dataset · Global

wildfire_global

Hugging Face dataset moritzrengert1/wildfire_global

Claim this listing

Represent wildfire_global? Send a claim for review so we can discuss listing updates. Claims do not automatically grant ownership or verification.

Data offered docs-verified: pre-fire and post-fire Sentinel-2 RGB chips (224×224 px) · dNBR severity map (grayscale PNG) · continuous mean dNBR ground truth · ordinal burn severity label (0 low, 1 moderate, 2 high) · burn vs paired no-fire negative control · continent, country, ecosystem group and latitude band metadata
Coverage docs-verified: Global: six continents, five ecosystem types
Resolution docs-verified: 10 m GSD; 224×224 px chips
Formats docs-verified: PNG · GeoTIFF
Access docs-verified: Free/open, CC-BY-4.0 licence (public Hugging Face dataset)

Use cases llm-inferred: Test VLM burn-severity classification against dNBR labels before using AI screening in post-fire ecology assessments · Benchmark geographic bias in image-based AI tools for multi-region environmental screening · Use pre/post-fire chip pairs to validate change-detection methods for land-management baselines
Audience llm-inferred: remote-sensing and GIS analysts · ML and VLM researchers · ecologists working on burn severity

Constraints llm-inferred: Global sample, not UK-specific; UK relevance needs checking · Small dataset (1K–10K samples) with 224×224 px chips · Built for VLM benchmarking, not operational wildfire or flood mapping · Burn events come from MODIS MCD64A1 (500 m) burned area · Hugging Face dataset viewer unavailable (split names not parsed) · No update frequency or time span stated

Verified 2026-10-15 via HF API: exists; 762 downloads/30d, 0 likes.

Metadata observed 14/10/2026 · research sweep: manual-2026-10-15 (Hugging Face datasets)