Abstract:

In this study, the false color images of visible and near-infrared spectra of FY3/MERSI data were used, combined with other digital maps of regional land use, vegetation type and distribution of forests. The multi-channel data were classified into 4 categories: forest, winter wheat, water and bare land, by visual interpretation and unsupervised classification. These features were then analyzed their spectral statistical characteristics of visible and near-infrared spectra. Based on all of these analyses, the multi-channel data were classified by using four supervised classification, namely parallelpiped, minimum distance, mahalanobis distance, Maximum Likelihood. After the feasibility of the forest classification was analyzed, the classification accuracy was assessed with digital forest map, and then the distribution of forests in the study area was determined. The results showed that the classification obtained by the Maximum Likelihood method had the highest accuracy, which was 85%, followed by the Minimum Distance and the Mahalanobis Distance method, and the results obtained by the Parallelepiped method were the worst. On the whole, with combining supervised classification and unsupervised classification method. It was found that extracting the macro distribution of forest resources with MERSI data was feasible, and the extracted distribution of forest resources in Shanxi Province of this study was accurate, therefore it could be a effective means for macro monitoring of regional forest resources.

Key words: MERSI data, supervised classification, forest