摘要
遥感水深监测能快速、经济地获取大范围水域的水深信息,为水资源管理、航道维护、生态环境保护等提供重要数据支持。然而,目前的水深反演研究主要在沿海水域,内陆湖泊的水深反演研究相对较少。本文以池州市东至县肖思湖为研究区,对Sentinel-2多光谱数据和实测水深数据进行相关性分析,并利用传统的单波段模型、多波段模型和对数比值模型确定最佳水深反演波段组合,分别建立BP神经网络、随机森林和支持向量机三种机器学习模型进行水深反演,并对反演结果进行精度评价。结果表明:以Sentinel-2影像的B2、B3、B4、B8波段和不同波段之间的比值共计10个变量为自变量,所构建的随机森林模型在湖泊水域的反演效果最佳,反演得到的肖思湖水深分布与实测水深基本一致,但反演精度随着水深的增大有所下降,且受水中物质分布异质性影响分布不连续。
Remote sensing water depth monitoring enables rapid and cost-effective acquisition of large-scale water body depth data,supporting water resource management,channel maintenance,and ecological protection.While current research predominantly focuses on coastal waters,inland lake depth inversion remains underexplored.This study addresses this gap by analyzing the correlation between Sentinel-2 multispectral data and measured depth data in Xiaosi Lake,Dongzhi County,Chizhou City.Traditional models(single-band,multi-band,and logarithmic ratio)were employed to identify optimal band combinations,followed by the development of three machine learning models—BP neural network,random forest,and support vector machine—for depth inversion.Results demonstrate that the random forest model,utilizing 10 variables derived from Sentinel-2 bands(B2,B3,B4,B8)and their interband ratios,achieves the highest inversion accuracy in lake waters.The inverted depth distribution aligns broadly with measured data,though accuracy declines with increasing depth and exhibits discontinuities due to water column heterogeneity.
作者
张树衡
赵萍
张辰
刘兴国
孟显卓
ZHANG Shuheng;ZHAO Ping;ZHANG Chen;LIU Xingguo;MENG Xianzhuo(Institute of Geophysical and Geochemical Exploration Technology of Anhui Province,Hefei,Anhui 230022,China;School of Resources and Environmental Engineering,Hefei University of Technology,Hefei,Anhui 230009,China;Institute of Geological Surveying and Mapping Technology of Anhui Province,Hefei,Anhui 230022,China)
出处
《安徽地质》
2025年第4期366-370,共5页
Geology of Anhui
基金
安徽省水资源基础调查项目“湖泊水储存量调查”(皖自然资调[2024]2号)资助。
