A comparative analysis of classification methods for burn scar mapping in wetlands

This project explores the role and history of fire in the Macquarie Marshes and Gwydir Wetlands in New South Wales with two key objectives; to compare the spectral indices and classification methods that can be utilised for burn scar mapping, and to understand the behaviour of vegetation regeneration and regrowth in post-fire wetland ecosystems, namely the rate of regeneration and its relation to variables influencing burn severity, such as wetland moisture content. The methodology consists of multiple stages, the first of which includes performing spectral indices such as the normalised vegetation differential index (NDVI) and NDVI-Difference, burned area index (BAI), and the leaf area index (LAI) using satellite imagery. These processes measure the red and near-infrared light reflected from a sensed surface, which can be used to calculate green vegetation cover and density, which will allow for the detection of the large vegetation losses caused by bushfires. These outputs will then be classified by different classification methods such as neural networks (NNs), support vector machines (SVM) and random forest (RF) to determine the accuracy of the delineation between ‘burned’ and ‘unburned’ pixels, with a comparative statistical analysis performed to ensure accurate results. Finally, these outputs will be sequenced to form a time-series analysis, highlighting vegetation regeneration over time.