基于Google Earth Engine和Sentinel-2 MSI数据的遥感水质参数反演

作者	单位
宋挺	江苏省无锡环境监测中心, 江苏无锡 214000
徐超	江苏省无锡环境监测中心, 江苏无锡 214000
严飞	江苏省无锡环境监测中心, 江苏无锡 214000
张军毅*	江苏省无锡环境监测中心, 江苏无锡 214000

通讯作者:张军毅* 江苏省无锡环境监测中心, 江苏无锡 214000

摘要点击次数: 396

全文下载次数: 662

中文摘要:

卫星遥感凭借大范围覆盖、高时效和低成本等优势,在水质监测方面相较于地面监测更具经济性和实用性。然而,传统遥感处理方式存在数据下载量大、存储要求高、处理时间长等问题,限制了该技术的推广应用。将相关处理程序集成至Google Earth Engine云平台,构建了一种基于Sentinel-2 MSI数据和随机森林机器学习算法的水质参数遥感监测流程。该流程涵盖高锰酸盐指数、总氮、总磷、蓝藻密度、浊度、电导率6项参数,并具备人机交互功能。以2023年3月14日数据为例,利用该流程分析太湖流域水质参数的空间分布特征。同时,对2022年太湖流域水质参数的逐月变化及其相互关系进行了研究,并评估了反演精度。研究结果表明,随机森林算法在水质参数反演中表现良好。其中,浊度的决定系数最高,为0.603 6,而蓝藻密度的决定系数相对较低,仅为0.199 4。电导率的平均绝对百分比误差最低,为14.11%,而蓝藻密度的平均绝对百分比误差最高,达118.68%。总体而言,该算法在不同数值区间均能有效反演水质参数,具备业务化应用的潜力。此外,2022年太湖流域水质参数的月度变化趋势存在差异。其中,蓝藻密度的变化幅度最大,达177.4%,而电导率的变化幅度最小,仅为8.8%。从蓝藻密度与其他水质参数的关系来看,其与高锰酸盐指数呈极显著正相关,与总磷呈显著正相关,而与总氮、电导率呈显著负相关,与浊度无明显相关性。该遥感监测流程可实现水质长期监测的自动化与一体化,特别适用于内陆水体尤其是小型湖泊和河流的水环境管理与污染溯源调查,可为水质监测提供技术支持。

英文摘要:

Satellite remote sensing,characterized by its expansive coverage,frequent observations,and cost-effectiveness,surpasses terrestrial monitoring in feasibility and economy.Yet,conventional remote sensing methods grapple with challenges,including excessive data download volumes,prolonged processing durations,and significant storage demands.These limitations obstruct the extended time-series analysis of water quality through remote sensing.To address these issues,specific processing applications were integrated into the Google Earth Engine cloud platform,devising a remote sensing monitoring workflow for water quality parameters based on Sentinel-2 MSI data and the Random Forest machine learning algorithm.This workflow covers six water quality metrics:permanganate index,total nitrogen,total phosphorus,cyanobacterial density,turbidity,and conductivity,and features human-computer interaction capabilities.This model was used to analyze the spatial distribution of these parameters in the Taihu Lake Basin on March 14,2023.Furthermore,the monthly fluctuations and interdependencies of these parameters throughout 2022 were investigated,and each parameter's inversion accuracy was evaluated.The outcomes revealed that the Random Forest algorithm effectively inverted water quality metrics.Specifically,turbidity recorded the highest R² value of 0.6036,whereas cyanobacterial density presented a lower R² of 0.1994.Conductivity demonstrated the minimal MAPE at 14.11%,contrasted by cyanobacterial density's peak at 118.68%.Notably,the Random Forest algorithm effectively inverted water quality parameters across different concentration ranges,underscoring its commercial viability.Throughout 2022,the Taihu Lake Basin exhibited diverse monthly distribution patterns for these metrics.Cyanobacterial density manifested the most pronounced fluctuation,with an amplitude of 177.4%,while conductivity registered a mere 8.8% variation.Assessing interdependencies revealed cyanobacterial density's strong positive association with the permanganate index and a significant positive relationship with total phosphorus.In contrast,it showed notable negative correlations with total nitrogen and conductivity,remaining unrelated to turbidity.In conclusion,the developed workflow offers a streamlined,comprehensive solution for extended remote sensing-based water quality surveillance.This methodology holds promise for efficiently monitoring inland aquatic systems,particularly minor lakes and rivers,fortifying efforts in water management and tracing pollution sources.

查看全文查看/发表评论下载PDF阅读器