Description:
Mapping forest types in a natural heterogeneous forest environment using remote sensing data is a long-standing challenge due to similar spectral reflectance from different tree species and significant time and resources are required for acquiring and processing the remote sensing data. The purpose of this research was to determine the optimum number of remote sensing images and map the Sal forest through the analysis of Vegetation Index (VI) signatures. We analyzed the eight days’ composite moderate resolution imaging spectroradiometer (MODIS) time series normalized differential vegetation index (NDVI), and enhanced vegetation index (EVI) for the whole year of 2015. Jeffries-Matusita (J-M) distance was used for the separability index. Performance of EVI and NDVI was tested using random forest (RF) and support vector machine (SVM) classifiers. Boruta algorithm and statistical analysis were performed to identify the optimum set of imageries. We also performed data level five-fold cross validation of the model and field level accuracy assessment of the classification map. The finding confirmed that EVI with SVM (F-score of Sal 0.88) performed better than NDVI with either SVM or RF. The optimum 12 images during growing and post monsoon season significantly decreased processing time (to one-fourth) without much deteriorating accuracy. Accordingly, we were able to map the Sal forest whose area is accounted for about 36% of the 82% forest cover in the study area. The proposed methodology can be extended to produce a temporal forest type classification map in any other location.