白洋淀水体总磷总氮高光谱估算模型比较  

Comparative study on the hyperspectral estimation models of TP and TN in Baiyangdian water body

作　　者：陈洁 张立福 张红明 张琳珊 岑奕 童庆禧 CHEN Jie;ZHANG Lifu;ZHANG Hongming;ZHANG Linshan;CEN Yi;TONG Qingxi(State Key Laboratory of Remote Sensing Science,Aero Information Research Institute,Chinese Academy of Sciences,Beijing 100101,China;University of Chinese Academy of Sciences,Beijing 100049,China;China Aero Geophysical Survey and Remote Sensing Center for Natural Resources,Beijing 100083,China)

机构地区：[1]中国科学院空天信息创新研究院遥感科学国家重点实验室,北京100101 [2]中国科学院大学,北京100049 [3]中国自然资源航空物探遥感中心,北京100083

出　　处：《遥感学报》2023年第7期1642-1652,共11页NATIONAL REMOTE SENSING BULLETIN

基　　金：国家自然科学基金(编号:41830108,41977154);兵团重大科技项目(编号:2018AA004);兵团重点领域创新团队(编号:2018CB004);国家重点研发计划(编号:2017YFC1500901)。

摘　　要：总磷(TP)、总氮(TN)是水质富营养化的重要指标,亦是水质监测的主要参数。具有快速高效、无二次污染等特点光谱法水质监测是当今水环境遥感分析研究的热点。针对水体TP、TN反演模型采用实验室标准液或野外全样本进行建模时,各种水质参数的相互影响及预测值超出建模样本值域的可能,使得实际的预测效果并不理想的情况。本文以白洋淀实验区的实际水体样本为反演模型的输入值,在确定最优相关波段和最佳反演模型的基础上,讨论了5种不同浓度范围场景下的样本建模对反演模型的影响,同时剖析了模型对超出建模浓度值样本的预测能力。结果表明:建模样本浓度覆盖预测样本时,反演模型决定系数R^(2)>0.6,TP、TN浓度预测值的平均偏离度ARE<20%;建模浓度高于预测样本时,R^(2)在0.6左右,对超过建模浓度范围12%以内的预测值,其ARE<25%;建模浓度低于预测值时,R^(2)介于0.4—0.5,预测值超过建模样本浓度一倍时,ARE≤30%;建模样本浓度位于预测值两侧时,R^(2)可达0.8,ARE<25%;建模样本浓度值介于预测值之间时,0.45<R^(2)<0.55,ARE>35%。通过本文的研究与讨论,可为水质参数监测的实际工程应用提供科学依据和参考。Total Phosphorus(TP)and Total Nitrogen(TN)are important indicators of water quality eutrophication and the main parameters for water quality monitoring.Water quality monitoring by spectroscopy has become a hot spot in the current remote sensing water environment research because it is rapid and efficient and has no secondary pollution.The usual TP and TN inversion models are established based on the laboratory configuration standard solution for spectral measurement or the modeling based on the full samples.The model constructed in this way has a good regression effect.However,the actual water body causes the mutual influence of various water quality parameters,the concentration distributions of TP and TN are not uniform,and the predicted value may exceed the training sample range,making the actual prediction effect always unsatisfactory.In this study,the actual water samples in the Baiyangdian area are used as the input values of the inversion model.First,the measured spectral data and the chemical analysis values of TP and TN are used to compare the relationship between the correlation values of different reflectances and water quality parameters.Various inversion models have been constructed for the best relevant bands.The most stable and accurate modeling method has been determined through comparison.Therefore,the modeling samples are divided into uniform,high-value,low-value,median-value,and max-min-value samples according to concentration.Then,the influence of the sample modeling with different concentration ranges on the inversion model is discussed.The model’s predictive ability for samples with concentration values beyond modeling is determined.The extraction results of the characteristic wavebands in the range of 400—100 nm indicate that the reflectance correlation coefficient of TP and TN corresponding to a single wavelength is less than 0.3,which is not high;the maximum correlation coefficient with the first-order value of reflectance is 0.76,which is a moderate correlation;the correlation coeffic

关 键 词：水质监测 总磷 总氮 浓度反演 偏最小二乘法 

分 类 号：P2[天文地球—测绘科学与技术]