Optimization of sampling schemes for vegetation mapping using fuzzy classification

This paper considers the design of an optimal sampling scheme for a multivariate fuzzy-k-means classifier. Fuzzy classification is applied to delineate vegetation patterns from remote sensing data. The confusion index distinguishes subareas with high uncertainty due to class overlapping from those with low uncertainty. These subareas govern allocation of sample points. A simulated annealing approach minimizes the mean of shortest distances between samples. Optimization was done by prioritizing the survey to areas with high uncertainty. The methodology is tested on a site located in the Amazonian region of Peru. It resulted into an almost equilateral triangular scheme at those parts of the area where uncertainty was highest. The study shows that optimal sampling can be successfully combined with fuzzy classification, using an appropriate weight function.