Data to reproduce results from: "Forecasted increases in fire occurrence in Victoria under climate change"

# README: Spatially-explicit predictions of future fire occurrence in Victoria, Australia<br>**Scripts and data for analysis of the manuscript:** *Spatially-explicit predictions of future fire occurrence in Victoria, Australia*<br>**Authors:** Charlie Hart, Andrew Dowdy, Sarah C. McColl-Gausden, Hamish Clarke, Luke Collins, Amelia French, Trent D. Penman, Nevil Amos, Angie Haslem, Cindy E. Hauser, Jim Thomson, Josephine MacHunter, Matt White &amp; William L. Geary<br>**Corresponding author:** William L. Geary ([e-mail])<br>---<br>## Description of the Data and File StructureThis repository contains the code and key data required to replicate the modelling and figures presented in the manuscript.<br>**Note on Replication:** While the full workflow is documented for transparency, Scripts 1–4 and 9 rely on external raw data (DEECA spatial layers, AGCD/AWO climate grids, etc.) that are not included in this repository due to size and licensing restrictions.<br>To replicate the core results:* **Run Step 5** to fit the Boosted Regression Trees (BRTs).* **Run Step 6** to prepare cross-validation data.* **Run Step 8** to consolidate and summarise the cross-validation performance.* **Run Step 10** to generate the spatial fire probability predictions.* **Run Step 11** to recreate the final publication figures.<br>---<br>## Folder: scriptsThis folder contains the sequential workflow for the analysis.<br>* **`fun_fit_forced_brt.R`** - Helper functions for fitting BRT models.* **`Step 1 Make analysis mask.R`** - Generates the 75m study region mask and stratified random sampling points.* **`Step 2_Make burned area rasters.R`** - Processes Victorian Fire History into annual binary burned area layers.* **`Step 3a_SPEI from AWO grids.R`** - Calculates 12 and 24-month SPEI from climate data.* **`Step 3b_Make covariates.R`** - Processes fire weather, ignitions, topography, and Time Since Fire (TSF) predictors.* **`Step 3c_Make covariate stacks.R`** - Harmonises all spatial layers into unified annual stacks.* **`Step 3d_Covariate VIc.R`** - Conducts Variance Inflation Factor (VIF) analysis.* **`Step 4a_Extract data.R`** - Extracts covariate values to sampling points for model training.* **`Step 5_Fit the model_lr_0.005.R`** - Fits primary BRT models (learning rate 0.005).* **`Step 6_Prep Cross validation Data.R`** - Prepares data folds for model validation.* **`Steps 7a, 7b, 7c (Cross Validation)`** - Batch submission and worker scripts to run model cross-validation in parallel on Spartan.* **`Step 8_Summarise CVs.R`** - Consolidates performance metrics from the cross-validation runs.* **`Steps 9a, 9b, 9c (Future Climate Prep)`** - Downscales climate projections, generates future SPEI maps, and compiles future prediction stacks.* **`Step 10_Current and Future Fire Predictions.R`** - Generates spatial probability rasters for baseline and future scenarios.* **`Step 11_Plots for publication.R`** - Produces final binned maps, delta maps, and regional boxplots.<br>**Note on additional_scripts:** This subfolder contains ad-hoc workflows not required to recreate the primary outputs. These include scripts for downloading raw AWO climate data, rolling cross-validation routines, and preliminary SPEI raster generation. They are included for additional context and methodological completeness.<br>---<br>## Folder: dataThis folder contains essential input files for the analysis.* **Modelling Data:** Extracted covariate values used to train the BRT models.* **Districts:** Spatial boundaries for Victorian fire management districts.* **Random Points:** The stratified random points used for data extraction and model training.<br>---<br>## Folder: outputsThis folder contains final datasets and model objects created at key stages.<br>* **Prediction_Stacks:** Multi-layer annual covariate stacks used to generate predictions.* **Predictions:** Baseline and future fire probability GeoTIFFs (`.tif`).* **Final_Models:** * `brt_base_lr0.005_ffdi_95.RDS` - Base model object. * `brt_full_rac_lr0.005_ffdi_95.rds` - RAC model object.* **Regional_Summaries:** * `regional_fire_risk_summary_RAC.csv` - Mean fire probabilities by district.<br>---<br>## Code and SoftwareAll analyses were performed in **R v4.3.0** or later. The complete, version-controlled codebase, including scripts for data pre-processing, model training (BRTs), and spatial prediction, is hosted on GitHub:<br>**GitHub Repository:** [https://github.com/charliehart4/future-fire-victoria](https://github.com/charliehart4/future-fire-victoria)<br>### Key R Libraries:* **Spatial Data:** `terra`, `sf`, `tidyterra`* **Modelling:** `dismo`, `gbm` (Boosted Regression Trees)* **Visualisation &amp; Utility:** `ggplot2`, `dplyr`, `tidyr`<br>### Usage:To replicate the study, clone the GitHub repository and ensure the 21 GB data archive from Figshare is extracted into the `/data` and `/outputs` directories as specified in the scripts.