The wildfire spread prediction problem addressed by CanadaWildFireDaily

Overview of the CanadaWildFireDaily dataset

Abstract

CanadaWildfireDaily is a large-scale dataset and benchmark for daily wildfire spread prediction across Canada, designed to support research on forecasting the evolution of active wildfires. It pairs daily satellite-derived fire perimeters with weather, fuel, and terrain covariates, and provides standardized train/val/test splits along with baseline models to enable reproducible comparison of spread-prediction methods.

Dataset and code

The dataset is available for download, and we release the code for building the dataset in other regions of the world and for the benchmark.

Authors

MMahacine EttahriMila: Quebec AI Institute, GERAD and Polytechnique Montreal, Canada
Hager Radi AbdelwahedHager Radi AbdelwahedMila: Quebec AI Institute, Canada (Project Lead)
Toluwaleke OlutayoToluwaleke OlutayoMila: Quebec AI Institute, Canada
Daniel AloiseDaniel AloiseGERAD and Polytechnique Montreal, Canada
Gaétan Marceau CaronGaétan Marceau CaronMila: Quebec AI Institute, Canada

For questions, please contact us at: hager.radi@mila.quebec

Citation

Coming soon