Research on the extraction of wetlands water based on unmixing mixed-pixel methods — A case study in Qinghai

Water was the crucial factor that kept wetlands stable and healthy. To fast and accurately extract wetlands water using remote sensing had become an important method on wetlands investigation, research and protection. Generally, the image which has lower spatial resolution, has higher temporal resolution, it is necessary for wetlands´ dynamic monitoring. However, the lower spatial resolution, the more mixed-pixel it has, which seriously affects classification accuracy. The mixed-pixel unmixing methods could get not only more accuracy water information but also capture dynamic changes of wetlands fast. A linear spectral unmixing model and maximum likelihood classification method are applied to extract wetlands water (including lakes, rivers, ponds et.) in Qinghai province using MODIS (Moderate Resolution Imaging Spectroradiometer) data which has high temporal resolution, hyperspectral, free download. The wetland map interpreted manually from CBERS-2B data are used to validate the classification results. The 2D scatter plots and Pixel Purity Index approach are compared in the selection of endmember. The result showed that the linear spectral mixing model could obtain wetlands water information more accuracy than maximum likelihood method; The results of 2D scatter plots has more accuracy than that of PPI approach.