Problem
Research questions and friction points this paper is trying to address.
Burn scar delineation
Sentinel-3 OLCI
Temporal modelling
Post-fire damage assessment
Image segmentation
Temporal Modelling for Burn Scars on Sentinel-3
🤖 AI Summary
This study addresses the loss of pre- and post-fire change signals caused by processing satellite imagery independently in burned area segmentation. We construct a temporally paired dataset comprising 246 fire events and systematically evaluate deep learning architectures, including U-Net, SegFormer, and ConvLSTM-enhanced models, using Sentinel-3 OLCI imagery. Our findings demonstrate for the first time that temporal modeling improves segmentation accuracy only when pre-fire frames are incorporated. Furthermore, we reveal that a five-band subset achieves performance comparable to the full 21-band configuration. By maintaining high accuracy while substantially reducing computational costs, this work establishes a new paradigm for efficient burned area monitoring.
📝 Abstract
Rapid and accurate burn scar delineation from satellite imagery is essential for post-fire damage assessment. Sentinel-3 OLCI, with daily revisit and 21 spectral bands, suits rapid mapping, yet most pipelines treat acquisitions independently, leaving the pre/post-fire change signal unexploited. We present a dataset of 246 wildfire activations (2016-2025) from the Copernicus Emergency Management Service, with Sentinel-3 OLCI temporally paired acquisitions. We benchmark spatial and temporal (ConvLSTM-augmented) variants of three backbones (U-Net, SegFormer, ConvNeXt-UPerNet) under two input modes and spectral configurations. Temporal modeling improves segmentation only when pre-fire frames are included, and a 5-band subset matches the full 21-band OLCI configuration.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Temporal Modelling
Burn Scar Segmentation
ConvLSTM
Sentinel-3 OLCI
Spectral Band Selection
🔎 Similar Papers
No similar papers found.
