Application Of A Classification And Matching Algorithm Based On Normalized Mutual Information For Estimating The Chlorophyll-a Concentration In Taihu Lake,China

Inland lake eutrophication has become one of the most serious environmental issues around the world.Chlorophyll-a(Chl-a)is an important indicator of the degree of pollution due to eutrophication.Chl-a concentration can be estimated in real-time and in a dynamic manner using remote sensing technology,which provides important information and technical support for the monitoring and early warning of water eutrophication.However,due to the complex optical properties of inland lakes,the accuracy of a single model is affected both spatially and temporally.Additionally,due to the limitation of the spatial and temporal resolution of existing sensors,the application of Chl-a inversion algorithms using various types of remote sensing data does not meet the demands of spatiotemporal variation.Taihu Lake,which is one of the most eutrophic lakes in China,was chosen as the study area.To improve the accuracy and universality of the single model,classification and matching algorithms were proposed to estimate the Chl-a concentration.Moreover,to satisfy the spatiotemporal demand of the inland Chl-a concentration,the advantages of various types of remote sensing data were combined by applying fusion models.Chl-a concentrations were then estimated from NMI classification and matching algorithm using fusion images.The studies above provided not only a theoretical principle for operational Chl-a production but also the theory and technology for real-time and dynamic monitoring of inland water eutrophication.Based on the objectives above,the major research results were as follows:1.A classification and matching algorithm based on normalized mutual information(NMI)was proposed to estimate Chl-a concentration.Based on the typical Chl-a inversion algorithms and the theory of NMI,classification and matching algorithms,which included an NMI direct matching algorithm and an NMI weighted matching algorithm,were proposed to estimate Chl-a concentration.Additional critical steps including sub-spectra extraction,distance function determination and weighted factor determination were completed in the process of model building.The NMI classification and matching algorithm provided the theory and method for Chl-a estimation using different types of remote sensing data.2.The NMI classification and matching algorithms were established separately using in situ data,GOCI data and HJ-1 CCD data.According to the NMI classification and matching algorithms,in situ spectral reflectance data were first classified using the NMI classification method.Then,the NMI direct matching and the NMI weighted matching algorithms were applied separately using in situ data,GOCI data and HJ-1 CCD data.Finally,the potentials of the NMI classification and matching algorithms were evaluated and demonstrated using these different types of remote sensing data.The results indicated that in situ normalized spectral reflectance data could be classified into three types,and the typical features of the three types were retained in the spectral reflectance data and were resampled in the GOCI and HJ-1 CCD bands.An accuracy evaluation of the non-classification method,the NMI direct matching algorithm and the NMI weighted matching algorithm indicated that the proposed methods could be applied to estimate the Chl-a concentration.The accuracies of the NMI direct matching algorithm and the NMI weighted matching algorithm were higher than those of the non-classification method.Moreover,most data were dominated by two or three Chl-a inversion algorithms.The performance of the NMI weighted matching algorithm was better than that of the NMI direct matching algorithm.The NMI weighted matching algorithm also reduced the splicing and discontinuous effects in the remote sensing images.3.Based on the NMI weighted matching algorithm,the Chl-a concentrations were estimated using multi-source remote sensing data.The Chl-a concentrations were estimated separately using the NMI weighted matching algorithm for GOCI and HJ-1 CCD images acquired at different times.According to the above results,the advantages and disadvantages were analyzed separately for the GOCI and HJ-1 CCD images.On this basis,a Spatial and Temporal Adaptive Reflectance Fusion Model and an Improved Unmixing Model(STARFM-IUM)were proposed to obtain fusion images that augmented the advantages of GOCI and HJ-1 CCD.Then,the NMI weighted matching algorithm was applied to the fusion data.The results indicated that this application was satisfactory for the different types of remote sensing images.Additionally,the application of the NMI weighted matching algorithm to the fusion images better reflected the spatial-diurnal distribution of Chl-a in Taihu Lake.