基于改进像元三分模型的植被覆盖度提取及时空变化分析  

Vegetation Fractional Cover Extraction and Spatiotemporal Variation Analysis Based on Improved Normalized Difference Vegetation Index(NDVI)and Dry Fuel Index(DFI)Model

作　　者：张凡 仇天昊 李欣悦 张姝茵 徐超[1] 谢治国 Zhang Fan;Qiu Tianhao;Li Xinyue;Zhang Shuyin;Xu Chao;Xie Zhiguo(Northwest Agriculture and Forestry University,Yangling 712100;Shaanxi Academy of Forestry,Xi’an 710016)

机构地区：[1]西北农林科技大学,杨凌712100 [2]陕西省林业科学院,西安710016

出　　处：《林业科学》2024年第12期13-26,共14页Scientia Silvae Sinicae

基　　金：陕西省林业科技创新重点专项(SXLK2023-02-3);陕西省林业科技创新重点专项(SXLK2022-02-7)。

摘　　要：【目的】引入DBSCAN聚类算法和QuickHull凸包检测算法,提出一种自适应端元特征值提取(AEEE)算法,解决像元三分模型中纯净像元指数结合二维散点图(PPI-2DSP)算法依赖人工圈选特征像元候选区的问题,利用AEEE算法评估神木市的光合植被覆盖度(f_(PV))、非光合植被覆盖度(f_(NPV))和裸土覆盖度(f_(BS)),并分析其时空变化,验证算法的有效性,为该地区生态环境评价和植被覆盖度变化规律研究提供参考。【方法】以Landsat系列卫星遥感影像为数据源,首先对遥感数据进行预处理,然后计算像元的归一化植被指数(NDVI)和干枯燃料指数(DFI),通过以下4个步骤获取特征像元候选区:1)应用随机采样模块减少数据规模;2)采用DBSCAN算法聚类,去除离群数据,得到最大簇;3)利用QuickHull算法计算凸包,构建特征三角形边界;4)计算由凸包点集中3个点所构成的最大面积三角形顶点,分别以3个顶点为中心,取端点阈值(θ)范围内区域作为特征像元候选区。预处理的影像经最小噪声分离变换减少计算量后,采用PPI算法计算纯净像元指数,利用特征像元候选区提取纯净像元指数大于5的纯净像元,将这些像元的NDVI和DFI算术平均值作为端元特征值代入像元三分模型计算f_(PV)、f_(NPV)和f_(BS),并分析其时空变化。【结果】AEEE算法提取的神木市2000—2022年端元特征值与PPI-2DSP算法选取的端元特征值相近,相对误差平均值约7.35%;将其应用于像元三分模型,估算得到神木市的f_(PV)和f_(NPV),与传统方法相比年平均误差分别为4.79%和5.05%,符合精度要求。在时间层面上,2000—2022年神木市的f_(PV)和f_(NPV)总体呈波动增长趋势,分别以年平均0.52%和0.22%的速率增长;在空间层面上,2000—2022年神木市的f_(PV)和f_(NPV)呈东南增长快速、西北增长缓慢的趋势,其中f_(PV)主要以增加(39.8%)和基本不变(28.0%)2种变化强度等级为主。【结论】AEEE�【Objective】By introducing the DBSCAN clustering algorithm and QuickHull convex hull detection algorithm,an adaptive endmember eigenvalue extraction(AEEE)algorithm is proposed to address the issue of manual selection of feature pixel candidate areas in the pixel ternary model combined with the pure pixel index-2D scatter plot(PPI-2DSP)algorithm.The AEEE algorithm is utilized to assess the photosynthetic vegetation coverage(f_(PV)),non-photosynthetic vegetation coverage(f_(NPV)),and bare soil coverage(f_(BS))in Shenmu city and analyze their spatiotemporal variations.The effectiveness of the algorithm is verified,providing reference for the evaluation of the ecological environment and the study of vegetation fractional cover change patterns in the region.【Method】Using Landsat series satellite remote sensing images as the data source,the remote sensing data is preprocessed first.Then,the normalized difference vegetation index(NDVI)and dry fuel index(DFI)of pixels are calculated.Feature pixel candidate areas are obtained through the following 4 steps:1)Reducing the data volume using a random sampling module;2)Clustering with the DBSCAN algorithm to remove outlier data and identify the largest cluster;3)Computing the convex hull using the QuickHull algorithm to construct the boundaries of feature triangles;4)Calculating the vertices of the largest area triangle formed by three points within the convex hull point set.Regions within a specified range centered at the three vertices are selected as feature pixel candidate areas with a vertex threshold(θ).After reducing the computational complexity through the Minimum Noise Fraction transformation of preprocessed images,the pure pixel index is calculated using the PPI algorithm.Pure pixels with a pure pixel index greater than 5 are extracted using feature pixel candidate areas.The arithmetic mean values of the NDVI and DFI for these pixels are used as endmember eigenvalues into the pixel ternary model to calculate f_(PV),f_(NPV),and f_(BS),followed by an analysis o

关 键 词：像元三分模型 植被覆盖度 LANDSAT 自适应端元特征值提取算法 陕西省神木市 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置] Q948[自动化与计算机技术—控制科学与工程]