Evaluating the potential of sentinel-2, landsat-8, and irs satellite images in tree species classification of hyrcanian forest of iran using random forest

Spatially explicit information on tree species composition of any forest provides valuable information to forest managers as well as to nature conservationists. In this study, the potential of three spaceborne sensors: (1) Landsat-8, (2) Sentinel-2, and (3) IRS-Pansharpened were compared by applying Random Forest (RF) classification algorithms to classify the three most common tree species: Pinus taeda, Alnus spp., and Populus spp., in Hyrcanian forest of Iran. Three RF models with optimized parameters of mtry and ntree were used for the classification of trees species. Based on our Overall Accuracy (OA) and Kappa Coefficient (KC) analysis, IRS-Pansharpened data showed the highest accuracy (OA = 84.9% and KC = 79.7%), followed by Landsat-8 (OA = 78.2% and KC = 70.6%), and Sentinel-2 (OA = 77% and KC = 70%). According to the Mean Decrease in Accuracy (MDA) criterion delivered as an output of RF, the near-IR spectral band was found on the top rank (high variable importance) as compared to all other spectral bands for tree species classification. The findings of this study can be used by the researcher, forest managers, economists and policy and decision makers in the context of sustainable forest management of Hyrcanian forest resources.