Robust Burned Area Delineation through Multitask Learning

In recent years, wildfires have posed a significant challenge due to theirincreasing frequency and severity. For this reason, accurate delineation ofburned areas is crucial for environmental monitoring and post-fire assessment.However, traditional approaches relying on binary segmentation models oftenstruggle to achieve robust and accurate results, especially when trained fromscratch, due to limited resources and the inherent imbalance of thissegmentation task. We propose to address these limitations in two ways: first,we construct an ad-hoc dataset to cope with the limited resources, combininginformation from Sentinel-2 feeds with Copernicus activations and other datasources. In this dataset, we provide annotations for multiple tasks, includingburned area delineation and land cover segmentation. Second, we propose amultitask learning framework that incorporates land cover classification as anauxiliary task to enhance the robustness and performance of the burned areasegmentation models. We compare the performance of different models, includingUPerNet and SegFormer, demonstrating the effectiveness of our approach incomparison to standard binary segmentation.