PyForeFire

July 28, 2026 · View on GitHub

PyForeFire Logo

PyForeFire provides Python bindings for ForeFire, an open-source wildfire simulation engine written in C++ and developed by CNRS at the Université de Corse Pascal Paoli.

The distribution is named forefire on PyPI; the importable module is pyforefire.

Installation

pip install forefire

Wheels are published for Linux (x86_64, aarch64) and macOS (Apple Silicon and Intel), on CPython 3.9 and newer. They are self-contained: NetCDF and its own dependencies are bundled inside the wheel, so there is nothing to install beforehand and nothing to configure.

Installing also puts the forefire command-line interpreter on your PATH:

forefire -v

What the published wheels do not include

Wheels are built for portability, which means they deliberately leave out two build-time features:

  • MPI coupling is disabled, so wheels cannot drive coupled fire-atmosphere runs with MesoNH.
  • CPU-specific optimisation (-march=native) is off, so the binary runs on any machine of the same architecture rather than only on the build machine.

If you need either, build from source (below).

Building from source

Any platform without a published wheel — Windows, musl-based Linux, or an unusual architecture — falls back to compiling the sdist, which needs a C++ compiler, CMake ≥ 3.15, and the NetCDF C and legacy C++4 libraries:

# Debian/Ubuntu
sudo apt install build-essential cmake libnetcdf-dev libnetcdf-c++4-dev
# Fedora/RHEL
sudo dnf install gcc-c++ cmake netcdf-devel netcdf-cxx4-devel
# macOS
brew install cmake netcdf netcdf-cxx

pip install forefire --no-binary forefire

To build with MPI support and native optimisation, pass the CMake options through:

pip install forefire --no-binary forefire \
  --config-settings=cmake.define.FOREFIRE_ENABLE_MPI=ON \
  --config-settings=cmake.define.FOREFIRE_NATIVE_ARCH=ON

If NetCDF lives somewhere CMake does not look, point at it with --config-settings=cmake.define.NETCDF_HOME=/path/to/netcdf (and NETCDF_CXX_HOME if the C++4 API is installed separately).

Usage

Verifying the installation

import pyforefire as forefire

ff = forefire.ForeFire()
ff.execute("FireDomain[sw=(0.,0.,0.);ne=(300.,200.,0.);t=0.]")
print("PyForeFire installed and domain created successfully.")

If this runs without an ImportError or linking error, your installation is working. Note: you may see warnings about missing fuel tables, which is expected at this stage.

Running a simple simulation

This example starts a fire in the centre of a domain and runs it for 1000 seconds.

import pyforefire as forefire

ff = forefire.ForeFire()

# 1. Define a 10km x 10km simulation domain
sim_shape = (10000, 10000)
ff.execute(f'FireDomain[sw=(0,0,0);ne=({sim_shape[0]},{sim_shape[1]},0);t=0]')

# 2. Set a simple propagation model (isotropic, i.e. a perfect circle)
ff.addLayer("propagation", "Iso", "propagationModel")

# 3. Start a fire in the center of the domain
ff.execute(f'startFire[loc=({sim_shape[0]/2},{sim_shape[1]/2},0.0)]')

# 4. Run the simulation forward by 1000 seconds
ff.execute("step[dt=1000]")

# 5. Print the state of the fire front to the console
print(ff.execute("print[]"))

This produces text output describing the location of the fire front nodes.

To generate a circle.kml file for visualization in Google Earth, set the dumpMode parameter before the final print command:

ff["dumpMode"] = "kml"
ff.execute("print[circle.kml]")

More advanced examples

For examples that use real-world data (fuel, topography, wind), see the scripts in the tests/python/ directory of the main repository.

Development

The Python bindings are built from the repository root, together with the C++ core:

git clone https://github.com/forefireAPI/forefire.git
cd forefire
pip install -e .

Re-run that command after touching _pyforefire.cpp or the C++ core. If you iterate often, install the build requirements once and let scikit-build-core recompile on import instead:

pip install scikit-build-core pybind11
pip install -e . --no-build-isolation --config-settings=editable.rebuild=true

To build a wheel without installing it:

pip wheel . -w dist/

The build is driven by scikit-build-core, configured in the root pyproject.toml; the extension module target itself lives in the root CMakeLists.txt behind FOREFIRE_BUILD_PYTHON.

Smoke testing a built wheel

python tests/python/test_wheel.py

Run this against an installed wheel rather than from a build tree: it checks that the extension loads, that its bundled NetCDF resolves, and that a trivial simulation advances.

Troubleshooting

  • NetCDF not found while building from source: install the two NetCDF packages listed above. The C library alone is not enough; ForeFire's DataBroker includes the legacy C++4 header <netcdf>, which ships in libnetcdf-c++4-dev / netcdf-cxx4-devel / netcdf-cxx.
  • Illegal instruction after copying a self-built install to another machine: it was compiled with -march=native. Rebuild with FOREFIRE_NATIVE_ARCH=OFF, or use the published wheel.

License

ForeFire is licensed under the GNU General Public License v3.0. See the LICENSE file.

Project URLs