paper.md

title	ForeFire: A Modular, Scriptable C++ Simulation Engine and Library for Wildland-Fire Spread
tags	C++ Python wildfire simulation wildland fire forest fire bushfire fire behaviour fire spread front‑tracking discrete‑event simulation DEVS combustion fire physics numerical solver numerical modeling high‑performance computing parallel computing MPI fire‑atmosphere coupling atmospheric modeling mesoscale meteorology weather coupling data assimilation fuel models hazard prediction risk assessment decision support emergency management operational forecasting software library python bindings command‑line tool simulation engine model validation benchmarking machine learning neural networks smoke modeling
authors	name orcid affiliation corresponding Jean-Baptiste Filippi 0000-0002-6244-0648 1 true name orcid affiliation corresponding Roberta Baggio 0000-0003-1531-0530 1 false name orcid affiliation corresponding Ronan Paugam 0000-0001-6429-6910 2 false name orcid affiliation corresponding Frédéric Bosseur 0000-0002-8108-0887 1 false name affiliation orcid corresponding Antonio Leblanc 4 0009-0001-9043-2039 false name orcid affiliation corresponding Alberto Alonso-Pinar 0009-0009-2051-9700 1, 3 false
affiliations	name index ror SPE, UMR 6134, CNRS, University of Corsica Pascal Paoli, Corte, France 1 016nwev19 name index ror Centre for Technological Risk Studies, Department of Chemical Engineering, Universitat Politécnica de Catalunya 2 03mb6wj31 name index ror Faculty of Science, The University of Melbourne, 4 Water St., Creswick, Australia 3 01ej9dk98 name index umgrauemeio 1.5°C, Brazil 4
date	25 May 2025
bibliography	paper.bib

Summary

Wildfire forecasting is both an active research area and an important need for decision support systems. ForeFire is a modular, high-performance, scriptable, discrete‑event‑driven simulation engine [@filippi2009] focusing computational effort on the active region of a fire front defined as a dynamic mesh (or multipolygons) of fire markers. It is designed to model the spread of wildfire perimeters over large landscapes at meter scale resolution in seconds, serving both as a research platform and a tool for operational forecasting. The core C++ library has Fortran and Python bindings and is accompanied by a lightweight scriptable interpreter (a custom FF language), can load, save and export data in NetCDF, GeoJson, KML, PNG and JPG, and includes a local HTTP service with customizable graphical user interface. ForeFire can also account for fire-atmosphere interaction by two-way coupling with the MesoNH [@lac2018] atmospheric model [@filippi2013].

Statement of need

Wildfire modeling tools have historically been split between complex combustion research models and streamlined operational tools, each with distinct limitations. Computational combustion and fluid dynamics (CFD) based models (e.g., FIRETEC [@linn2005] or WFDS [@mell2007]) are highly computationally intensive and yet unable to provide large wildfire forecasts faster than real time. Atmospheric coupled codes, such as WRF/SFire [@mandel2011] must be run within an atmospheric model and require a large amount of processing power and data. Operational wildfire simulators, such as widely used Farsite [@finney1998] (now Flammap [@Finney2023FlamMap]), or Canadian Prometheus [@garcia2008], are able to simulate fire fronts spanning tens of kilometers in a matter of seconds, but have definite built-in modeling assumptions and are distributed as compiled software with graphical interfaces with limited scriptability. Other open source libraries are ElmFire [@lautenberger2013] or Cell2Fire [@pais2021] that are tied to a single spread model and do not include scripting language, or deep learning based [@XIA2025106401].

ForeFire was developed as a community tool to fill the gap between highly complex customizable models and more rigid operational tools: a unified wildfire simulator that is both adaptable (highly scriptable with multiple bindings) and high-performing (discrete‑event‑driven simulation with dynamic mesh allows to concentrate computation at meter scale resolution only on the active part of the front to perform speed over 100Ha per second on a single CPU). It is intended to serve both as a research platform and a tool for operational forecasting.

Typical Use Cases

Rapid prototyping of new models

ForeFire implements several standard fire flux and spread rate models, such as Rothermel [@andrews2018] or Balbi [@balbi2009], but also makes it trivial to switch, extend or add to this base with a single .cpp using any existing model file as a template. Internally data is handled as layers that can come from a NumPy array, read from NetCDF or generated on the fly by ForeFire (e.g. slope derived from the elevation layer, fuel loaded as index map with tabulated fuel — with part of [@Scott2005] fuel table already available). Developing a Rate Of Spread wildfire model was the original purpose of this simulation code and helped to iterate versions of the Balbi Rate Of Spread formulation on case studies in [@balbi2009] and [@santoni2011]. It also served to implement various heat and chemical species flux models used for volcanic eruption in [@filippi2021], plume chemistry [@strada2012] or industrial fires in [@baggio2022]. In addition, the code includes a generic ANNPropagationModel, which implements a feedforward artificial neural network (ANN) that expects a pre-trained graph file.

Batch simulations with the ForeFire scripting

Custom FF language allows users to easily generate multiple scenarios, including fire-fighting strategies, model evaluation [@filippi2014], ensemble forecasts [@allaire2020] or generate a deep learning database [@allaire2021]. A FF script is a set of scheduled instructions that are interpreted real-time, advancing the simulation clock with a step[dt=] or a goTo[t=] command. Each of these commands (such as goTo[t=42], include[state.ff], startFire[lonlat=(-8.1, 39.9,0)]@t=42, setParameter[propagationModel=Rothermel] or plot[parameter=speed;filename=ROS.png]) can also be called from HTTP, C++, Fortran or Python, and constitutes the core logic of the library. Help and autocompletion are directly available in the interactive shell interpreter that also includes a batch mode. The graphical user interface is web‑based through an embedded HTTP service (command listenHTTP[host:port]) with user‑defined or default pages as shown in \autoref{fig:gui}.

By utilizing pre-compiled datasets over extensive regions, this approach supports continent-wide operational forecasting services. It has been deployed to identify optimal escape routes [@kamilaris2023], integrated into the French National WildFire Decision Support System OPEN DFCI, showcased on the FireCaster demonstration platform, and also currently used in commercial simulation services AriaFire Firecaster, umgrauemeio Pantera and Ororatech FireSpread.

Two-way coupling with the MesoNH atmospheric model

The same scripts can be executed in coupled mode with the Open-Source atmospheric model MesoNH [@lac2018] with fire propagating using surface fields (wind) from MesoNH and forcing heat and other flux fields into the atmosphere. An idealized coupled simulation can be run on a laptop at field scale [@filippi2013], but also on a supercomputer to forecast fire-induced winds of large wildfires [@filippi2018], fire-induced convection [@couto2024], [@campos2023] or even to estimate wildfire spotting [@alonsopinar2025].

Coupled simulations generate gigabytes of 3D data that can be converted to VTK/VTU files using Python helper scripts to visualize in the open-source tool ParaView as shown in \autoref{fig:coupled}.

Acknowledgements

This work has been supported by the Centre National de la Recherche Scientifique and French National Research Agency under grants ANR-09-COSI-006-01 (IDEA) and ANR-16-CE04-0006 (FIRECASTER). The authors thank all contributors and collaborators who have assisted in the development and testing of the ForeFire software.

References