基于哨兵 3A-OLCI 影像的内陆湖泊藻蓝蛋白浓度反演算法研究
投稿时间:2017-08-04  修订日期:2017-11-30  点此下载全文
引用本文:
摘要点击次数: 13
全文下载次数: 
作者单位E-mail
苗松 南京师范大学 njnums1214@163.com 
吕恒 南京师范大学 Heng.Lyu@njnu.edu.cn 
中文摘要:蓝藻是内陆富营养水体水华发生的主要优势藻种,而藻蓝蛋白(Phycocyanin,PC)是蓝藻的标志性色素,因此利用遥感估算水体中藻蓝蛋白浓度从而对蓝藻水华预警具有重要意义。本文利用太湖、滇池、洪泽湖的实测数据,构建藻蓝蛋白随机森林遥感估算模型,并将模型应用到哨兵3A-OLCI影像。通过对随机森林的输入自变量进行重要性分析,发现第7波段(620nm)、第8波段(665nm)和第9波段(675nm)三个波段对藻蓝蛋白反演的影响程度最大。同时,反演结果表明,随机森林反演的藻蓝蛋白浓度平均绝对百分比误差(MAPE)为34.86%,均方根误差(RMSE)为38.67ug/L,与simis等半分析算法和齐琳的PCI(Phycocyanin Index)指数模型相比,平均绝对百分比误差(MAPE)分别提高了85.06%和15.65%,均方根误差分别提高了26.08ug/L和19.86ug/L。利用地面实测数据对同步卫星影像大气校正进行精度评价,发现MUMM(The Management Uint Mathematical Model)算法可以用于OLCI影像的大气校正,尤其在560nm-779nm处共8个波段的MAPE低于30%,光谱曲线与实测光谱曲线形状保持一致。结果表明本文所构建的基于哨兵3A-OLCI影像的藻蓝蛋白随机森林反演模型,可以成功的应用于我国的内陆富营养化湖泊,为我国内陆湖泊藻蓝蛋白浓度的遥感反演提供一个新的算法。
中文关键词:藻蓝蛋白  OLCI  随机森林  遥感  反演
 
Retrieval algorithm of phycocyanin concentration in inland lakes from Sentinel 3A-OLCI images
Abstract:Abstract:Cyanobacteria is the dominant algae species in inland eutrophic water bodies, and the phycocyanin (PC) is its unique pigment which can be used as an indicator of its presence. Therefore, the retrieval of PC concentration by remote sensing is of great significance to early warning of cyanobacteria bloom. In this paper, the Random Forest retrieval Model for estimating PC concentration based on the sentinel 3A-OLCI bands was developed using in situ data collected from Taihu Lake, Dianchi Lake and Hongzehu Lake. The results of the importance analysis of input variables in random forest demonstrated that the seventh band(674nm), the eighth band(665nm) and the ninth band (620nm) have significant impact on the PC estimation. The accuracy assessment showed that the Mean Absolute Percentage Error(MAPE) of this PC retrieval model is only 34.86% with the Root Mean Square Error(RMSE) of 38.67ug/L. The comparison between the mode developed by this paper and other models, i.e., Simis semi-analytic algorithm and PCI exponential model was extensively conducted, and it was found that compared with other two models, the MAPE was improved by 85.06% and 15.65% respectively, and the RMSE was improved by 26.08ug/L and 19.86ug/L respectively. The atmospheric correction accuracy was further analyzed using the in situ samples and synchronous satellite image, and the result showed that the Management Uint Mathematical Model (MUMM) method can be successfully used for the OLCI image. The atmospheric corrected spectral curves are consistent with the measured spectral curves, and the MAPEs of 8 bands are all less than 30% at the wavelength range between 560 and 779 nm. The random forest model developed for estimating PC concentration in this paper can be successfully applied to Sentinel 3A-OLCI images, which provides a new algorithm for remote estimation of phycocyanin concentration in inland lake.
keywords:phycocyanin(PC)  OLCI  Random Forest(RF)  remote sensing  inversion
  查看/发表评论  下载PDF阅读器

版权所有:《红外与毫米波学报》编辑部