaptability and stability to water extraction in various complex environments.The SMDPSO method is not suitable for plateau mountainous areas with a lot of ice and snow, but it can have better extraction effect on mountain water in areas without ice and snow interference.Sentinel-2A is higher than Landsat 8 in all kinds of precision indexes obtained by confusion matrix, which is more suitable for the study of water extraction methods in mountainous areas.