Logo image
Back
A Sentinel-2 Benchmark and Deep-Learning Study for Wildfire Damage Mapping
Conference proceeding

A Sentinel-2 Benchmark and Deep-Learning Study for Wildfire Damage Mapping

Valeria Martin, K. Brent Venable and Derek Morgan
Proceedings of the 8th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery, pp.135-145
ACM Conferences
GeoAI '25: 8th ACM SIGSPATIAL International Workshop on AI for Geographic Knowledge Discovery ( Minneapolis, Minnesota, USA, 11/03/2025–11/06/2025)
11/03/2025
DOI: https://doi.org/10.1145/3764912.3770829
Web of Science ID: WOS:001665788100020

Metrics

3 Record Views

Abstract

Wildfires are increasing in frequency and severity due to climate change, creating a growing need for accurate, large-scale damage assessment tools. In this work, we describe a new multi-temporal benchmark dataset for wildfire segmentation using Sentinel-2 imagery and wildfire polygon data, which we call CalFireSeg-50. For 50 of the largest recent wildfires in California (2019 to 2023), we collect RGB, NBR, and NDVI images. Ground truth masks are, then, generated using a conditional approach that combines spatial fire boundaries with burn-sensitive vegetation indices. Using images from this benchmark, we conduct an evaluation of four deep learning (DL) segmentation models for detecting wildfire damage: UNet, UNet++, TransUNet, and SegFormer. Our analysis compares model performance across multiple established quality metrics (e.g., IoU and Dice), training time, and data efficiency using a consistent GPU setup. TransUNet achieves the best performance across most quality metrics, including the highest Dice score (88%) and accuracy (94%), while UNet++ marginally obtains the top IoU score (82%), indicating stronger contour-level precision. However, TransUNet requires significantly more time to train than the CNN-based approaches. SegFormer shows promise in few-shot settings, but is outperformed by the other models when training involves larger datasets. These findings provide novel insights into the strengths and limitations of CNN, hybrid, and transformer-based architectures for wildfire segmentation, significantly advancing DL approaches for wildfire damage mapping.

Details

Logo image