Soft Classification Based Remote Sensing Estimation Of Chlorophyll-a Concentration And Long Time Series Analysis In Taihu Lake, China

Water pollution, especially eutrophication of inland waters, has become more and more serious in China in recent years. Chlorophyll-a (Chla) is the most important pigment in phytoplankton for photosynthesis, and its concentration is an important index of eutrophication. Satellite remote sensing plays a more and more important role in water environmental monitoring with the advantages of low cost, wide range, fast speed, good continuity, and so on. Chlorophyll-a concentration (Cchla) estimation is an important content of water color remote sensing, while it is always difficult in eutrophic turbid inland water. The main content of this study was to estimate Chla concentration of eutrophic turbid inland water, such as Taihu lake in eastern China, by using the technology of remote sensing. And the spatial distribution rules and long time trend of Cchla of Taihu Lake were further analyzed.In this study, we presented a new Chla estimation strategy, which was called soft classification (fuzzy classification) based estimation strategy. In comparision, we renamed the existing algorithms as tradition estimation strategy and hard classification based estimation strategy. We selected the eutrophic turbid inland water Taihu lake as the research area, and obtained a large number of data in Taihu lake, including water surface spectra and MERIS satellite images. We actualized the estimation strategies used these data, and evaluated the soft classification based estimation strategy by comparing with tradition estimation strategy and hard classification based estimation strategy. Then, we developed a Chla estimation technological process for massive and long time MERIS data by soft classification based estimation strategy, and solved some key technical problems in the process. Finally, we estimated Cchla in Taihu lake from2002to2012by MERIS data, and analyzed the characteristics of space distribution and the annual, seasonal and monthly variation and trend of of Cchlas in Taihu lake.We found the following phenomenon from the results of estimation strategies accuracy evaluation by water surface spectra and MERIS satellite images. The accuracy of hard classification based estimation strategy was better than tradition estimation strategy; the accuracy of soft classification based estimation strategy was the best, and its universality was also the best. Over the last decade, the spatial distribution of Chla decreased progressively from north to south in Taihu lake; the annual variation obeyed the shape of "W"; the seasonal variation was remarkable: lowest in winter, rising in spring, highest in summer, and reducing in autumn; the monthly variation obeyed the shape of "A" which showed low in the months of winter and high in the months of summer, and the fluctuant variation on one year cycle.The mainly contributions of this study are as follows:Firstly, we summarized and presented four Chla estimation strategy. Then, we comprehensively evaluated the accuracies of the four estimation strategy with five Chla estimation algorisms based on water surface spectra, and with twelve algorisms based on MERIS images. We found optimal algorisms and models of different type of waters by comparasion.Secondly, We demonstrated the suitability of MERIS2P remote sensing reflectance product in Taihu lake by comparing the MERIS2P data and the field measured reflectance.Thirdly, We estimated Cchla to1932scenes of MERIS images from2002to2012in Taihu lake, and then analyzed the spatial distribution, annual change, seasonal change and lunar change of Cchla in Taihu lake.The main innovation of this study are as follows:Firstly, We developed a soft classification based estimation strategy for Cchla estimation in Taihu lake. The strategy can improve the stability, reliability and smoothness of Cchla estimation results. This estimation strategy had solved the regional and seasonal limitations of the traditional Cchla estimation methods. Secondly, We developed an automatic water extraction methods assisted by water vector boundary data. This method greatly improved the water extraction accuracy and the ability to process mass data.