A research team has introduced ACFIRE version 1.0, a new mesoscale modeling framework designed to advance the understanding of the complex, two-way interactions between wildland fires and the atmosphere. By integrating a canopy-resolving atmospheric model with a surface fire behavior submodel, the system addresses a significant gap in current modeling capabilities. While existing wildland fire models vary in scale and complexity, few can simultaneously simulate the effects of a fire on the atmosphere across distances ranging from hundreds of meters to hundreds of kilometers while explicitly resolving scales of atmospheric turbulence within and above a forest canopy. The development of ACFIRE, detailed in Geoscientific Model Development, provides a robust research tool for exploring these dynamics through a coupled approach that computes heat release from a fire dynamically rather than relying on user-imposed estimates.
The ACFIRE framework combines three primary components: the Advanced Regional Prediction System (ARPS), a modified version known as ARPS-CANOPY, and the DEVS-FIRE fire behavior model. ARPS provides the fully compressible, non-hydrostatic dynamical core necessary for simulating atmospheric phenomena across various scales. ARPS-CANOPY builds upon this foundation by incorporating forest canopy drag terms in the momentum and sub-grid-scale turbulent kinetic energy equations, alongside canopy element heating and cooling terms in the thermodynamic equation. These modifications allow the model to account for the aggregate effect of individual trees on atmospheric flows without requiring the computational cost of resolving flow around every individual leaf or branch. The integration of DEVS-FIRE, a raster-based cellular automata model, enables the simulation of surface fire spread based on Rothermel’s equations, with fire-atmosphere coupling facilitated by exchanging wind and heat flux data at each model grid point.
To evaluate the model, the researchers simulated a management-scale prescribed burn conducted in the New Jersey Pine Barrens in March 2019. This 11.2-hectare burn experiment provided an extensive observational dataset, including ground-level fire-tracker sensors for characterizing fire spread, multi-spectral sensors for flame intensity, and six instrumented flux towers to measure turbulence and energy fluxes above the fire front. The simulation was initialized at 14:00 EDT and active burning was simulated until 17:00 EDT, with the fire ignited along the southwestern perimeter of the unit between 14:57 and 15:07 EDT. By utilizing 25 Intel Xeon CPUs on a local cluster, the team performed this three-hour simulation, which served as a preliminary test of the coupled model’s performance in representing atmospheric responses to a surface fire in a forest setting.
Comparing the simulated results to observations from the six flux towers, the study found that the ACFIRE simulation successfully reproduced the mean and turbulent flow through the forest canopy. A notable improvement was observed in the representation of upper-canopy turbulent kinetic energy (TKE) compared to previous simulations using an imposed heat source. Specifically, the model showed that the introduction of dynamic two-way coupling significantly better captured these turbulent flow characteristics, which are crucial for accurately simulating the fire-perturbed atmosphere. While there was some overestimation of temperature and other variables in the lower canopy at certain towers, the researchers attributed this to the specific implementation of the sensible heat flux into the lowest model layer, highlighting an area for future optimization.
Regarding fire spread, the ACFIRE simulation correctly predicted the general southwest-to-northeast movement of the fire through the burn unit. The fire front reached the northeastern perimeter between 16:27 and 16:37 EDT, closely matching the timing observed during the field experiment. However, the simulation did not capture all the finer-scale complexities of the observed fire perimeter. For example, it failed to reproduce two specific unburned islands and several smaller “fingers” of fire spread observed by the fire-tracker array. Furthermore, the range of fire spread rates in the model, which fell between approximately 2.3 and 3.7 meters per minute, was narrower than the observed range of 0.4 to 6.6 meters per minute. The researchers noted that these differences are likely linked to the use of horizontally homogeneous fuels in the simulation and the limitations of the 30-meter grid spacing.
The study also examined the spatiotemporal evolution of sensible heat flux and wind speed. The analysis of the coupling between the 3-meter grid used for fire behavior and the 30-meter atmospheric grid showed that the interpolation procedures successfully maintained the physical integrity of the fire-perturbed atmospheric flow. Animations and static frames revealed the development of complex features, such as a bowed fire front similar to parabolic shapes observed in real-world fires, although these features were transient. The researchers emphasized that the absence of aliasing or distortion during the data exchange process between the two models confirms the stability of the coupled framework for this preliminary case study.
The researchers acknowledged several limitations in the current ACFIRE framework. Most notably, the model is currently restricted to surface fire spread and does not simulate torching or crown-fire propagation, nor does it account for forest fuel heterogeneity in this initial version. Additionally, the study’s reliance on a 30-meter grid resolution means that only very large turbulent eddies are explicitly resolved, while all finer-scale turbulence is parameterized. The authors emphasized that these are limitations of the current study’s configuration rather than inherent flaws in the ACFIRE framework, which is designed to be flexible enough for higher spatial resolution in future applications.
Future work is planned to address these limitations and enhance the model’s capabilities. Key next steps include implementing crown fire modeling approaches, such as those developed by Scott and Reinhardt, and introducing heterogeneous fuel data to better capture the observed unburned islands and variations in spread rates. The team also intends to distribute sensible heat flux vertically within the forest canopy rather than placing it entirely in the lowest grid layer, which is expected to improve lower-canopy atmospheric predictions. Testing the model with fires in complex terrain and those covering larger burn areas remains a priority for determining the long-term utility of ACFIRE for operational fire management and smoke dispersion studies.
Ultimately, the successful development and evaluation of ACFIRE version 1.0 demonstrate the feasibility of dynamically coupled fire-atmosphere modeling as a tool for research. By providing a controlled environment to explore the mechanisms of fire-fuel-atmosphere interactions, ACFIRE contributes to the broader goal of building more accurate, operationally viable fire-prediction systems. The researchers hope that insights gained from this framework will help refine the parameterizations used in simpler, high-efficiency models, enabling them to better anticipate the dynamic and transient fire behaviors that occur in real-world landscapes.
To construct the unified framework, the developers resolved several structural and numerical hurdles between the Fortran-based atmospheric model and the Java-based fire model. In DEVS-FIRE, the authors replaced a finite-difference approximation of fuel mass loss with an exact time derivative, preventing the data exchange interval from distorting calculated sensible and latent heat fluxes. They also corrected a scheduling flaw that previously barred ignitions after initialization. Within ARPS-CANOPY, the team fixed a meridional wind interpolation defect that generated artificial kinks along simulated fire perimeters, and introduced global array declarations alongside message-passing interface routines to enable parallel execution across high-performance computing clusters. Furthermore, they added an explicit vertical interpolation step to evaluate horizontal wind fields at exactly six meters above ground level. This modification ensured direct consistency with the reference height required by standard fire behavior equations, departing from earlier configurations that extracted winds directly from the second-lowest atmospheric grid layer regardless of its actual vertical elevation.
The operational setup of the New Jersey simulation incorporated specific adjustments to bridge atmospheric turbulence and ground-level fire spread. The models communicated every second using disk-based text files, establishing a continuous record of exchanged flux and wind fields while remaining shorter than both the flaming residence time and the vertical turbulent timescale. To counteract a known tendency of ARPS-CANOPY to overestimate near-surface winds, the researchers introduced a case-specific bias correction factor of 0.7 to the interpolated six-meter winds before transferring them to DEVS-FIRE. Sensitivity experiments demonstrated that this adjustment lowered simulated spread rates across the burn unit by approximately 33 percent, producing closer agreement with field observations. The model then applied an unsheltered wind adjustment factor of 0.4 for standard fuel model 7, representing southern rough with a 40 percent moisture of extinction, to compute mid-flame winds while allowing overstory vegetation drag to remain explicitly governed by canopy terms.
Statistical comparisons provided deeper quantitative validation of the coupled system alongside defined operating bounds. Across the thermocouple array, the simulated fire arrival times exhibited an overall bias error of 0.9 minutes and a root-mean-square error of 17.6 minutes, showing that the model tracked the real-world burn front with an average delay of under one minute. At the western flux tower, the root-mean-square difference for upper-canopy turbulent kinetic energy dropped from 42.4 percent under the previous imposed-heat formulation to 36.6 percent under ACFIRE. Because the underlying governing equations assume small perturbations relative to the base state and constant canopy drag coefficients, the authors advised restricting ACFIRE to surface fires with one-minute mean sensible heat fluxes at or below roughly 155 kilowatts per square meter to preserve physical validity.
Subject of Research: Development and preliminary evaluation of the ACFIRE mesoscale wildland fire-atmosphere modeling system.
Article Title: Development of ACFIRE version 1.0: a mesoscale model with forest canopy and fire behavior submodels
Article References: Kiefer, M. T., Zhong, S., Charney, J. J., Bian, X., Heilman, W. E., Seitz, J., Skowronski, N. S., Clark, K. L., Gallagher, M. R., Patterson, M., Cole, J., Mueller, E. V., & Hu, X. (2026). Development of AC FIRE version 1.0: a mesoscale model with forest canopy and fire behavior submodels. Geoscientific Model Development, 19(22), 10897-10915. https://doi.org/10.5194/gmd-19-10897-2026
Image Credits: AI Generated
DOI: 10.5194/gmd-19-10897-2026
Keywords: ACFIRE, wildland fire, mesoscale modeling, fire-atmosphere interactions, forest canopy, turbulent kinetic energy, prescribed burn, New Jersey Pine Barrens, two-way coupling, DEVS-FIRE, ARPS-CANOPY, fire spread modeling, atmospheric turbulence, sensible heat flux, surface fire, scientific simulation, computational fluid dynamics
Cite Scienmag News
Russell Cooper. (November 17, 2026). New Mesoscale Model ACFIRE Successfully Simulates Forest Fire-Atmosphere Interactions. Scienmag. https://scienmag.com/new-mesoscale-model-acfire-successfully-simulates-forest-fire-atmosphere-interactions/
Russell Cooper. "New Mesoscale Model ACFIRE Successfully Simulates Forest Fire-Atmosphere Interactions." Scienmag, 17 November 2026, https://scienmag.com/new-mesoscale-model-acfire-successfully-simulates-forest-fire-atmosphere-interactions/. Accessed 17 November 2026.
Russell Cooper. "New Mesoscale Model ACFIRE Successfully Simulates Forest Fire-Atmosphere Interactions." Scienmag. November 17, 2026. https://scienmag.com/new-mesoscale-model-acfire-successfully-simulates-forest-fire-atmosphere-interactions/








