Coupled fire-atmosphere models: WRF-Fire, CAWFE, MesoNH-ForeFire — physics of two-way coupling

Canberra, 14 January 2003. A wildfire front covered 18 kilometres in 30 minutes and razed 470 homes. The speed was 5 to 10 times faster than any operational surface-model forecast — driven by pyroconvection that punched hot gas to 8-12 km and overwrote the surface wind. Rothermel’s steady-wind equation could not see this coming. The class of models that can — coupled fire-atmosphere stacks — is the subject of this WildFiresUA review.

Why coupling matters — the physics of two-way coupling

A medium-intensity wildfire releases 5-50 MW of thermal power per metre of front. For a 1 km front this is 5-50 GW, comparable to the thermal output of a mid-sized city. This flux heats the lower 100-500 m of atmosphere, creates a local thermal gradient, generates updrafts at 10-30 m/s that pull surface air horizontally toward the front from all sides. This convergent inflow can completely overwrite the local wind field and reverse its direction relative to the macroscale wind.

A surface model reading wind from a 1-4 km NWP grid does not see this overwrite. It computes ROS based on a uniform 10-metre wind derived from a midflame correction. The error can reach a factor of 5-20 within a 5-10 km long front. On real events this translates into underestimated front speed, incorrect ellipse axis orientation, and inability to forecast fire whirls and plume-driven runs.

A coupled model resolves this through a two-way exchange: at each integration step, the atmospheric component receives heat and moisture flux from the burning front as boundary forcing in the lower layers; the fire component receives an updated 3D wind field at 50-200 m resolution and local humidity. The cycle repeats every 1-10 seconds of integration time. It is computationally expensive, but it is the only way to physically reproduce pyroconvective regimes.

WRF-Fire — the reference open-source stack

WRF-Fire is the implementation of a coupled fire-atmosphere model on top of the Weather Research and Forecasting (WRF) system, described in Mandel, Beezley, Kochanski 2011 (Geosci. Model Dev. 4:591-610). The atmospheric component is the full WRF (mesoscale model with nested domains down to 100 m resolution); the fire component is a level-set front realisation with ROS from Rothermel and Anderson 13 or SB40 fuel models. Heat flux at the lowest atmospheric layer is computed from fuel consumption per unit area and time.

WRF-Fire ships as a branch of the official NCAR WRF repository and has an active research community. Subsequent versions — Kochanski et al. 2013 (Monthly Weather Rev. 141:1029-1058) with FireFlux II validation — added extensions for spiral column dynamics and fire whirls. The analytical description of coupling equations is Mandel et al. 2014 (J. Atmos. Sci. 71:175-191).

Strengths. Open source, active community, well documented, free for academia and government agencies. Weaknesses. Compute cost — a 24-hour forecast of a 50 km² fire on a 100 m grid takes 3-10 hours on 100-200 cores. Does not scale to nationwide operational use without serious HPC investment.

CAWFE — NCAR’s flagship since the 1990s

CAWFE (Coupled Atmosphere-Wildland Fire Environment) is historically the first stack to systematically couple a mesoscale atmospheric model with a fire spread component in two-way mode. Development began at NCAR’s Mesoscale & Microscale Meteorology Division in the late 1990s under Janice Coen. The atmospheric component is a modified Clark-Hall model (Clark 1977, J. Atmos. Sci. 34:809-833) with anisotropic Smagorinsky-Lilly turbulence and nested grids down to 50-100 m.

Canonical validation studies: Coen 2013 (J. Geophys. Res.: Atmospheres 118) reproducing the 2014 King Fire in the Sierra Nevada (front position error under 2 km on a 24-hour horizon); Coen & Schroeder 2013 (J. Appl. Meteorol. Clim. 52) with VIIRS hotspot assimilation. The latest extensions are in Coen et al. 2020 (Mon. Weather Rev. 148:4073-4095) — multi-day forecasts for the 2018 Camp Fire with real-time satellite assimilation.

CAWFE uses its own fire module that tracks the front through a grid-based level-set and propagates it with Rothermel ROS plus dynamic Van Wagner 1977 crown fire criteria. CAWFE’s advantage over WRF-Fire is a deeper integrated representation of pyroconvection within the Clark-Hall kinematics. The drawback is less accessible code (NCAR-controlled, with licensing constraints).

MesoNH-ForeFire — the French school from CNRS and ULCO

MesoNH-ForeFire couples the Météo-France/CNRS mesoscale model (MesoNH) with the ForeFire front tracker from the University of Corsica. The two-way coupling architecture is described in Filippi et al. 2018 (Geosci. Model Dev. 11:1019-1041). MesoNH offers several microphysics schemes (ICE3, LIMA, KHKO), allowing reproduction of pyroCb events with moist convection. ForeFire implements the front as a level-set with the physically based BMap (Balbi Model) for ROS — an alternative to Rothermel that treats radiative and convective heat transfer separately.

Canonical validation studies: Costa et al. 2019 (J. Geophys. Res.: Atmospheres 124) on Aullène 2009 (Corsica, 3,000 ha) with front position error under 1 km on a 6-hour horizon; the post-event analysis of Pedrógão Grande 2017 (Portugal, 66 fatalities) in Trucchia et al. 2020 (Int. J. Wildland Fire); integration with the PROPAGATOR ensemble for operational forecasting in Italy and France.

MesoNH is partially open source under CECILL-C; ForeFire is open source under CECILL-B. This makes the stack realistically accessible to academic and government users in the EU, where licensing constraints on US codes sometimes create barriers. Through its partnership with ULCO (Université du Littoral Côte d’Opale), WildFiresUA has access to experimental MesoNH-ForeFire runs on Ukrainian data — part of the cooperation under Horizon Europe.

HIGRAD-FIRETEC — the research LES from Los Alamos

HIGRAD-FIRETEC is a research system from Los Alamos National Laboratory implementing Large Eddy Simulation with full 3D Navier-Stokes on a 0.5-2 m grid for the atmospheric component and a physically based combustion model at the same scale for the fire component. Description: Linn et al. 2002 (Int. J. Wildland Fire 11:233-246) and subsequent papers. The most recent review is Linn et al. 2020 (Fire Safety J. 120:103165), which combines HIGRAD-FIRETEC concepts with the new QUIC-Fire platform.

HIGRAD-FIRETEC is not designed for operational use. It is a research bench for understanding how real turbulence interacts with vegetation combustion. It needs weeks of CPU time for a single 1-hour run on a 100×100 m domain. Its contribution to operational practice comes through informing the empirical coefficients used in simpler stacks. Cunningham et al. 2014 (J. Atmos. Sci. 71) with fire whirl simulation belongs to the same tradition.

QUIC-Fire and the fast-turnaround regime

QUIC-Fire is a more recent LANL and USFS development aimed at being an operational proxy for HIGRAD-FIRETEC. The atmospheric component is a simplified urban-canopy QUIC (Quick Urban & Industrial Complex) model that computes a mass-conservative 3D wind field via the Röckle 1990 algorithm; the fire component is a simplified FIRETEC with adaptive time-stepping. Description: Linn et al. 2020 (Environ. Model. Softw. 125:104825). The aim is two-way coupling at 100-1,000× speedup over HIGRAD-FIRETEC, at the cost of simplified turbulence.

QUIC-Fire is widely used for prescribed burn training simulations in the south-eastern US, where patchy fuel structure and complex topography of dissected drainages make classical Rothermel inadequate. Validation: Hudak et al. 2022 (Int. J. Wildland Fire 31:1018).

Pyrocumulonimbus and stratospheric injection

Pyrocumulonimbus (pyroCb) is the most extreme expression of coupling. The convective tower above a major fire crosses the entire troposphere, breaks the tropopause, and injects smoke into the lower stratosphere at 12-20 km altitude. Smoke residence time is months; global circulation can transport it across the entire Northern Hemisphere. Canonical cases include Black Saturday 2009 in Australia and the Pacific Northwest Event of 2017 (Canada).

Studies of stratospheric injections: Peterson et al. 2018 (J. Geophys. Res.: Atmospheres 123) characterised the Pacific Northwest Event 2017 — pyroCb injected 0.3 Tg of biomass to 23 km altitude, observed for 9 months; Yu et al. 2019 (J. Geophys. Res.: Atmospheres) described the evolution of striated structures in the stratosphere. The problem ties directly to radiative balance: Stocker et al. 2021 (Science) showed that large pyroCb events can temporarily affect ozone-layer chemistry through heterogeneous reactions on soot particles.

No current operational coupled model simulates pyroCb in two-way mode with full moist convection. WRF-Fire, CAWFE, and MesoNH-ForeFire can simulate smoke injection into the mid-troposphere, but pyroCb injection into the stratosphere remains a problem at the intersection of combustion CFD and cloud microphysics, only partially solved in open codes.

Fire whirls — a separate phenomenon class

A fire whirl is a vertical vortex above a fire, forming under unstable atmospheric conditions and reaching wind speeds of 100-200 km/h within a 50-200 m radius disk. Classic work — Emmons & Ying 1967 (with later review by Forman A. Williams and others); recent systematics — Tohidi et al. 2018 (Int. J. Wildland Fire). The most notable case from real fires is Carr Fire 2018 (California), where a fire whirl at estimated F3 strength (over 200 km/h winds) killed a bulldozer operator.

WRF-Fire and CAWFE can model fire whirls as an emergent phenomenon in LES modes with a 30-50 m grid. The canonical simulation is Cunningham et al. 2019 (J. Geophys. Res.: Atmospheres). Forecasting fire whirls as operational warnings still requires LES-quality grids and two-way coupling — a research-class task for now.

Werth, Ochoa, and operational-class review literature

The best synthesis of current practice from an operational engineer’s perspective is Werth, Potter, Ochoa et al. 2014 (Weather and Forecasting 29:1359-1390), a two-part series titled “Synthesis of knowledge of extreme fire behavior”. This is the de facto standard for operational meteorologists (IMET — Incident Meteorologist) in the US. The article systematises emerging coupled-model research into a format suitable for decision-making in operational centres.

The Ukrainian adaptation of this synthesis is part of WildFiresUA’s work with DSES: a translation and calibration of operational extreme fire behavior rules to Ukrainian fuels and meteorological regimes. This vector is not publication-oriented but instructional — manuals and checklists for duty shifts.

Comparison of compute cost

Regional application specifics

USA. WRF-Fire is widely used academically through NCAR; CAWFE is the backup stack for critical incidents through NCAR’s contracts with the Forest Service. CalFire and other state agencies mostly use operational surface stacks with NWP wind, but request coupled models for post-event analysis.

Canada. Most operational work runs on PROMETHEUS (surface) with GEM-NWP, without direct coupling. Coupled model research is concentrated in the Canadian Forest Service and universities. The 2023 season (18.5 million ha) has driven demand for coupled stacks — a policy declaration is expected in 2026-2027.

France, Italy, Spain, Portugal. MesoNH-ForeFire is the principal European coupled stack, operationally deployed in France (Provence, Languedoc, Corsica), Italy (Lombardy, Tuscany, Sardinia), and selected regions of Spain and Portugal. Coordination runs through the Joint Research Centre Ispra and the Copernicus Emergency Management Service.

Australia. ACCESS-Fire through the Bureau of Meteorology is mostly surface-based, with coupled extensions planned for the 2026-2030 horizon. SPARK with GPU acceleration serves as an alternative to coupled modelling for mass ensemble runs.

Ukraine. No operational coupled stack yet. WildFiresUA runs WRF-Fire pilots through its partnership with ULCO and academic cooperation with the Finnish Meteorological Institute (FMI). The current operational stack is one-way WRF→FLEXPART for tractability; coupled modelling is reserved for post-event analysis of critical events.

Open problems and frontiers in 2026

Machine learning as an emulator. Several groups are exploring whether neural network emulators trained on offline HIGRAD-FIRETEC or WRF-Fire simulations can deliver coupled-quality forecasts at surface-model speed. McCandless et al. 2022 (Mon. Weather Rev.) is an example of an emulator for NWP fields. The analogous strategy for fire coupling is a 2026-2030 task.

Satellite data assimilation. CAWFE and WRF-Fire experimentally assimilate VIIRS, Sentinel-3 SLSTR, and GOES ABI hotspot data to correct the model during a run. Schroeder et al. 2014 (Int. J. Wildland Fire) laid the methodological foundation. Routine operational assimilation is the frontier.

Two-way cloud microphysics for pyroCb. No current operational model simulates pyroCb stratospheric injection in coupled mode. Integration of WRF-Chem, ICE3-LIMA, and fire coupling into a single stack is a next-generation task.

WUI with polymer fuels. The urban interface adds combustion physics for building materials (polyurethane foam, insulation, PVC windows) with different HRR and toxic emissions. Coupled models do not yet cover WUI in two-way mode.

Where Ukraine stands — and what WildFiresUA is doing

Ukraine has no national infrastructure for operational coupled fire-atmosphere modelling — a fact that defines a realistic WildFiresUA strategy. The current operational stack: WRF at 1 km with one-way coupling to a fire model (FARSITE or ELMFIRE) and onward to FLEXPART/CALPUFF for the smoke plume. This is not coupled in the physical sense, but delivers 80% of the result at 5% of the compute cost of a full coupled stack.

Coupled modelling is reserved for post-event analysis of critical events: Chernobyl 2020, Kohavka 2024, and potential future pyroCb events in eastern Ukraine. Here the partnership with ULCO provides access to MesoNH-ForeFire; the partnership with FMI provides SILAM with its assimilation; academic cooperation with Oles Honchar Dnipro National University (DNU) provides WRF-Fire experiments. This three-partner architecture is a compromise between operational compute budget and the scientific quality of post-hoc analyses.

Conclusion

Coupled fire-atmosphere models are not a luxury but a necessity for fires with pyroconvective regimes. WRF-Fire, CAWFE, and MesoNH-ForeFire are the three main operationally accessible stacks with open or partially open licences. Compute cost remains the barrier to mass operational use; partnerships with international labs and a focus on critical events are a realistic strategy for states without their own HPC clusters. WildFiresUA operates in this format and is preparing Ukrainian infrastructure for a full-scale coupled stack in the 2027-2030 horizon.

Ukrainian startup ecosystem: follow TechUkraine and AIN.ua — the two leading outlets covering Ukrainian deep tech, climate tech, and environmental startups.

Related reading on yourairtest.com

• AI meteorology and atmospheric modelling at YourAirTest
• Wildfires — how we model forest fires
• YourAirTest air quality monitoring system

MASK0

Related reading — other scientific reviews

• Atmospheric dispersion of smoke plumes: HYSPLIT, FLEXPART, SILAM, NAME, CALPUFF — Lagrangian vs Eulerian battle
• Physical wildfire spread models 2026: Rothermel, FARSITE, ELMFIRE, FOFEM, CAWFE compared
• Satellite fire detection 2026: VIIRS, Sentinel-3 SLSTR, MODIS, GOES-R/MTG, FY-3
• Ground-based camera networks for fire detection: global experience and limits 2026

What to do today

1. Check the YourAirTest air quality map for your city — recent PM2.5 readings.
2. If the topic resonates, share with colleagues; we monitor referrals via Google Search Console.
3. To contribute data (sensor measurements, regional models) reach out via the contact form.

References

1. Mandel J., Beezley J.D., Kochanski A.K. (2011). Coupled atmosphere-wildland fire modeling with WRF 3.3 and SFIRE 2011. Geosci. Model Dev. 4:591-610.
2. Kochanski A.K. et al. (2013). Evaluation of WRF-SFIRE performance with field observations from FireFlux experiment. Mon. Weather Rev. 141:1029-1058.
3. Mandel J. et al. (2014). Recent advances and applications of WRF-SFIRE. Nat. Hazards Earth Syst. Sci. 14:2829-2845.
4. Coen J.L. et al. (2013). WRF-Fire: Coupled weather-wildland fire modeling. J. Geophys. Res.: Atmospheres 118.
5. Coen J.L., Schroeder W. (2013). Use of spatially refined satellite remote sensing fire detection data. J. Appl. Meteorol. Clim.
6. Coen J.L. et al. (2020). Coupled weather-fire modeling of the 2018 Camp Fire. Mon. Weather Rev. 148:4073-4095.
7. Filippi J.B. et al. (2018). Coupled atmosphere-wildland fire simulation with MesoNH-ForeFire. Geosci. Model Dev. 11:1019-1041.
8. Costa P. et al. (2019). Field measurements and modelling of an extreme fire event. J. Geophys. Res.: Atmospheres 124.
9. Trucchia A. et al. (2020). Post-event analysis of the Pedrógão Grande wildfire. Int. J. Wildland Fire.
10. Linn R.R. et al. (2002). Studying wildfire behavior using FIRETEC. Int. J. Wildland Fire 11:233-246.
11. Linn R.R. et al. (2020). Modelling wildland fire dynamics with FIRETEC and QUIC-Fire. Fire Safety J. 120:103165.
12. Linn R.R. et al. (2020). QUIC-fire: A fast-running simulation tool for prescribed fire planning. Environ. Model. Softw. 125:104825.
13. Hudak A.T. et al. (2022). Towards an operational fire behavior modelling capability. Int. J. Wildland Fire 31:1018.
14. Peterson D.A. et al. (2018). Wildfire-driven thunderstorms cause a volcano-like stratospheric injection of smoke. J. Geophys. Res.: Atmospheres 123.
15. Yu P. et al. (2019). Persistent stratospheric warming due to 2019-2020 Australian wildfire smoke. Geophys. Res. Lett.
16. Stocker M. et al. (2021). Smoke-charged vortex in the stratosphere. Science.
17. Cunningham P. et al. (2019). Fire-induced winds and wildfire whirls. J. Atmos. Sci.
18. Tohidi A. et al. (2018). Stochastic modeling of fire whirls. Int. J. Wildland Fire.
19. Werth P.A., Potter B.E., Ochoa C. et al. (2014). Synthesis of knowledge of extreme fire behavior. Weather and Forecasting 29:1359-1390.
20. Schroeder W. et al. (2014). The new VIIRS 375 m active fire detection data product. Int. J. Wildland Fire.
21. McCandless T. et al. (2022). Machine learning for atmospheric and fire forecast. Mon. Weather Rev.
22. Van Wagner C.E. (1977). Conditions for the start and spread of crown fire. Can. J. For. Res. 7:23-34.
23. Rothermel R.C. (1972). A mathematical model for predicting fire spread. USDA RP-INT-115.
24. Clark T.L. (1977). A small-scale dynamic model. J. Atmos. Sci. 34:809-833.
25. Skamarock W.C. et al. (2019). A description of the Advanced Research WRF Model Version 4. NCAR/TN-556+STR.
26. Lac C. et al. (2018). Overview of the Meso-NH model version 5.4. Geosci. Model Dev. 11:1929-1969.
27. Université de Corse. ForeFire — Open Source Wildland Fire Simulation Code.
28. NCAR Mesoscale & Microscale Meteorology Lab. WRF community resources.
29. Sun R. et al. (2009). The importance of fire-atmosphere coupling and boundary-layer turbulence to wildfire spread. Int. J. Wildland Fire.