Post-fire burned forest area mapping with Sentinel-2 imagery: evaluating object-based machine learning models and transferability

Abstract

Forest fires pose significant threats to ecosystems and cause irreversible damage, contributing to environmental challenges beyond the burned areas. Timely detection is critical for assessing damage and guiding effective post-fire response. This study aims to identify, and map burned forest areas using uni-temporal Sentinel-2 imagery and to evaluate the spatial and temporal transferability of object-based machine learning models. An object-based classification approach was applied by integrating Object-Based Image Analysis (OBIA) with Random Forest (RF) and Gradient Tree Boosting (GTB) algorithms on the Google Earth Engine (GEE) platform. Additional indices—Mid-Infrared Burn Index (MIRBI), Normalized Burn Ratio 2 (NBR2), Burn Area Index (BAI), and Normalized Difference Vegetation Index (NDVI)—were calculated and included as input bands. Image segmentation was performed using the Simple Non-Iterative Clustering (SNIC) algorithm. The analysis focused on four Mediterranean forest areas in Türkiye, with transferability tested in selected sites in Greece and Spain. RF and GTB models achieved overall accuracies between 93.5 and 97.7% and 92.9–96.8%, respectively. Both models maintained overall accuracies above 83% when transferred to ecologically similar Mediterranean study areas in Greece and Spain. The findings demonstrate the potential of object-based RF and GTB models as transferable tools for burned-forest mapping within comparable Mediterranean environmental conditions.