Sensing Pixel Decomposition Methods And Applied Research In The Dianchi Lake Water Quality Monitoring

The project Water Pollution Monitoring and Water Ecology Environment Management System for Dianchi Lake in Yunnan Province was a sub-project whose main objective was to develop a Spatial Decision Support System (SDSS) and apply to water pollution of Dianchi Lake analysis and decision with remote sensing and GIS techniques.It is a well-known fact that pixels in remotely sensed imagery are typically mixed pixels due to both the limited spatial resolution of sensors and the heterogeneous surfaces of ground covers. Mixed pixels are a major source of inconvenience in conventional classification. Their presence severely degrades the performance of classification (or target detection) systems. Thus, pixel unmixing becomes a major concern in many remote sensing classification applications. In this research, we explore pixel unmixing to analyze the water pollution of Dianchi lake.Over the last couple of decades, scientists have been researching on ways of unmixing pixels to determine the proportions of their component endmembers. The usual approach employed to achieve this is through modeling of spectral mixtures. Several types of models have been proposed. The linear mixture model (LMM) has been a dominant mathematical model in pixel unmixing analysis because of its simplicity and effectiveness. In the other way, once we have selected one of the spectral mixture models, the determination of component endmembers plays a key role in the pixel unmixing.We propose an improved linear mixture model and apply it to our research. At the same time, we explore two different methods to determine the component endmembers and discuss the results. One advantage of our model is that it doesn't require the correction for atmospheric and topographic effects, which is usually complex and often gives approximate results.In this paper, we respectively applied ISODATA, maximum likelihood technique and the improved linear mixture model to classify the pixels and analyze the pollution of Dianchi lake. The results prove that the improved linear mixture model can give a better understanding of the pollution of Dianchi lake. It can not only give the abundance images of component endmembers constituting the area of a pixel, but also get the classification image.Our research also show that, compared to the regular regression analysis of water quality parameters, the improved linear mixture model can better simulate the distribution of the water pollution of Dianchi lake when the concurrent in situ measurements are absent.In the end, we use the results to analyze the temporal and spatial change of the water pollution of Dianchi lake...