Classifying Pinus roxburghii Using an Innovative Training Approach of Fuzzy Models While Handling Heterogeneity Within Class in Western Himalayan Forests

Remote sensing can be used for effectively mapping plant species, thereby aiding in their sustainable management. Pinus roxburghii (PR), also known as Chir Pine, is often found alongside Quercus leucotricophora and Rhododendron arboreum around Dudhatoli range, Uttarakhand. It holds immense ecological and economic importance in the Himalayan region. It is observed that PR exhibits heterogeneity within an image likely due to varying aspect, shadows, and canopy coverage. This study focuses on mapping PR using an innovative individual sample as mean (ISM) approach embedded in the framework of the Possibilistic c-means (PCM) and noise clustering (NC) fuzzy classifier, specifically addressing heterogeneity within the class while comparing it with the conventional mean training parameter approach. The research utilises the Modified Soil Vegetation Index 2 (MSAVI2) from a semi-hypertemporal (SH) dataset consisting of 17 images acquired by the 8-band PlanetScope data. This study also experiments with different numbers of training samples to understand their impact on the output. Results of PCM with an m value of 2.1 and NC with δ value of 50,000 show good classified outputs. It was also found that a training sample size of 11 showed the best result. This study showcases progress in using the SH dataset, ISM-based PCM, and NC models with a limited number of training samples to overcome challenges posed by class heterogeneity.