Tarazona et al. (2026) Advancing burned area mapping using the Normalized Radar Burn Ratio (NRBR)

Identification

• Journal: Remote Sensing Applications Society and Environment
• Year: 2026
• Date: 2026-01-01
• Authors: Yonatan Tarazona, Vasco Mantas
• DOI: 10.1016/j.rsase.2026.101938

Research Groups

• University of Coimbra, Department of Earth Sciences, and Centre for Earth and Space Research-CITEUC, Portugal
• University of Coimbra, Department of Earth Sciences, eoLab, Centre for Earth and Space Research-CITEUC, Portugal

Short Summary

This study assesses the effectiveness of the new Normalized Radar Burn Ratio (NRBR) index, derived from Sentinel-1 C-band Synthetic Aperture Radar (SAR) data using a U-net deep learning model, for burned area detection across diverse land cover types in the Mediterranean climate zone. It demonstrates that radar-based NRBR can match the performance of the widely used optical-based delta Normalized Burn Ratio (dNBR), particularly under adverse atmospheric conditions.

Objective

• Assess the sensitivity of the Normalized Radar Burn Ratio (NRBR) in detecting burned areas across different land cover types, identifying where it demonstrates higher accuracy.
• Demonstrate that NRBR can achieve comparable performance to optical-based delta Normalized Burn Ratio (dNBR) for burned area mapping, leveraging its cloud independence.
• Evaluate the spatial variations of NRBR and dNBR across land cover types.

Study Configuration

• Spatial Scale: Continental Portugal, covering approximately 88572.98 square kilometers, subdivided into 50 km x 50 km tiles for analysis. Regional-scale assessment.
• Temporal Scale: The 2017 fire season. Pre-fire periods were May 15 - June 15, 2017 (optical) and October 15 - November 15, 2016 (radar). Post-fire periods were July 01 - December 30, 2017 (optical) and October 15 - November 15, 2017 (radar).

Methodology and Data

• Models used: U-net deep learning architecture for semantic image segmentation.
• Data sources:
  • Copernicus Sentinel-1 C-band Synthetic Aperture Radar (SAR) data (VV and VH polarizations, ascending pass, 10 m spacing, Ground Range Detected format).
  • Copernicus Sentinel-2 Multispectral Instrument (MSI) optical data (10 m, 20 m, and 60 m spatial resolutions, surface reflectance).
  • Reference burned area data for 2017 from Direção-Geral do Território (DGT), Portugal (vector layers, converted to 10 m raster).
  • European Space Agency (ESA) land cover proportions at 10 m spacing.
  • Google Earth Engine (GEE) for data processing and mosaic generation.

Main Results

• NRBR achieved an Intersection over Union (IoU) of 64.8% and a Dice Coefficient (DC) of 79.3% in forested areas.
• dNBR achieved higher values in forested areas with an IoU of 85.8% and a DC of 91.6%.
• In grassland covers, NRBR outperformed dNBR, achieving an IoU of 57.6% (vs 53.4% for dNBR) and a DC of 73.0% (vs 69.5% for dNBR).
• NRBR provides spatially continuous burned area estimates, unaffected by cloud cover, whereas dNBR predictions were partially affected by cloud cover (up to 12% in some training tiles and 2.4% in a testing tile).
• The accuracy of burned area detection is significantly influenced by land cover proportions, with models performing better when trained and tested on similar landscape compositions.
• Unburned areas consistently exhibited higher NRBR values than burned areas, with the strongest contrasts observed in forests, shrublands, and grasslands. Burned areas consistently showed lower dNBR values across all land cover types.

Contributions

• Provides the first regional-scale assessment of the Normalized Radar Burn Ratio (NRBR) across diverse land cover types.
• Demonstrates that the radar-based NRBR can achieve performance comparable to the widely used optical-based delta Normalized Burn Ratio (dNBR).
• Highlights NRBR's potential as a SAR-based index to complement optical approaches, improving burned area mapping reliability, especially under adverse atmospheric conditions (e.g., cloud cover).
• Offers a robust, all-weather monitoring system for wildfire management, enabling rapid initial assessment crucial for emergency response.
• Facilitates integration into existing multi-source monitoring frameworks due to its compatibility and complementary nature with optical indices.
• Supports scalable and potentially automated burned area mapping through its computational efficiency using the Google Earth Engine and U-Net pipeline.

Funding

• "Fundaç˜ao para a Ciˆencia e a Tecnologia” (FCT) - Portugal, through UI/BD/154831/2023 (PhD project of Yonatan Tarazona Coronel).

Citation

@article{Tarazona2026Advancing,
  author = {Tarazona, Yonatan and Mantas, Vasco},
  title = {Advancing burned area mapping using the Normalized Radar Burn Ratio (NRBR)},
  journal = {Remote Sensing Applications Society and Environment},
  year = {2026},
  doi = {10.1016/j.rsase.2026.101938},
  url = {https://doi.org/10.1016/j.rsase.2026.101938}
}