metadata
license: cc-by-2.0
pretty_name: PhiSat-2 Burned Area Dataset
size_categories:
- 1K<n<10K
Burned Area Dataset Overview
This dataset is based on Sentinel-2 imagery and provides simulated burned area data at a 4.75-meter spatial resolution. It allows for detailed post-fire landscape analysis in different ecosystems around the world.
- Wildfire events: 115
- Continents covered: 6
- Fire types included: Boreal, temperate, tropical
Classes
Each image is labeled with four classes:
- Background
- Burned Area
- Clouds
- Waterbodies
Including Clouds and Waterbodies helps models handle real-world issues like cloud cover and similar-looking features, improving model performance and generalization.
Class Distribution by Dataset Split
| Class | Test (%) | Validation (%) | Training (%) |
|---|---|---|---|
| Background | 20.0 | 20.4 | 59.7 |
| Burned Area | 19.8 | 20.1 | 60.1 |
| Clouds | 19.4 | 20.1 | 60.5 |
| Waterbodies | 20.8 | 19.4 | 59.8 |
This dataset is well-suited for wildfire mapping, monitoring, and ecological research.