荒漠植被覆盖度遥感提取研究进展  

Research Progress on Extraction of Fractional Vegetation Cover in Desert Area Based on Remote Sensing

作　　者：韩婕 曹晓明[1,2] HAN Jie;CAO Xiaoming(Institute of Ecological Protection and Restoration,Chinese Academy of Forestry,Beijing 100091,China;Institute of Desertification Studies,Chinese Academy of Forest,Beijing 100091,China)

机构地区：[1]中国林业科学研究院生态保护与修复研究所,北京100091 [2]中国林业科学研究院荒漠化研究所,北京100091

出　　处：《环境科学研究》2024年第10期2288-2298,共11页Research of Environmental Sciences

基　　金：新疆第三次科学考察项目(No.2021xjkk0304);国家自然科学基金项目(No.41971398);科技基础资源调查专项项目(No.2023FY100703)。

摘　　要：植被覆盖度(Fractional Vegetation Cover,FVC)是植被生长状况监测、荒漠化监测与评价等方面的重要指标,但荒漠植被的生理和光谱特征均有别于普通健康绿色植被,为其FVC遥感提取带来了诸多难点。本文从数据、方法及应用等方面总结分析了提取荒漠地区FVC各类方法的应用进展。结果表明:①在荒漠区应用的遥感数据源包括多光谱影像、高光谱影像、雷达影像以及无人机高分影像等,融合多源遥感数据、结合各方法可明显提高荒漠FVC的估算精度。②模型上,经验回归模型与像元分解模型应用较广泛,但经验回归模型过度简化了FVC与遥感数据之间的复杂关系,像元分解模型则忽略了荒漠植被显著的空间异质性,故二者都无法充分考虑遥感数据的时空特征和地物背景的影响。近年来,机器学习与数据挖掘算法广泛用于荒漠FVC遥感提取研究中,此类算法能够处理大规模数据,使处理效率大幅提高,并在一定程度上减少人为干扰等因素。然而,此类算法对地表环境变化的复杂性和非线性关系建模能力有限,可能面临样本不平衡、特征选择困难以及模型泛化能力不足等挑战。未来荒漠FVC遥感提取研究将是一个基于多尺度、多源数据、融合多种方法实现大尺度、高精度且时序兼具的估算模式,这种新形式可有效弥补单一卫星遥感影像数据在时间、空间及光谱分辨率信息方面的不足,但荒漠植被特殊的生理和光谱特征、新型的遥感指标和理论策略以及算法的改进程度有待进一步探索。Fractional Vegetation Cover(FVC)is a crucial indicator for monitoring vegetation growth status,desertification monitoring and evaluation,among other aspects.However,due to the differences in physiological and spectral characteristics between desert vegetation and typical healthy green vegetation,extracting FVC information through remote sensing faces many challenges.This paper summarizes and analyzes the progress in the application of various methods for extracting FVC in desert areas from the aspects of data,methods,and applications.The research results show that:(1)Remote sensing data sources applied in desert areas include multispectral imagery,hyperspectral imagery,radar imagery,and high-resolution UAV imagery.Integrating multi-source remote sensing data and combining multiple methods can significantly improve the estimation accuracy of desert vegetation coverage.(2)The empirical regression model and the pixel decomposition model are the most widest used models;however,the empirical regression model oversimplifies the complex relationship between FVC and remote sensing information,while the pixel decomposition model ignores the spatial heterogeneity of vegetation,both of which fail to fully consider the spatiotemporal characteristics of background remote sensing data.In recent years,machine learning and data mining algorithms have been widely used in desert vegetation cover extraction.This type of algorithm can process large-scale data,significantly improve processing efficiency,and reduce human interference and other factors to certain extent.However,they have limited modeling capabilities for the complexity and non-linear relationships of surface environmental changes,and may face challenges such as sample imbalance,difficulty in feature selection,and insufficient model generalization capabilities.Future desert FVC remote sensing extraction research will adopt multi-scale,multi-source data,and multi-method fusion approach to achieve large-scale,high-precision and time-sequential estimation models.This new f

关 键 词：荒漠植被 植被覆盖度 遥感 模型算法 

分 类 号：X87[环境科学与工程—环境工程] TP79[自动化与计算机技术—检测技术与自动化装置]