Application of spectral decomposition algorithm for mapping water quality in a turbid lake (Lake Kasumigaura, Japan) from Landsat TM data

详细信息    查看全文

• 作者：Youichi Oyama ; Bunkei Matsushita ; Takehiko Fukushima ; Kazuo Matsushige ; Akio Imai
• 关键词：Spectral decomposition algorithm ; Remote sensing ; Turbid lakes ; Bio-optical model
• 刊名：ISPRS Journal of Photogrammetry and Remote Sensing
• 出版年：2009
• 出版时间：January 2009
• 年：2009
• 卷：64
• 期：1
• 页码：73-85
• 全文大小：3967 K

文摘

The remote sensing of Case 2 water has been far less successful than that of Case 1 water, due mainly to the complex interactions among optically active substances (e.g., phytoplankton, suspended sediments, colored dissolved organic matter, and water) in the former. To address this problem, we developed a spectral decomposition algorithm (SDA), based on a spectral linear mixture modeling approach. Through a tank experiment, we found that the SDA-based models were superior to conventional empirical models (e.g. using single band, band ratio, or arithmetic calculation of band) for accurate estimates of water quality parameters. In this paper, we develop a method for applying the SDA to Landsat-5 TM data on Lake Kasumigaura, a eutrophic lake in Japan characterized by high concentrations of suspended sediment, for mapping chlorophyll-a (Chl-a) and non-phytoplankton suspended sediment (NPSS) distributions. The results show that the SDA-based estimation model can be obtained by a tank experiment. Moreover, by combining this estimation model with satellite-SRSs (standard reflectance spectra: i.e., spectral end-members) derived from bio-optical modeling, we can directly apply the model to a satellite image. The same SDA-based estimation model for Chl-a concentration was applied to two Landsat-5 TM images, one acquired in April 1994 and the other in February 2006. The average Chl-a estimation error between the two was 9.9 % , a result that indicates the potential robustness of the SDA-based estimation model. The average estimation error of NPSS concentration from the 2006 Landsat-5 TM image was 15.9 % . The key point for successfully applying the SDA-based estimation model to satellite data is the method used to obtain a suitable satellite-SRS for each end-member.