为分析三屯河流域植被覆盖时空变化特征,基于Landsat卫星遥感数据、像元二分模型、图像差值法及变异系数等方法手段对2000—2020年20 a间植被覆盖时空分布特征、趋势及稳定性进行分析。三屯河流域植被覆盖度呈现逐年增长趋势,特别是2012—2015年的年均增长速度为4.93%/(10 a);植被覆盖分布在空间上表现出两端高、中间低的规律性;三屯河流域植被覆盖改善区域占比39.58%,整体呈现改善的趋势;三屯河流域内植被覆盖度较为稳定,区域绝大部分处于弱变异与中等变异。三屯河流域植被覆盖演化受到自然因素和人类因素影响,自然因素的影响主要体现在温度与降水量的改变,人类活动因素的影响主要体现在土地利用类型的改变上。研究成果可为促进三屯河流域生态恢复、经济发展提供科学依据。

To analyze the spatiotemporal characteristics of vegetation cover in the Santun River Basin, a comprehensive study was conducted using a combination of Landsat satellite remote sensing data, pixel-based binary models, image differencing methods, and coefficient of variation. The period from 2000 to 2020, spanning 20 years, was examined to assess the distribution patterns, trends, and stability of vegetation cover. The results reveal a consistent annual increase in vegetation cover within the Santun River Basin, with a notable average growth rate of 4.93% per decade from 2012 to 2015. Spatially, the distribution of vegetation cover follows a distinctive pattern, with higher values observed at the two ends and lower values in the central regions. Additionally, approximately 39.58% of the Santun River Basin exhibits improved vegetation cover, indicating an overall positive trend. Notably, the majority of the basin displays relatively stable vegetation cover, characterized by minor to moderate variations.The evolution of vegetation cover in the Santun River Basin is influenced by a combination of natural and human factors. Changes in temperature and precipitation patterns constitute the primary natural influences, while alterations in land use types serve as the main drivers of change induced by human activities. These findings provide scientific foundation for promoting ecological restoration and facilitating sustainable economic development within the Santun River Basin.

在灌区用水管理中，灌溉面积及其空间分布信息非常重要。传统的灌溉面积提取手段耗费大量的人力物力和时间，已经不能满足灌区的现代化管理。自遥感技术应用于灌溉面积提取以来，经过几十年的发展，已经有很多的研究成果。但现今在应用于灌溉面积提取的遥感技术中，前沿的方法多数采用多个传感器数据或长时间序列的数据，且往往针对某一特定的区域，很难具体的应用在实际的灌区工作中。为了在灌区管理的实际应用中准确高效地提取灌溉面积和分布，开展了一种基于光学卫星多时相差值数据的神经网络算法的灌溉面积提取技术研究。以山东省淄博市桓台县的试验田为研究区域，首先利用随机森林对Sentinel-2卫星数据的所有波段以及一些与土壤水含量以及植被相关的指数进行重要性排序，不同地区的地情下重要性排序结果也不同，所以利用重要性排序可以快速的获取适合此地区的波段以及指数的组合。选取重要性较高的波段或指数作为神经网络模型输入层进行灌溉面积提取。然后根据实际样本田的数据对提取结果进行检验，所得到的总体灌溉面积提取精度达到了76.7%。Kappa系数为0.74。此外，对研究区域进行植被覆盖度分级，并分析了在不同植被覆盖度下的灌溉面积提取结果精度变化。其中，在中等和较高的植被覆盖度地区具有更高的精度。研究区大部分地区为农业地区，作物以冬小麦和夏玉米为主，使用数据为3月中下旬卫星影像，研究区此时期植被覆盖度较高，符合在此地情下进行灌溉面积提取在中等和较高植被覆盖度地区具有更高精度的结果。基于神经网络的多时相光学卫星数据差值提取灌溉面积研究可以在不同地区的地情下通过重要性排序获取适合该研究区波段的组合，得到更高的灌溉面积提取结果精度。

Irrigation area and its spatial distribution information are very important in water management in irrigated areas. The traditional method of extracting irrigated area consumes a lot of manpower and material resources and time, which can not satisfy the modern management of irrigated area. Since the application of remote sensing technology in the study of irrigation area extraction, after decades of development, there have been a lot of research results. However, in the current remote sensing technology applied in irrigation area extraction, most of the cutting-edge methods adopt multiple sensor data or long time series data, which are often targeted at a specific area, and it is difficult to be specifically applied in the actual irrigation area. In order to extract irrigation area and distribution accurately and efficiently in the practical application of irrigation area management, a neural network algorithm based on multi-temporal difference data of optical satellites was developed in this paper. The experimental field in Huantai County, Zibo City, Shandong Province was taken as the study area. Firstly, random forest was used to sort the importance of all bands of Sentinel-2 satellite data and some indices related to soil water content and vegetation. The importance ranking results were different in different geographical conditions, so the combination of bands and indices suitable for this region could be quickly obtained by using importance ranking. In this study, the band or index with high importance was selected as the input layer of the neural network model to extract the irrigation area, and then the extraction results were tested according to the actual sample data, and the overall extraction accuracy of the irrigation area reached 76.7%. The Kappa coefficient was 0.74. In addition, the vegetation coverage of the study area was graded, and the precision changes of irrigation area extraction results under different vegetation coverage were analyzed. Among them, the accuracy is higher in medium and high vegetation coverage areas. Most of the study area is an agricultural area, with winter wheat and summer corn as the main crops. The data used are satellite images in mid-to-late March. The vegetation coverage of the study area is relatively high in this period, which accords with the result that the extraction of irrigated area has higher accuracy in the areas with medium and high vegetation coverage under the situation in this area. The research results of multi-temporal optical satellite data difference extraction of irrigation area proposed in this paper based on neural network can obtain the combination of bands suitable for the study area through importance ranking under different geographical conditions, and obtain higher accuracy of irrigation area extraction results.