内陆水体水质参数遥感反演集合建模方法

Ensemble modeling methods for remote sensing retrieval of water quality parameters in inland water

以微山湖为研究对象,利用2015年6月11~13日获取的实测高光谱和水体叶绿素a浓度、总悬浮物浓度和浊度数据,构建3种水质参数遥感反演常用的经验模型和PSO-SVM模型并进行精度评价,确定参与3种水质参数集合建模的反演模型,分别利用以熵权法(EW-CM)、集对分析法(SPA-CM)为代表的确定性集合建模方法和以贝叶斯模型平均(BMA)为代表的概率性集合方法构建反演3种水质参数的EW-CM、SPA-CM和BMA集合模型.通过贝叶斯平均方法获取各模型和BMA集合模型反演3种水质参数的不确定性区间,对比3种水质参数各模型和集合模型反演结果.结果表明:(1)确定性集合模型中SPA-CM模型精度整体高于EW-CM模型;(2)BMA概率性集合模型建模精度整体上要优于SPA-CM和EW-CM集合模型,验证精度稍低于SPA-CM模型,和EW-CM模型相当;(3)概率性集合建模可以给出集合模型和各模型反演水质参数的不确定性区间;(4)确定性和概率性集合模型可以综合各模型信息,使得集合模型同时具有较高的建模和验证精度,降低单一模型反演水质参数的不确定性,并在一定程度上提高水质参数反演精度.

Based on the measured hyperspectral data and concentration of chlorophyll a, total suspended matter（TSM） and turbidity obtained during June 11 to 13, 2015 in Weishan Lake, empirical models and PSO-SVM model were established to retrieve the three water quality parameters. Meanwhile, the performance of those models was evaluated to determine the models applied to ensemble modeling. The ensemble models containing EW-CM, SPA-CM and BMA were established to retrieve the three water quality parameters by using deterministic ensemble method and probabilistic ensemble method. The deterministic and probabilistic ensemble method was based on the entropy weight method along with pair analysis method and Bayesian averaging method, respectively. Bayesian averaging method was employed to obtain the retrieval uncertainty range of the three water quality parameters by using the single model and the BMA ensemble model, and the retrieval uncertainty range of these models was compared. These results demonstrated that（1） the accuracy of SPA-CM model was better than that of EW-CM model in deterministic ensemble models;（2） the modeling accuracy of BMA probabilistic ensemble model was better than that of SPA-CM and EW-CM model; the verification accuracy of BMA probabilistic ensemble model was similar with that of EW-CM model but slightly lower than that of the SPA-CM model;（3） Probabilistic ensemble modeling could obtain the retrieval uncertainty range of water quality parameters by using the ensemble model and the single model;（4） The deterministic and probabilistic ensemble model associated with the single model information showed a higher modeling and verification accuracy, which could be used to reduce the uncertainty of water quality parameters retrieval compared with single model and promote the retrieval accuracy of water quality parameters in a manner.