Dataset for paper titled "Scalable, Explainable, and Economically Valuable: A Sub-Hourly Data-Driven Wildfire Potential Forecast for Diverse Australian Climates"

Dataset for paper titled "Scalable, Explainable, and Economically Valuable: A Sub-Hourly Data-Driven Wildfire Potential Forecast for Diverse Australian Climates"Author: Alberto Ardid (University of Canterbury, Christchurch, New Zealand)Contact: [e-mail] ------------------------------------------------------------------------------- OVERVIEW--------This codebase provides the scripts, data, and models used to develop and evaluate a machine-learning (ML) wildfire potential forecast system for diverse Australian climates. The focus is on sub-hourly prediction and quantifying the economic value of forecasts, as described in the associated paper. The ML model is trained on weather station data from Australia's Sunshine Coast, Brisbane, and Hobart, and is benchmarked against the Australian Fire Behaviour Index (FBI). The repository includes tools for model evaluation, cross-validation, and a Potential Economic Value (PEV) framework to assess operational benefit in financial terms. ------------------------------------------------------------------------------- FOLDER STRUCTURE---------------- code_to_share_natcomms_fire_gen/│├── main_cross_correlation_cve_cross_regional.py├── main_cross_correlation_cvf_one_sta.py├── environment.yml│├── pev/│ ├── _pev_one_region/│ ├── files/│ ├── pev_fpr_tpr_ml_fbi_one_region.py│ └── pev_one_region/│├── fire/├── forecasts/├── models/├── features/└── data/ ------------------------------------------------------------------------------- KEY COMPONENTS-------------- Main Scripts:- main_cross_correlation_cve_cross_regional.py * Core script for cross-regional evaluation of the ML fire potential forecast. * Compares ML model performance against the FBI across different regions. * Handles data loading, model evaluation, and result aggregation. - main_cross_correlation_cvf_one_sta.py * Focuses on single-station (site-specific) cross-validation. * Useful for assessing model performance and transferability at individual weather stations. Potential Economic Value (PEV) Analysis:- pev/ * Contains all code and results related to the economic value assessment of the fire potential forecasts. * pev_fpr_tpr_ml_fbi_one_region.py: Main script for calculating the Potential Economic Value (PEV) of the ML and FBI forecasts. Computes ROC curves, net savings, and other metrics for a given region. Generates plots and summary statistics for inclusion in the paper. * _pev_one_region/ and pev_one_region/: Output folders containing figures and results from the PEV analysis (e.g., ROC curves, net savings plots). * files/: Contains input data for the PEV analysis, such as fire ignition records and weather data. Data and Features:- data/: Raw input data: weather station records and fire ignition times for each region.- features/: Precomputed feature files (e.g., fuel moisture, fire indices) used as model inputs. Models and Forecasts:- models/: Trained ML models and feature sets for each region and cross-validation fold.- forecasts/: ML and FBI forecast outputs, including consensus predictions and performance metrics. fire/: Utility scripts for data processing, downloading, and feature engineering. ------------------------------------------------------------------------------- GETTING STARTED--------------- 1. Environment Setup - Use the provided environment.yml to create a conda environment with all required dependencies: conda env create -f environment.yml conda activate fireforecast 2. Running Analyses - For cross-regional evaluation: python main_cross_correlation_cve_cross_regional.py - For single-station evaluation: python main_cross_correlation_cvf_one_sta.py - For PEV analysis (within pev/): cd pev python pev_fpr_tpr_ml_fbi_one_region.py 3. Data - All necessary data for reproducing the results is included in the data/ and features/ folders. ------------------------------------------------------------------------------- CITATION--------If you use this code or data, please cite the associated paper. ------------------------------------------------------------------------------- CONTACT-------For questions or collaborations, please contact:Alberto [e-mail]