CanadaFireSat: High-Resolution Multimodal Wildfire Forecasting Benchmark for Canada

Provider
EPFL Environmental Computational Science and Earth Observation Laboratory (ECEO)
License
MIT (source)
Size
3,003 files
Format
Parquet (in sampled files)
Stored at
Hugging Face (https://huggingface.co/datasets/EPFL-ECEO/CanadaFireSat)
Access
Free and public. No account with the data provider is needed.
Category
Environmental Science
Source
https://arxiv.org/abs/2506.08690

About this dataset

Benchmark for high-resolution (100 m) wildfire occurrence forecasting across Canada with multimodal deep learning. Fire targets come from the National Burned Area Composite (NBAC) polygons, mapped to a 2.8 km grid in 8-day windows. Inputs for each sample are a Sentinel-2 Level-1C satellite image time series over the 64 days before prediction (cloud cover under 40 percent) plus environmental predictors: ERA5-Land hydrometeorology, MODIS land surface temperature, MODIS NDVI, EVI, LAI and FPAR, and Canadian Fire Weather Index and Drought Code from Copernicus EMS. The card reports 177,801 samples from 2016 to 2023, with fire in 39 percent of samples, split by year (78,030 train 2016 to 2021, 14,329 validation 2022, 85,442 test 2023), and NALCMS land cover for analysis. The listing (measured 2026-11-14) shows one folder per province or territory (Alberta alone holds 502 Parquet shards, about 182 GB) plus images/ and metadata/; the repository uses about 1.16 TB. Load with Hugging Face datasets or pyarrow. CNN and ViT baselines are on GitHub; an NPY version is EPFL-ECEO/CanadaFireSat-Raw. Released under the MIT licence; cite the ISPRS paper (arXiv:2506.08690).

How to load this dataset in Python

from datasets import load_dataset
ds = load_dataset("EPFL-ECEO/CanadaFireSat")
print(ds)

Data structure

File types in sampled files: Parquet

Files

Example file: Alberta/test-00000-of-00208.parquet

Analyze this dataset without downloading it

On Scigantic, CanadaFireSat: High-Resolution Multimodal Wildfire Forecasting Benchmark for Canada opens in a hosted JupyterLab notebook with the data already mounted, so you can read files where they are stored instead of copying them first. Each notebook includes an AI assistant that can see the mounted files and write and run analysis code. Open the dataset page to launch a notebook, or browse the full catalog.

This page is also available as Markdown and as JSON.

Related Environmental Science datasets