使用光谱平滑提高浑浊水体叶绿素a浓度估算模型的应用精度  

Using Spectral Smoothing Method to Improve the Validation Precision of the Chlorophyll-a Estimation Model in Turbidity Water

作　　者：程春梅[1] 韦玉春[1] 王国祥[1] 张静[1] 夏晓瑞[1] 

机构地区：[1]南京师范大学虚拟地理环境教育部重点实验室,江苏南京210023

出　　处：《遥感技术与应用》2013年第6期941-948,共8页Remote Sensing Technology and Application

基　　金：江苏省普通高校自然科学研究计划项目(09KJA420001;07KJB420062)资助;国家自然科学基金资助项目"湖泊藻类不同色素组分的高光谱定量反演研究"(40771152);江苏高校优势学科建设工程资助项目;江苏省普通高校研究生科研创新计划项目(CXLX12_0394)

摘　　要：水体叶绿素a浓度估算是水质参数遥感监测的重要内容,由于采样时间和地点的限制,传统估算模型的参数和形式具有较大的时间和空间依赖性。光谱平滑可以突出不同数据集的共同特征,从而增加模型的预测精度,因此考虑使用平滑方法来提高水体叶绿素a浓度估算模型的应用精度。利用太湖2004年夏季和2011年春季共4个月的数据,对比分析了移动平均、多项式平滑和核回归平滑处理前后浑浊水体实测反射光谱的变化,以及该变化对叶绿素a浓度三波段遥感估算模型和模型应用精度的影响。结果表明:核回归平滑处理后的光谱数据建立的三波段模型的残差正态分布更好,估算模型更为稳健。将2004年7月数据建立的模型用于8月数据,估算的叶绿素a浓度的RMSE从平滑前的33.56mg/m3降低到了平滑后的25.60mg/m3;将2011年3月建立的模型用于4月数据,估算的叶绿素a浓度的RMSE从平滑前的16.68mg/m3降低到了平滑后的10.57mg/m3。由此可以认为,实测光谱的核回归平滑处理有助于提高叶绿素a浓度三波段模型的应用精度,且对于叶绿素a浓度变化较大的夏季数据的改进效果更显著。Chlorophyll-a concentration estimation is an important part of the remote sensing monitoring of water quality parameters, traditional estimation model largely depends on place and time because of sam- piing limit. Spectral smoothing can improve the common features of different datasets, thereby increasing model prediction precision,smoothing method was used to improve the application accuracy of the chloro phyll estimation model. Based on the datasets of four months in Taihu lake-two months in summer of 2004 and two months in spring of 2011 ,this study compared the spectrum difference above the turbid water sur- face before and after Moving average smoothing,Savitzky-Golay smoothing and Kernel Regression smoot- hing, and discussed its influence on three-band estimation model and model application precision. The result shows that model residual of estimation model after Kernel Regression spectrum smoothing fits the normal distribution better,indicated that the estimation model is more stable. When model in July was used directly in August,2004,RMSE of Chla estimated decreased from 33.56 mg/m3 before smoothing to 25.60 mg/m3 after spectrum smoothing;when model in March was used directly in April,2011 ,RMSE of Chla estimated decreased from 16.68 mg/m3 before smoothing to 10.57 mg/m3 after spectrum smoothing. It can be conclu- ded that Kernel Regression smoothing of in-situ spectrum can increase the application precision of Chla three-band estimation model, and the improvement is more significant in summer data which has a large chlorophyll-a concentration range.

关 键 词：光谱平滑 核回归 叶绿素A 太湖 水色遥感 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]