Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

CaBuAr: California Burned Areas dataset for delineation

Daniele Rege Cambrin, Luca Colomba, Paolo Garza

TL;DR

The paper presents CaBuAr, an open dataset for burned-area delineation using Sentinel-2 pre- and post-fire imagery over California, paired with rasterized ground-truth masks from CALFIRE. It establishes three baselines—spectral-index analysis, SegFormer, and U-Net—to explore segmentation and change-detection tasks. Experimental results show deep learning models, particularly U-Net and SegFormer-B0, outperform index-based methods, while index-based approaches remain limited in reliability. The work provides a valuable resource for automated monitoring and recovery planning and plans for future expansion to other regions, modalities, and resolutions.

Abstract

Forest wildfires represent one of the catastrophic events that, over the last decades, caused huge environmental and humanitarian damages. In addition to a significant amount of carbon dioxide emission, they are a source of risk to society in both short-term (e.g., temporary city evacuation due to fire) and long-term (e.g., higher risks of landslides) cases. Consequently, the availability of tools to support local authorities in automatically identifying burned areas plays an important role in the continuous monitoring requirement to alleviate the aftereffects of such catastrophic events. The great availability of satellite acquisitions coupled with computer vision techniques represents an important step in developing such tools. This paper introduces a novel open dataset that tackles the burned area delineation problem, a binary segmentation problem applied to satellite imagery. The presented resource consists of pre- and post-fire Sentinel-2 L2A acquisitions of California forest fires that took place starting in 2015. Raster annotations were generated from the data released by California's Department of Forestry and Fire Protection. Moreover, in conjunction with the dataset, we release three different baselines based on spectral indexes analyses, SegFormer, and U-Net models.

CaBuAr: California Burned Areas dataset for delineation

TL;DR

The paper presents CaBuAr, an open dataset for burned-area delineation using Sentinel-2 pre- and post-fire imagery over California, paired with rasterized ground-truth masks from CALFIRE. It establishes three baselines—spectral-index analysis, SegFormer, and U-Net—to explore segmentation and change-detection tasks. Experimental results show deep learning models, particularly U-Net and SegFormer-B0, outperform index-based methods, while index-based approaches remain limited in reliability. The work provides a valuable resource for automated monitoring and recovery planning and plans for future expansion to other regions, modalities, and resolutions.

Abstract

Forest wildfires represent one of the catastrophic events that, over the last decades, caused huge environmental and humanitarian damages. In addition to a significant amount of carbon dioxide emission, they are a source of risk to society in both short-term (e.g., temporary city evacuation due to fire) and long-term (e.g., higher risks of landslides) cases. Consequently, the availability of tools to support local authorities in automatically identifying burned areas plays an important role in the continuous monitoring requirement to alleviate the aftereffects of such catastrophic events. The great availability of satellite acquisitions coupled with computer vision techniques represents an important step in developing such tools. This paper introduces a novel open dataset that tackles the burned area delineation problem, a binary segmentation problem applied to satellite imagery. The presented resource consists of pre- and post-fire Sentinel-2 L2A acquisitions of California forest fires that took place starting in 2015. Raster annotations were generated from the data released by California's Department of Forestry and Fire Protection. Moreover, in conjunction with the dataset, we release three different baselines based on spectral indexes analyses, SegFormer, and U-Net models.
Paper Structure (11 sections, 7 figures, 4 tables)

This paper contains 11 sections, 7 figures, 4 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Dataset
  4. Preprocessing
  5. Manual inspection
  6. Task
  7. Experiments
  8. Experimental settings
  9. Spectral Indexes
  10. Deep learning models
  11. Conclusion

Figures (7)

  • Figure 1: California administrative boundaries (red) vs satellite tiles of the proposed dataset (blue).
  • Figure 2: Geographical distribution of wildfires (red) within the California boundaries (blue).
  • Figure 3: Example of pre-fire and post-fire RGBs and relative masks.
  • Figure 4: Sample of post-fire RGBs and masks with the associated comments.
  • Figure 5: Burned pixels percentage per image per fold.
  • ...and 2 more figures