Using Fuzzy Logic and Landsat TM-derived Vegetation Indices to Detect Forest Disturbances with Uncertainty Measures

摘要

Detecting forest cutting and other disturbances from multi-year satellite images over large areas has been challenging. This paper motivates to present new techniques for detecting forest changes from time series images and compare the efficiency of three widely used indicators: Normalized Difference Vegetation Index (NDVI), Normalized Difference Water Index (NDWI), and Vegetation Condition Index (VCI). A fuzzy logic method for change detection is presented to generate spatio-temporal fuzzy change and uncertainty maps of forest disturbances. This approach overcomes the difficulty of selecting proper thresholds for traditional image differencing. The proposed method was applied to detect the forest disturbances in the north of South Carolina and the south of North Carolina, U.S.A. using the Landsat Thematic Mapper (TM)images from 1984 to 2005. The produced disturbance maps showed not only forest disturbances but also the uncertainty of the detected changes. Furthermore, the results for forest disturbances using NDVI, NDWI, and VCI were compared and analyzed. The results showed that both NDWI and VCI were more effective in detecting forest disturbances than NDVI was. However, the fuzzy detection maps of forest disturbances calculated from NDWI and VCI contained more noises than the maps generated from NDVI.