Analysis of Various Spectral Indices and their Weighted Fusion Techniques for the Accurate Extraction of Forest Burn Scar in a Tropical Deciduous Forest

The present study was conducted in the tropical dry deciduous forests of Vidarbha region, Maharashtra, India. Burn scar discrimination capabilities of seven widely used spectral indices (SIs) such as burn area index (BAI), burned area index modified-LSWIR (BAIML), burned area index modified-sSWIR (BAIMs), normalized burn ratio (NBR), normalized difference vegetation index (NDVI), normalized difference moisture index (NDMI), modified soil-adjusted vegetation index (MSAVI) and their weighted fusion were examined in a multi-temporal domain. This study generated different images during fire season using individual SIs, change vector analysis and weighted fusion normalized difference image technique. Comparative analysis was performed between these approaches for burn scar discrimination with M statistics, burned and unburned class distribution and evaluation of confusion matrix. The study demonstrates that the weighted fusion of BAI, MSAVI and BAIMs can more accurately discriminate burn scars with good overall accuracy (86.61%).