A key tool used in the State of Wildfire report is now publicly available.
The newly released Probability of Fire (PoF)-Toolbox enables researchers, practitioners and institutions to build, train and deploy their own AI-based wildfire risk models using publicly available or local data.
Wildfire risk is shaped by local climate, fuel conditions, land management and human activity. While the globally trained PoF model used in the State of Wildfire report helps compare events across regions, it cannot always capture regional specificities. The PoF-Toolbox enables users to develop locally trained machine-learning models tailored to their own environments.
Built around a modular and reproducible Jupyter workflow, the toolbox guides users through three core steps:
- Build – assemble weather, fuel and ignition predictors
- Train – fit and validate a machine-learning Probability of Fire model
- Deploy – generate probabilistic fire-risk estimates at the desired spatial and temporal scale
By making this workflow openly available, we aim to support greater transparency, local ownership and scientific innovation in wildfire risk modelling. The PoF-Toolbox empowers users not only to generate predictions, but to explore the drivers of fire risk, test sensitivities and better understand how fire behaviour may evolve under changing environmental and societal conditions.
