2022 年克鲁伦河流域 10 米分辨率植被覆盖度月度数据集

精确获取流域范围内的植被覆盖度(Fractional Vegetable Cover,FVC)，对于深入研究流域生态环境、湿地健康状况及其生态保护策略具有至关重要的意义。克鲁伦河流域是横跨中蒙边境的重要生态区，具有很高的生物多样性价值，对支撑并维护该区域生态系统平衡具有重要作用，鉴于此，本数据集以克鲁伦河流域作为研究区，基于 10 米空问分辨率的Sentine1-2 多光谱遥感影像获取高精度植被覆盖度数据集，为流域生态环境保护提供数据支撑。为克服像元二分法、线性回归方法、随机森林回归模型等传统植被覆盖度反演方法在光谱特征间细微差异挖掘有效性与高维特征问复杂非线性关系发现不足等问题。为更精准估算该流域植被覆盖度，论文比较了基于深度学习的双向长短时记忆网络(Bidirectional longShort-Term memory,BiLSTM)模型、随机森林回归、多层感知机与LSTM 四个模型的性能以确定数据处理方法。所用特征数据以 Sentine1-2 多光谱数据为基础，综合光谱指数与高程数据，所反映植被相关信息包括叶绿素含量、水分状况以及地形地貌等。该特征数据集将进一步划分为训练集和测试集。比较结果表明，BiLSTM的 R2和 RNSE分别为 0,716 和 0.103，综合性能最优。论文基于该模型生成了 2022年克鲁伦流域的月度植被覆盖度数据集,包括 12 个月的克鲁伦河流域植被覆盖度反演结果组成，全部数据已经完成拼接和掩膜提取等操作。该数据集可用于评价克鲁伦河流域地表植被生长状况和生态系统健康状况，并为相关流域的生态保护研究提供支持。

Accurate acquisition of Fractional Vegetation Cover (FVC) within a river basin is critically important for in-depth studies of the basin’s ecological environment, wetland health status, and ecological protection strategies. The Kherlen River Basin is an important ecological area spanning the China-Mongolia border, with high biodiversity value, and plays a significant role in supporting and maintaining the regional ecosystem balance. In view of this, this dataset takes the Kherlen River Basin as the study area, and acquires a high-precision FVC dataset based on Sentinel-1 and Sentinel-2 multispectral remote sensing images with 10-meter spatial resolution, to provide data support for river basin ecological environment protection. To address the shortcomings of traditional FVC inversion methods such as the dimidiate pixel model, linear regression, and random forest regression, including their insufficient effectiveness in mining subtle differences between spectral features and their poor ability to capture complex nonlinear relationships among high-dimensional features, this study compared the performance of four deep learning-based models: Bidirectional Long Short-Term Memory (BiLSTM), random forest regression, Multi-Layer Perceptron (MLP), and Long Short-Term Memory (LSTM), to determine the optimal data processing method. The feature dataset used is based on Sentinel-1 and Sentinel-2 multispectral data, combined with spectral indices and elevation data, which captures vegetation-related information including chlorophyll content, water status, and topography, among other factors. This feature dataset is further divided into a training set and a test set. The comparison results show that the BiLSTM model achieved the best comprehensive performance, with an R² of 0.716 and an RNSE of 0.103. Based on this optimal model, this study generated a monthly FVC dataset for the Kherlen River Basin in 2022, which consists of FVC inversion results for 12 months. All data have undergone operations such as mosaicking and mask extraction. This dataset can be used to evaluate the surface vegetation growth status and ecosystem health status of the Kherlen River Basin, and provide support for ecological protection research of relevant river basins.