A 28-class fuel map for Mudge Island, British Columbia, aligning with the Scott and Burgan (2005) Fire Behavior Fuel Model. The map was generated by applying a 1D-CNN to Sentinel-2 multispectral imagery, integrating 12 spectral bands and three vegetation indices. The dataset was created by Ka Hong Lee and last updated in May 2026.
Use Cases
- Prioritizing fuel treatments based on detailed 28-class fuel characterization.
- Assessing community-scale wildfire risk on small islands using the generated fuel map.
- Supporting fire behavior modeling with spatially detailed fuel type data.
- Evaluating CNN classification frameworks for fuel mapping using multispectral data.
Strengths
- Internal validation achieved an overall accuracy of 86.9% and an F1 score of 87.3%.
- The framework integrates 12 spectral bands and three vegetation indices (NDVI, EVI, NDWI) for classification.
- The final output is a detailed 28-class fuel map aligned with an established fire behavior model.
Limitations
- External validation using four independent datasets showed most accuracy metrics falling below 70%.
- Column-level documentation is absent; field semantics must be inferred after download.
- The dataset's performance varied significantly, with accuracy for some classes ranging from 0% to 18.2%.
Provenance
- Source
- Borealis Harvested Dataverse
- Collection Method
- Generated by applying a 1D-CNN to Sentinel-2 multispectral imagery, integrating spectral bands, vegetation indices, and biophysical features.
- Freshness
- Last updated 2026-05-02 04:11:32; freshness should be verified.
- Geography
- Mudge Island, British Columbia, Canada