Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Sen2Fire: A Challenging Benchmark Dataset for Wildfire Detection using Sentinel Data

Yonghao Xu, Amanda Berg, Leif Haglund

TL;DR

Sen2Fire introduces a challenging wildfire-detection benchmark by fusing Sentinel-2 multi-spectral data with Sentinel-5P aerosol information, comprising 2466 512×512 patches labeled via MOD14A1 V6.1. The study systematically evaluates how different band selections and spectral indices, notably $NBR$ and $NDVI$, affect detection when used alone or in combination, using a U-Net baseline. Key findings show that carefully chosen band combinations can outperform the traditional all-bands approach, and that incorporating aerosol data generally improves performance, highlighting a new avenue for remote-sensing-based wildfire detection. The dataset and accompanying code provide a testbed for transferability and method development in wildfire monitoring with high practical relevance for environmental monitoring and disaster response.

Abstract

Utilizing satellite imagery for wildfire detection presents substantial potential for practical applications. To advance the development of machine learning algorithms in this domain, our study introduces the \textit{Sen2Fire} dataset--a challenging satellite remote sensing dataset tailored for wildfire detection. This dataset is curated from Sentinel-2 multi-spectral data and Sentinel-5P aerosol product, comprising a total of 2466 image patches. Each patch has a size of 512$\times$512 pixels with 13 bands. Given the distinctive sensitivities of various wavebands to wildfire responses, our research focuses on optimizing wildfire detection by evaluating different wavebands and employing a combination of spectral indices, such as normalized burn ratio (NBR) and normalized difference vegetation index (NDVI). The results suggest that, in contrast to using all bands for wildfire detection, selecting specific band combinations yields superior performance. Additionally, our study underscores the positive impact of integrating Sentinel-5 aerosol data for wildfire detection. The code and dataset are available online (https://zenodo.org/records/10881058).

Sen2Fire: A Challenging Benchmark Dataset for Wildfire Detection using Sentinel Data

TL;DR

Sen2Fire introduces a challenging wildfire-detection benchmark by fusing Sentinel-2 multi-spectral data with Sentinel-5P aerosol information, comprising 2466 512×512 patches labeled via MOD14A1 V6.1. The study systematically evaluates how different band selections and spectral indices, notably and , affect detection when used alone or in combination, using a U-Net baseline. Key findings show that carefully chosen band combinations can outperform the traditional all-bands approach, and that incorporating aerosol data generally improves performance, highlighting a new avenue for remote-sensing-based wildfire detection. The dataset and accompanying code provide a testbed for transferability and method development in wildfire monitoring with high practical relevance for environmental monitoring and disaster response.

Abstract

Utilizing satellite imagery for wildfire detection presents substantial potential for practical applications. To advance the development of machine learning algorithms in this domain, our study introduces the \textit{Sen2Fire} dataset--a challenging satellite remote sensing dataset tailored for wildfire detection. This dataset is curated from Sentinel-2 multi-spectral data and Sentinel-5P aerosol product, comprising a total of 2466 image patches. Each patch has a size of 512512 pixels with 13 bands. Given the distinctive sensitivities of various wavebands to wildfire responses, our research focuses on optimizing wildfire detection by evaluating different wavebands and employing a combination of spectral indices, such as normalized burn ratio (NBR) and normalized difference vegetation index (NDVI). The results suggest that, in contrast to using all bands for wildfire detection, selecting specific band combinations yields superior performance. Additionally, our study underscores the positive impact of integrating Sentinel-5 aerosol data for wildfire detection. The code and dataset are available online (https://zenodo.org/records/10881058).
Paper Structure (12 sections, 2 equations, 5 figures, 2 tables)

This paper contains 12 sections, 2 equations, 5 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Dataset Description
  3. Study Areas
  4. Satellite Data
  5. Data Curation
  6. Methodology
  7. Spectral Indices
  8. Input Strategies
  9. Experiments
  10. Implementation Details
  11. Experimental Results
  12. Conclusions and Discussions

Figures (5)

  • Figure 1: Study areas of the Sen2Fire dataset: Four bushfires happened in the 2019--2020 Australian bushfire season.
  • Figure 2: The relative frequency distribution of the digital number (DN), index, or aerosol values for fire and non-fire samples in the training set. There exist high overlaps (denoted in dark red) between the distributions of fire and non-fire samples.
  • Figure 3: Visualization for patches in the training set. RGB, SWIR, NBR, and NDVI denote different color composites.
  • Figure 4: The ratio of the fire samples to the non-fire samples.
  • Figure 5: Wildfire detection result on the test set. The input patches are concatenated to reconstruct the complete image tile.