PREPFIRE

PREPFIRE (PREParation for FIRE simulation) is a Python package for preparing materials for fire simulation and analysis, providing tools for processing fire data, weather conditions, and landscape characteristics. The package implements a complete pipeline for fire risk assessment and simulation.

The output files are intended for the fire spread simulation in FlamMap algorithm.

Features

• Fire data processing and analysis
• Landscape file (lcp) component processing
• Weather data integration with ERA5-Land dataset
• Weather type clustering and analysis
• Interactive API key management for weather data
• Ignition probability map
• Fuel Moistures File (.fms) generation
• Project structure setup utility

Installation

From PyPI (recommended)

pip install prepfire

From GitHub Release

Download the .whl file from the Releases page and install it:

pip install prepfire-0.1.1-py3-none-any.whl

Or install directly from the release URL:

pip install https://github.com/CYglume/PREPFIRE/releases/download/v0.1.1/prepfire-0.1.1-py3-none-any.whl

From GitHub

Install the latest development version directly from the repository:

pip install git+https://github.com/CYglume/PREPFIRE.git

For development

Clone the repository and install in editable mode:

git clone https://github.com/CYglume/PREPFIRE.git
cd PREPFIRE
pip install -e ".[test]"

Project Structure

prepfire_project/
├── input_data/
│   └── region/
│       ├── fires.gpkg              # (Required) Fire ignition points — GeoPackage preferred; .shp also accepted
│       ├── cropping_polygon/       # Optional bounding polygon for the processing extent (one feature)
│       └── lcp_Fuel/
│           ├── elevation.tif       # (Required) File names must match the corresponding string
│           ├── slope.tif           # (Required) File names must match the corresponding string
│           ├── aspect.tif          # (Required) File names must match the corresponding string
│           ├── fuel.tif            # (Required) uint8 fuel-model codes; file names must match
│           ├── canopyCover.tif     # (Required) File names must match the corresponding string
│           ├── canopyHeight.tif    # (Required) File names must match the corresponding string
│           ├── cbh.tif             # (Required) File names must match the corresponding string
│           └── cbd.tif             # (Required) File names must match the corresponding string
└── Processed_data/
    └── region/
        ├── Fires/                  # Spatially filtered fire records (fires_region.gpkg)
        ├── Landscape/              # Reprojected and resampled landscape layers
        ├── Ignition/               # Ignition probability surface (ig_kde.tif) and sample points
        ├── Weather/
        │   ├── CDS/                # Raw ERA5-Land monthly NetCDF downloads
        │   ├── FMS/                # Fuel moisture scenario files (.fms)
        │   ├── p95_extreme_weather_types.csv   # Single-percentile scenario (extreme_percentile=int)
        │   ├── all_mean_weather_types.csv
        │   ├── Extensive_percentile_group_weather_scenarios.csv  # Multi-percentile (extreme_percentile=list)
        │   └── fire_weather_records.csv        # Per-fire extracted weather
        ├── Final/
        │   └── lcp_{region}.tif    # Multi-band landscape file (8 bands)
        └── Sim/                    # Reserved for FlamMap/MTT simulation outputs

Usage

import prepfire

# Set up initial project folder if data is not prepared as desired folder structure
prepfire.setup_project_structure(
    "your_region",               # Required: Name of the region folder
    root_dir="path/to/project",  # Optional: Project root directory (defaults to current directory)
)

# Initialize the pipeline object
user_pipeline = prepfire.PrepFirePipeline(
    region="your_region",           # Required: Name of the region folder
    root_dir="path/to/project",     # Optional: Project root directory (defaults to current directory)
    bound_coords=None,                 # Optional: Bounding box coordinates [xmin, ymin, xmax, ymax]. Default by using extent of fire.shp or cropping polygon (if provided)
    output_crs="EPSG:3035",         # Optional: Output coordinate system
    col_fire_size="Area_ha",        # Optional: Column name for fire size in hectares
    col_fire_date="Date",           # Optional: Column name for fire dates
    extreme_percentile=95,          # Optional: int → single-percentile extreme (e.g. 95 for 95th/5th); list → percentile-group intervals, e.g. [30,60,90] → (0,30],(30,60],(60,90],(90,100]
    lcp_components=[                # Optional: Landscape components for LCP file (Follow the order of landscape file for USGS FlamMap)
        'elevation', 'slope', 'aspect', 'fuel',
        'canopyCover', 'canopyHeight', 'cbh', 'cbd'
    ],
    cds_api_key="your_key",         # Optional: API key for weather data
    time_of_day=["12:00"],          # Optional: Times for weather data download
    buffer_size=20000,              # Optional: Buffer size in meters for weather data
    min_clusters=4,                 # Optional: Minimum number of weather clusters
    max_clusters=10,                # Optional: Maximum number of weather clusters
    fire_months=[5,6,7,8,9,10],     # Optional: Months to consider for fire season
    lcp_resolution=100,             # Optional: Resolution in meters for LCP raster
    done_cds_download=False,        # Optional: Set True to skip ERA5-Land download (merged .nc files already present)
    done_lcp=False,                 # Optional: Set True to skip landscape processing (lcp_{region}.tif already present)
    livePlantMoist = [60, 90],      # Optional: Two values for setting moisture of [live Herbaceous, live Woody] plants in fms file
    weather_variable = [            # Optional: Variable names in fire weather table for data clustering 
        'T', 'RH', 'WS', 'DFMC'
    ],
    fire_weather  = None,           # Optional: Path to an external csv for fire weather information (Only used if skipping CDS download)
    log_level="INFO",               # Optional: Logging level
    random_state=None,              # Optional: Integer seed for reproducible ignition point generation
)

# Start running the process of the pipeline object
results = user_pipeline.prepare_simulation()

# ---------------------------------------------------------------------------------- #

# Or use the quick run function (Run starting from setup_project_structure)
results = prepfire.run_prepare(
    region="your_region",        # Required: Name of the region folder
    root_dir="path/to/project",  # Optional: Project root directory
    **kwargs                     # Optional: Additional keyword arguments for PrepFirePipeline
)

Pipeline Methods

The module provides three main functions:

• setup_project_structure: Create the folder structure for input datasets
• PrepFirePipeline: Create the pipeline object for quick processing datasets (load local data)
  • PrepFirePipeline.prepare_simulation() Main function under pipeline object to run all processing functions
• run_prepare: Incorporate the above two functions to provide a single line function for all processes (use all default values from the package)

The PrepFirePipeline class object provides the following methods:

1. process_fire_data(): Load, spatially filter, and save fire ignition records
2. process_weather_data(): Download (or skip) ERA5-Land data, derive weather variables, extract fire weather
3. generate_weather_types(scenario_tag=None): Cluster fire weather and write scenario CSVs
4. generate_ignition_points(): Generate uniformly random ignition points (default 500,000) within the study boundary. Points are spatially shuffled to remove the west-to-east ordering that GeoSeries.sample_points() introduces internally. Pass random_state to the pipeline for reproducible output.
5. process_landscape(): Reproject, resample, and stack all LCP layers into lcp_{region}.tif
6. generate_fmd_and_simpoints(): Write .fms fuel moisture files and produce the KDE ignition probability surface

• prepare_simulation(): Run the complete pipeline (steps 1–6) in sequence

See function description for full parameter configuration.

generate_weather_types — multi-version scenario outputs

By default each call overwrites any existing scenario CSV with the same name. To keep multiple scenario runs side by side, pass a scenario_tag:

# Keep single-percentile and multi-percentile outputs simultaneously
pipeline.extreme_percentile = 95
pipeline.generate_weather_types(scenario_tag="p95")
# → Weather/p95_extreme_weather_types_p95.csv

pipeline.extreme_percentile = [30, 60, 90]
pipeline.generate_weather_types(scenario_tag="groups")
# → Weather/Extensive_percentile_group_weather_scenarios_groups.csv

CDS API Key

The Copernicus Climate Data Store (CDS) API key can be provided in three ways:

1. Directly in the pipeline initialization:

pipeline = PrepFirePipeline(region="your_region", cds_api_key="your_api_key")

2. Environment variable:

import os
os.environ["CDS_API_KEY"] = "your_api_key"
pipeline = PrepFirePipeline(region="your_region")

3. Interactive prompt: If no API key is provided, the program will prompt you to enter it interactively.

Bounding box

The argument bound_coords in prepfire.PrepFirePipeline takes care of the extent being used for cropping the processed area:

1. bound_coords = [xmin, ymin, xmax, ymax]: Provide quick solution to manually set up bounding coords in degree with default EPSG:4326

2. bound_coords = None (default value): leave the argument blank to trigger the following options:

   1. Extra extent geometry (.shp or .gpkg) existing in folder input_data/region/cropping_polygon/: Use the polygon (must be only one feature in the file) to set the extent for the whole process
   2. No extra extent set up: Use the whole extent from input fire data for the process

Fire Weather

The pipeline downloads ERA5-Land data via the Copernicus Climate Data Store (CDS) and extracts per-fire weather at each ignition location. The extracted records are saved to Processed_data/region/Weather/fire_weather_records.csv (one row per fire event). Two aggregated scenario tables are also saved:

extreme_percentile type	Output files
int (e.g. 95)	p95_extreme_weather_types.csv + all_mean_weather_types.csv
list (e.g. [30, 60, 90])	Extensive_percentile_group_weather_scenarios.csv

If you have your own fire weather table, provide it via fire_weather to skip ERA5 extraction entirely:

prepfire.PrepFirePipeline(
    col_fire_size    = "Fire_size_in_ha",
    col_fire_date    = "Date",
    weather_variable = ['Temperature', 'RH', 'Wind_speed'],
    fire_weather     = "path/to/fire_weather.csv"   # bypasses CDS download and extraction
)

The CSV must contain at minimum one column for fire date, one for fire size in hectares, and columns matching weather_variable.

Testing

Tests verify that the package imports correctly and core functions produce expected outputs using synthetic data — no network access or large input files required.

Setup (one-time): install the package in editable mode with test dependencies. This lets you modify source code and test changes without reinstalling.

cd path/to/PREPFIRE
pip install -e ".[test]"

Windows note: If pip is not recognized, use python -m pip instead (e.g., python -m pip install -e ".[test]"). This ensures the correct Python interpreter is used, especially when multiple versions are installed.

GDAL on Windows: If installation fails with build errors on rasterio or geopandas, these packages require the GDAL C library. The easiest fix is to install pre-built wheels before installing prepfire:

pip install --only-binary :all: rasterio geopandas
pip install -e ".[test]"

The --only-binary :all: flag forces pip to use pre-compiled wheels instead of attempting to build from source. If no compatible wheel is found for your Python version, try upgrading pip (pip install --upgrade pip) or use conda which bundles GDAL automatically.

Run tests:

pytest tests/ -v              # Run all tests with verbose output
pytest tests/ -v -k "Clustering"  # Run only tests matching a keyword
pytest tests/ -v -x           # Stop on first failure (useful for debugging)

A successful run prints green PASSED for each test. If a test fails, pytest shows the exact assertion and traceback to help locate the issue.

Local build test: To verify the full build-and-install cycle of the development version (e.g., when the PyPI release has not yet caught up with recent changes), run the script for your platform from the repository root:

Windows (Command Prompt):

scripts\local_build_test.bat

Linux / macOS / Git Bash:

bash scripts/local_build_test.sh

Prerequisites: python must be on your PATH and the repository must be cloned locally (git clone https://github.com/CYglume/PREPFIRE.git).

The script performs the following steps automatically:

Step	Action
1	Creates a clean .venv_test virtual environment
2	Upgrades pip and installs the build backend
3	Builds the wheel (python -m build) and prints the output filenames
4	Installs the built .whl into the test environment
5	Imports prepfire and all sub-modules from outside the source tree, printing the installed version
6	Calls setup_project_structure() in a temporary directory to verify the project scaffold
7	Removes .venv_test and the temporary directory

A passing run ends with === BUILD AND INSTALL TEST PASSED ===. Any import error or missing dependency is surfaced at step 5.

Cleanup: Testing generates cache files that are safe to delete:

# Remove pytest and Python caches (auto-regenerated on next run)
rm -rf .pytest_cache
find . -type d -name "__pycache__" -exec rm -rf {} +

Acknowledgments

• Copernicus Climate Data Store for weather data
• Contributors and maintainers

Funding and supporting projects

This work was financed by the projects FireCycle (CNS2023-144228) funded by the Spanish Ministry of Science and Innovation (MCIN/AEI/10.13039/501100011033) and by the “European Union NextGenerationEU/PRTR”; and project “Understanding Wildfire Risk: A Local-Scale Assessment Framework in Chile”, financed by the AXA Research Fund. The work has been also supported by project FLARE (PID2023-148568OB-I00).