基于机器学习的东湖富营养化研究

Assessment of eutrophication status in East Lake based on machine learning methods

利用遥感技术建立影像波段反射率与地面监测水质参数的定量反演模型,能实现高效率、大尺度湖泊富营养化水平的监测。机器学习通过高度的非线性映射,能很好地利用已知信息,模拟复杂因素之间的关系。以武汉东湖为例,基于资源一号02C卫星影像,利用K-近邻法、支持向量机、随机森林、梯度提升决策树、人工神经网络等5种经典的机器学习算法建立了叶绿素a(Chl-a)、透明度(SD)、总磷(TP)、总氮(TN)、高锰酸盐指数(COD_(Mn))5个水质参数与影像反射率间的定量反演模型,并采用综合营养状态法对东湖富营养化程度进行评价。以2014年11月22个采样点和2015年1月23个采样点的数据为基准,与基于机器学习算法训练水质参数建模对比,富营养化分级正确率为分别为95.5%和82.6%;以武汉市环境保护局数据为基准,对比反演的东湖各子湖营养等级,正确率均为71.4%。

Remote sensing is a powerful and efficient tool to monitor water eutrophication status at large scale by developing the quantitative relationship between water reflectance and water quality parameters.Recently,machine learning methods have been increasingly used for analyzing relationships between complex factors and making full use of known information.In this study,based on the spectral bands of Ziyuan-1 02C imagery,K Near Neighborhood,Support Vector Machine,Random Forest,Gradient Boosted Decision Trees,Artificial Neural Network were used to develop the models for estimating the water quality parameters including Chlorophyll-a,Secchi Disk,total phosphorus,total nitrogen,COD Mn.The best models were then successfully applied to map the spatial pattern of these five parameters.The eutrophication status of East Lake was evaluated by Comprehensive nutritional State method.Compared with the data of 22 sampling sites in November 2014 and 23 in January 2015,the accuracy rates of eutrophication classification of the water quality parameters based on machine learning method were 95.5%and 82.6%,respectively.Compared with the data of Wuhan Environmental Protection Bureau,the correct rates of eutrophication of sub-lakes in East Lake were both 71.4%.