Burned Areas Mapping Using Sentinel-2 Data and a Rao’s Q Index-Based Change Detection Approach: A Case Study in Three Mediterranean Islands’ Wildfires (2019–2022)

Abstract

This study explores the application of remote sensing-based land cover change detection techniques to identify and map areas affected by three distinct wildfire events that occurred in Mediterranean islands between 2019 and 2022, namely Sardinia (2019, Italy), Thassos (2022, Greece), and Pantelleria (2022, Italy). Applying Rao’s Q Index-based change detection approach to Sentinel-2 spectral data and derived indices, we evaluate their effectiveness and accuracy in identifying and mapping burned areas affected by wildfires. Our methodological approach implies the processing and analysis of pre- and post-fire Sentinel-2 imagery to extract relevant indices such as the Normalized Burn Ratio (NBR), Mid-infrared Burn Index (MIRBI), Normalized Difference Vegetation Index (NDVI), and Burned area Index for Sentinel-2 (BAIS2) and then use (the classic approach) or combine them (multidimensional approach) to detect and map burned areas by using a Rao’s Q Index-based change detection technique. The Copernicus Emergency Management System (CEMS) data were used to assess and validate all the results. The lowest overall accuracy (OA) in the classical mode was 52%, using the BAIS2 index, while in the multidimensional mode, it was 73%, combining NBR and NDVI. The highest result in the classical mode reached 72% with the MIRBI index, and in the multidimensional mode, 96%, combining MIRBI and NBR. The MIRBI and NBR combination consistently achieved the highest accuracy across all study areas, demonstrating its effectiveness in improving classification accuracy regardless of area characteristics.