Palisades & Eaton Fires

Satellite Burn Severity Analysis

A Sentinel-2 Optical Remote Sensing Study

Los Angeles, California | January 2025

Using multispectral satellite imagery to map fire extent and assess burn severity

Analysis completed: August 26, 2026

Executive Summary

115.18 km² (28,462 acres) of burn area identified

Key Findings

Low Severity
47.65 km²
41.3% of burned area
Moderate-Low
50.33 km²
43.7% of burned area

Data Quality

  • Post-fire imagery: 0.02% cloud cover (excellent)
  • Spatial resolution: 10 meters
  • Temporal window: 10 days
Source: Sentinel-2A multispectral satellite imagery (ESA Copernicus)

Pre-Fire vs. Post-Fire: Visual Comparison

True-color satellite imagery acquired by Sentinel-2A showing landscape conditions before and after fire event

Pre-fire imagery
Pre-Fire Baseline (Jan 2, 2025)
Healthy vegetation, unburned landscape | 3.0% cloud cover
Post-fire imagery
Post-Fire Assessment (Jan 12, 2025)
Burned and brown vegetation areas visible | 0.02% cloud cover

Interpretation: The post-fire image clearly shows burned areas as brown/darker pixels compared to the pre-fire green vegetation. This spectral change is the basis for computing Normalized Burn Ratio (NBR) and assessing severity.

📥 Image Files:

truecolor_pre.png (493 KB) | truecolor_post.png (496 KB)

Sentinel-2A bands 4 (Red), 3 (Green), 2 (Blue) | 10m resolution

Burn Severity Classification Map

Spatial distribution of fire severity across both fire areas

Burn severity map
Classification Based on dNBR Values

Map Interpretation

  • Green areas: Unburned vegetation, unchanged from pre-fire baseline
  • Yellow areas: Low severity burn, partial vegetation loss
  • Orange areas: Moderate-low severity, significant vegetation damage
  • Red areas: Moderate-high and high severity, complete vegetation loss

📥 Image File: burn_severity_map.png (313 KB)

Classification: 10m resolution | Projection: EPSG:32611 (UTM Zone 11N)

Methodology: The Science

Remote Sensing Approach

  • Normalized Burn Ratio (NBR): Indices derived from NIR and SWIR bands measure vegetation change
  • Change Detection (dNBR): Differenced NBR quantifies burn severity
  • Cloud Masking: Scene Classification Layer removes cloud/shadow pixels
  • Classification: USGS standard severity thresholds applied

Mathematical Foundation

NBR = (NIR - SWIR2) / (NIR + SWIR2)

dNBR = NBRpre - NBRpost

Processing Steps

  • Remote windowed COG read via HTTPS (~3.7 sec per scene)
  • Resample 20m SWIR2 band to 10m NIR grid
  • Apply cloud/shadow mask using SCL classes
  • Compute NBR and dNBR with vectorized operations
  • Classify severity and generate visualizations
Key et al. (1995); USGS GeoMAC burn severity classification

Satellite Data & Results

Sentinel-2 Imagery Used

Role Scene ID Date Cloud Cover
Pre-fire baseline S2A_T11SLT_20250102T183754_L2A Jan 2, 2025 3.0%
Post-fire S2A_T11SLT_20250112T183727_L2A Jan 12, 2025 0.02%

Burn Severity Results

Class Area (km²) Area (acres) % Burned
Low 47.65 11,773 41.3%
Moderate-Low 50.33 12,436 43.7%
Moderate-High 16.63 4,109 14.4%
High 0.58 143 0.5%
Total Burned Area (dNBR ≥ 0.1): 115.18 km² | 28,462 acres
USGS GeoMAC burn severity classification

Validation & Accuracy

Official fire sizes: ~37,000 acres

Our analysis: 28,462 acres (77% match)

Why These Differences?

  • Temporal lag: Post-fire imagery (Jan 12) captured 3+ days after peak activity (Jan 7-9)
  • Threshold sensitivity: dNBR ≥ 0.1 is conservative; very low-severity burns may be missed
  • AOI boundary: Analysis area slightly smaller than official fire perimeters
  • Spectral limitations: Optical data cannot detect partial burns under dense canopy

Data Quality Metrics

Post-fire cloud cover: 0.02%

Exceptional imagery quality for vegetation damage assessment

Cloud/shadow contamination: ≤0.1%

Rigorous masking ensures reliable classification

USGS NIFC InciWeb | NASA FIRMS active fire detections

Limitations & Considerations

Data Limitations

  • Temporal lag: Peak fire activity (Jan 7-9) vs. post-fire imagery (Jan 12) = 3-5 day gap
  • Satellite revisit cycle: Sentinel-2 revisits every 5-6 days
  • Single sensor: Optical data only; Sentinel-1 SAR integration deferred
  • Cloud dependency: Method fails with > 20% cloud cover

Responsible AI Notes

Dataset includes residential properties; maps should be shared with privacy protections

Results should be validated with independent ground truth before operational use

Croissant 1.1 RAI schema

Future Work & Enhancement

Phase 1: Immediate (1-2 weeks)

  • Ground truth validation in high/moderate-high severity zones
  • Multi-temporal Sentinel-2 acquisition to monitor vegetation recovery
  • Cross-validation with NIFC official perimeters and MODIS FIRMS

Phase 2: Integration (1 month)

  • Sentinel-1 SAR coherence for structural damage assessment
  • SRTM DEM terrain analysis for slope/aspect normalization
  • Hydrological risk mapping for erosion/debris-flow hazards

Phase 3: Operationalization (2-3 months)

  • Near-real-time automated service: severity maps within 48 hours
  • Interactive web dashboard for emergency management access
  • Historical analysis (2015-2025) for trend assessment
Sentinel-1 SAR | SRTM elevation | PostGIS spatial database

Reproducibility & Reuse

Reusable Pipeline

  • Parameterized script: analyze_burn_severity.py adaptable to any fire event
  • Configuration-driven: Edit AOI bounding box and scene IDs; run automatically
  • Containerized: Conda environment file + Docker image for reproducibility

Documentation

  • Technical report with full methodology
  • Colleague handoff guide (setup, execution, troubleshooting)
  • Reusable workflow template for other fire events
  • GitHub repository with version control

Adaptation Timeline

For a new fire event:

15-30 min to locate scene IDs | 5-10 min to run analysis

GitHub | Git version control | GitHub Actions

Conclusion

Project Outcomes

  • ✓ Identified cloud-free satellite imagery bracketing fire event
  • ✓ Computed burn severity classification for both fires
  • ✓ Validated results (77% agreement with official records)
  • ✓ Generated georeferenced outputs for GIS analysis
  • ✓ Created reproducible pipeline for operational reuse

115.18 km² (28,462 acres) mapped using free satellite data and open-source tools

Key Takeaways

  • Accessibility: Modern remote sensing is practical and affordable
  • Scalability: Pipeline deployable globally for any fire in near-real-time
  • Collaboration: Reproducible workflows enable knowledge transfer
⚠️ CAUTION: This analysis is NOT ready for operational emergency management deployment. Results require field validation, independent verification, and integration with official fire management protocols before operational use.
1 / 11