摘要
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米空间分辨率的Sentinel-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间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 Long Short-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Sentinel-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RMSE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。
Precisely obtaining the Fractional Vegetable Cover(FVC)at the river basin scale is of immense importance for delving into the ecological environment,wetland health,and ecological conservation strategies within watersheds.The Kherlen River Basin is an important ecological area across the border between China and Mongolia.It has high biodiversity and is essential for supporting and maintaining the balance of ecosystems in the region.Thus,this dataset focuses on the Kherlen River Basin,leveraging Sentinel-2 multispectral remote sensing imagery with a spatial resolution of 10 m to derive FVC with high precision.The dataset provides vegetable cover data to support the ecological protection of the Kherlen River Basin.In order to overcome the problem,traditional vegetation coverage inversion methods,such as pixel dichotomy,linear regression,and random forest regression models,could be more effective in mining subtle differences between spectral features and finding complex nonlinear relationships between high-dimensional features.To estimate the vegetation coverage more accurately in the watershed,this paper compares the performance of four models:the Bidirectional Long Short-Term Memory(BiLSTM)model based on deep learning,Random Forest Regression,Multilayer Perceptron,and LSTM,to determine the optimal data processing method.The feature data used are based on Sentinel-2 multispectral data,integrating spectral indices and elevation data.The vegetation-related information reflected includes chlorophyll content,moisture status,and topography.This feature dataset is further divided into training and testing sets.The comparison results show that BiLSTM achieved an R2 of 0.716 and an RMSE of 0.103,indicating the best overall performance.This model generated a monthly vegetation coverage dataset for the Kherlen River Basin in 2022,comprising vegetation coverage inversion results for 12 months.All data have undergone operations such as mosaicking and mask extraction.This dataset can assess the vegetation growth status and ecosystem health of the Kherlen River Basin and support ecological protection research in related watersheds.
作者
牛博文
冯权泷
张毓
高秉博
SUKHBAATAR Chinzorig
冯爱萍
杨建宇
NIU BoWen;FENG QuanLong;ZHANG Yu;GAO BingBo;SUKHBAATAR Chinzorig;FENG AiPing;YANG JianYu(College of Land Science and Technology,China Agricultural University,Beijing 100093,China;Institute of Geography and Geoecology,Mongolian Academy of Sciences,Ulaanbaatar 15170,Mongolia;Ministry of Ecology and Environment Center for Satellite Application on Ecology and Environment,Beijing 100094,China)
出处
《农业大数据学报》
2025年第1期59-68,共10页
Journal of Agricultural Big Data
基金
国家重点研发计划项目克鲁伦河流域面源污染遥感监测与评估技术研发(2021YFE0102300)。
