Map regenerating forest stands based on DST and DSmT combination rules

Our results demonstrated the ability of the Free DezertSmarandache (DSm) model to improve thematic classification of forest regeneration over the use of Dempster-Shafer Theory (DST) and a classical Maximum Likelihood Algorithm (MLA). Overall, a classification accuracy of 82.75% was obtained with the reference method, MLA but it was improved by 7.4% by applying the fusion method DST (90.14%). Further improvement of 1% (to 91.13%), compared to those from the DST, was modest but noticeable when using the free DSm model. The study also showed the critical aspect of the design of the mass functions of each ancillary source and the difficulty to model the associated vagueness and uncertainty. Finally, the ability of the algorithms to take advantage of data fusion provided an excellent tool to test various combinations. After testing series of potential inputs, we found that drainage and surface deposit were the two best ancillary inputs in addition to spectral information to improve classification on the growth potential of regenerating forest stands in Southern Québec.