Improvement of remotely sensed vegetation coverage in heterogeneous environments with an optimal zoning approach

The high spatial heterogeneity forms a major uncertainty in accurately monitoring of vegetation coverage. In this study, an optimal zoning approach with dividing the whole heterogeneous image into relatively homogeneously segments was proposed to reduce the effects of high heterogeneity on vegetation coverage estimation. With the combination of the spectral similarity of the adjacent pixels and spatial autocorrelation of the segments, the optimal zoning approach accounted for the intrasegment uniformity and intersegment disparity of improved image segmentation. In comparison, vegetation coverage in the highly heterogeneous karst environments tended to be underestimated by the normalized difference vegetation index (NDVI) and overestimated by the normalized difference vegetation index-spectral mixture analysis (NDVI-SMA) model. Hence, when applying remote sensing for highly heterogeneous environments, the influence of high heterogeneity should not be ignored. Our study indicates that the proposed model, using NDVI-SMA model with improved segmentation, is found to ameliorate the effects of the highly heterogeneous environments on the extraction of vegetation coverage from hyperspectral imagery. The proposed approach is useful for obtaining accurate estimations of vegetation coverage in not only karst environments but also other environments with high heterogeneity.