pa-carbon-matching: Effectiveness of protected areas for carbon conservation

Statistical sample matching analysis to assess the effectiveness of Spanish protected areas in conserving aboveground biomass carbon (2017-2024).

Quick Start

1. Setup environment:

conda env create -f env.yml
conda activate biomass_pas

2. Run analysis:

# Quick test (1 replica)
python src/runner.py 1

# Full analysis (100 replicas)
python src/runner.py 100

3. Export results:

python src/export_results.py

Results will be in results/ directory as CSV and Parquet files.

What It Does

Compares carbon trajectories inside vs outside protected areas using statistical matching across five scenarios:

1. No Control - Random pairs (baseline)
2. Biomass Only - Match on initial biomass
3. Biomass + Climate - Add climate variables
4. Biomass + Climate + Topography - Add elevation and slope
5. Biomass + Climate + Topography + Accessibility - Add accessibility/remoteness

For each scenario, the code:

• Creates matched PA/non-PA pairs with similar characteristics
• Calculates biomass trends (2017-2024) for each pair
• Compares PA vs non-PA carbon trajectories
• Repeats across 100 replicas for statistical robustness

Repository Structure

├── src/
│   ├── runner.py                    # Main analysis script
│   ├── core_analysis.py             # Clustering and matching functions
│   ├── matching_scenarios.py        # Five matching scenarios
│   ├── dataloaders.py               # Data loading functions
│   ├── utils.py                     # Statistics and output
│   ├── export_results.py            # Convert results to CSV/Parquet
│   ├── config.py                    # Configuration parameters
│   └── visualization/               # Plotting scripts
├── data/                            # Input data (biomass, climate, PAs)
├── results/                         # Output files
├── env.yml                          # Conda environment
└── README.md

Data Requirements

Required data (place in data/ directory):

• Biomass time series (2017-2024, 100m resolution)
• PA boundaries shapefile
• Climate baseline and anomalies
• Topography (elevation, slope)
• Accessibility layer

See the Zenodo repository https://doi.org/10.5281/zenodo.17610642 for detailed data descriptions and access to full dataset.

Configuration

Edit src/config.py to change:

• Edge buffer size (default: 1 km two-sided)
• Cluster size (default: 2000 pixels)
• Pool size (default: 50 pixels)
• PA stratification type (designation or size)

Command Line Options

# Run with different stratifications
python src/runner.py 100 --stratification designation  # By PA type (default)
python src/runner.py 100 --stratification size         # By PA size
python src/runner.py 100 --stratification both         # Both (2x computation)

# Quick test
python src/runner.py 1

Output Files

Main results saved to results/pa_effectiveness_matching_results.npz

After running export_results.py:

• pool_results.parquet - Pool-level results (main dataset)
• replica_summaries.csv - Summary per replica
• overall_statistics.csv - Aggregated statistics
• cluster_quality.parquet - Matching quality metrics

See results/README.md for details on output files.

Requirements

• Python 3.10+
• Main packages: numpy, pandas, geopandas, rasterio, scikit-learn, scipy
• See env.yml for complete list

Contact

diego.bengocheapaz@pm.me