Low-intensity fires mitigate the risk of catastrophic wildfires in California's forests

This repository contains all code and configuration to reproduce the analyses in the paper:

Wu, X., Sverdrup, E., Mastrandrea, M.D., Wara, M.W., and Wager, S., 2023. Low-intensity fires mitigate the risk of high-intensity wildfires in California's forests. Science Advances, 9(45), p.eadi4123. DOI: 10.1126/sciadv.adi4123

Overview

We use a synthetic control approach to analyze twenty years of satellite-based fire activity data across 124,186 km² of forests in California, providing evidence that low-intensity fires substantially reduce the risk of future high-intensity fires.

Repository Structure

• data_processing/ — Scripts to process and harmonize raw geospatial/tabular data for analysis
• balancing/ — Covariate balancing synthetic control (CBPS) and related utilities
• analysis/ — Outcome analysis, regression, and figure/table generation
• Embeddings/scripts/ — Embedding generation, K-selection, and integration with CBPS (Python & R)
• diagnostics/ — Diagnostics scripts and results for pipeline validation
• config/ — Central configuration files (edit config/config.yaml for all parameters)
• env/ — Environment and dependency files for R and Python
• infra/ — Infrastructure utilities (e.g., HPC scripts)
• data/ — Raw, processed, and output data directories (with README files for structure)
• figures/, tables/, docs/ — Project documentation, figures, and tables
• tests/ — Unit and integration tests for pipeline validation

Getting Started

1. Clone the repository
2. Install dependencies
   • R: env/environment-capstone-r-spatial.yml
   • Python: env/requirements-extra.txt
3. Edit configuration
   • Set all parameters in config/config.yaml
4. Run the pipeline
   • Use main.R for end-to-end execution

Pipeline Workflow

1. Data preparation (data_processing/)
2. Embedding extraction and K-selection (Embeddings/scripts/)
3. CBPS with selected controls (balancing/, Embeddings/scripts/)
4. Outcome analysis and figure/table generation (analysis/, figures/, tables/)
5. Diagnostics and reporting (diagnostics/)

Data Sources

Data	Source	Spatial Resolution	Time Resolution	Time Periods
Active Fires	MODIS FIRMS	1 km²	daily	2000–
Meteorological	Daymet	1 km²	daily	2000–
Disturbance Agents	Dataverse	30 m²	yearly	2000–
Fractional Vegetation Cover	Dataverse	30 m²	yearly	2000–
Vegetation Class	CAL FIRE	30 m²	one time	1990–2014
Topography	GMTED	1 km²	one time	2010
Fire Severity	MTBS	30 m²	yearly	2000–
Fire Severity	RAVG	30 m²	yearly	2012–
Prescribed fires	Federal FACTS	unspecified	yearly	2000–
Prescribed fires	CAL FIRE	unspecified	yearly	2000–

All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials and Online Repository. Those interested in the original data can contact the corresponding author.

Reproducibility & Best Practices

• All scripts use logging/message for output (no print/cat)
• All scripts have module-level and function-level docstrings
• All configuration is centralized in config/config.yaml
• All data directories contain README files describing structure
• .gitignore excludes large/intermediate data and outputs

Usage Example

Rscript Embeddings/scripts/04_run_cbps_with_selected_controls.R <year> <selected_units_csv> <output_prefix> <train_start> <train_end> <test_start> <test_end> [flags]

Or run the full pipeline with:

Rscript main.R

Citing This Work

If you use this code or data, please cite:

• Wu, X., Sverdrup, E., Mastrandrea, M.D., Wara, M.W., and Wager, S., 2023. Low-intensity fires mitigate the risk of high-intensity wildfires in California's forests. Science Advances, 9(45), p.eadi4123. DOI: 10.1126/sciadv.adi4123

Contact

• xw2892@cumc.columbia.edu
• erikcs@stanford.edu
• swager@stanford.edu

License & Terms of Use

See LICENSE file. Authors/funders retain copyright.

---

For detailed documentation, see the docs/ directory and in-script docstrings.