Eulerian modelling of spotting using a coupled Fire-Atmosphere approach

Alonso-Pinar, Alberto; Filippi, Jean-Baptiste; Guyot, Adrien; McCarthy, Nicholas; Tulet, Pierre; Filkov, Alexander

doi:10.5194/egusphere-2025-4855

Preprints

https://doi.org/10.5194/egusphere-2025-4855

Preprints

17 Dec 2025

| 17 Dec 2025

Status: this preprint is open for discussion and under review for Natural Hazards and Earth System Sciences (NHESS).

Eulerian modelling of spotting using a coupled Fire-Atmosphere approach

Alberto Alonso-Pinar, Jean-Baptiste Filippi, Adrien Guyot, Nicholas McCarthy, Pierre Tulet, and Alexander Filkov

Abstract. Spotting, the process by which burning firebrands are lifted by convection and transported downwind igniting secondary fires. Spotting can become a critical driver of rapid wildfire spread and presents major challenges for prediction and suppression. Coupled fire–atmosphere models, which simulate the two-way interaction between fire behaviour and local atmospheric dynamics, offer a promising avenue to capture such complex processes. In this study, we introduce a computationally efficient Eulerian formulation for firebrand transport and spotting, implemented within the coupled MesoNH–ForeFire modelling framework. Two case studies were analysed: an idealized scenario over flat and hilly terrain to assess wind influence, and a realistic simulation of the 2016 Mt Bolton wildfire in southeastern Australia. The model captured key spotting dynamics and fire spread patterns, producing realistic downwind distances with spot fire timing that slightly preceded observations. A full 8-hour forecast, including spotting, simulates in just less than 3 hours, without optimization. Results demonstrate that this simplified approach provides a credible and time-efficient spotting forecast, supporting its potential for operational wildfire modelling and decision-making.

Received: 01 Oct 2025 – Discussion started: 17 Dec 2025

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Alberto Alonso-Pinar, Jean-Baptiste Filippi, Adrien Guyot, Nicholas McCarthy, Pierre Tulet, and Alexander Filkov

Status: open (until 28 Mar 2026)

Post a comment Subscribe to comment alert

RC1:
'Comment on egusphere-2025-4855', Akli Benali, 24 Feb 2026 reply
The paper shows the implementation of simplified Eulerian-based spotting model over two distinct cases: 1) "synthetic" case studies 2) real case study. The paper needs substancial reviews to ensure it provides a solid contribution. At this stage there is some confusion and the objetives and associated workflow need to be clarified.
Main Comments
Fuel description. L66: “Four main fuel classes were identified: grassland, forest, artificial surfaces and constructions, and water bodies”, these are not fuel classes, these are land cover classes. I don’t understand why this was used, instead of Corine Land Cover. Shrubs are missing. Moreover, I don’t understand why the authors need fuel data from EU land cover considering that they are running idealized cases (you can simply assign a fuel class) and a fire in Australia that one would expect to have detailed fuel data. Which fuel models were used?

L98 “We acknowledge that the Rothermel model was developed for, and is typically applied to, Northern Hemisphere coniferous forests, but our analysis focuses on ForeFire’s performance in spotting transport rather than firebrand generation or forward rate of spread”. This should be in the discussion section under the limitations of the fire spread simulation.

Simulation setup (section 2.3) would benefit from having a clearer separation between the idealized simulations vs. the real case simulation.

L194-195: Rate of spread is expressed in m/min, not ha/min. Please correct.

“Verification of the model” (section 2.5). This section encapsulates two main concerns:
It looks like the authors did a “model comparison” where they compare the results of this new model with another model. It is not clear (from the beginning) why this was done and what kind of added-value this brings to paper. The model comparison should ideally be stated in the objectives of the paper. Reading the paper further, it looks like one of the objectives is to compare computational costs…

It is stated in the objetives that the model will be validated against the extensive observations of Mt Bolton fire. Reading the Methods section there is no reference to “validation” and Figure 3 does not show spotfire landing location. Moreover, “verification”, “model intercomparison” and “validation” are separate things that should be distinguished.

Please revise both aspects so that it is clear to the reader what is intended in each step of the analysis. Verify is not the best term. Use evaluate\assess in the first case, validate in the second.
There are some Methods missing, regarding how the assessment and validation were done. One example: first paragraph of section 3.2.3 refers to Methods and not Results.

Figure 1: where are the “orange contours”? Where is the ignition? What do these range values mean? (this is never mentioned in the paper)

I do not understand why the authors compare surface progressions nor surface rate of spread (section 3.2.1) if the objective of the work is to implement and evaluate a spotting model. This is stated in the objectives (Section 1) but also on the sentence “but our analysis focuses on ForeFire’s performance in spotting transport rather than firebrand generation or forward rate of spread” (section 2.1). Additionally, it doesn’t seem to make much sense to evaluate the progressions considering that “the available combustible fuel was limited to the burn area of the observed fire".

Figure 10: if plume measurements were done on the field, shouldn’t they be displayed here?

Why didn’t you compare the spotting simulations vs. observations considering the two models (Eulerian and LAgrangian)? This would allow a more complete analysis of the quality of the model you propose, besides the gains in computation time.

“it is likely that the ERA5 re-analysis has a time shift with respect to the actual wind behaviour, leading to a time difference in terms of fire progression.” No weather station data available for comparison?

A 30min difference between alert and ignition is quite large. What background information did you use to support this?

Regarding the quality and usefulness of the Method implement to represent spotting: I believe the paper would benefit from a quantitative assessment of the distribution of the estimated accumalated mass over the observed spot fires. Is the distribution over the observed spot fires significantly different than the distribution over "non-spot" locations?

Minor Comments:
- Check Reference formatting. Example (L34): “offered by (Sullivan, 2009))” as opposed to “offered by Sullivan (2009),”.
- What do you mean by “all six degrees of freedom”? L55
- L56…references…
- L70: it does not make sense to say that it spans from grassland to very large fires. Maybe grassland and shrub-forest fires or small to very large fires. Grassland fires can be large fires.
- Methods described what “was” done (L104). Change the remaning text accordingly.
- Equation 1: not all terms are described. Correct “C_i” to match the equation.
- L174: kW.m-2
- L178: looks like this sentence is repeated.
- Don’t understand this sentence “The Mt Bolton namefiles are published accompanying the MesoNH-ForeFire repository as an open-source dataset.”
- L194: “During this phase, the fuel supported intense short range (less than 1 km) spotting”, to avoid discussions around the importance of fuel, I suggest you simply state that intense short range spotting occurred.
- Suggest moving the paragraph in L200 to the top.
- L210: what was the extent (in ha) of the Mt Bolton fire? Move this paragraph upwards, it is generic information about the fire which contrasts with specific information, for example, regarding plume measurements.
- L216-17: repeated.
- Figure 2 shows before it is mentioned
- L307: “had already shifted”
- Figure 4: larger numbers on the axis. What is “Wind Y”? Never mentioned before. What are the units?
- L185: rephrase the last sentence to make it clearer.
- First 2 paragraphs of 3.2.4: Methods. Also part of the 3^rd paragraph
- L452: do you mean “slope”?
- L454: I don’t think your results allow you to state “including the topography in a spotting model would enhance its overall accuracy.” You could do that, if you run Mt Bolton Fire without topography, compared the results and concluded that accuracy was higher when topography was used.
- L455: Make it clear that you are making the transition from the idealized cases to the Mt Bolton fire.
- L458-460: also short distance spotting can be rapidly absorbed by the fire front without being detected. In the first hours your highest “mass deposition” values are close to the fire front.
- L499: I believe the term is “new ignitions” and not “reignition”
- L539: “radar”
- L542: rewrite sentence “consistent advection field in which to embed firebrand transport”
- L544 “with observations”

Reply
Citation: https://doi.org/10.5194/egusphere-2025-4855-RC1
- CC1: 'Reply on RC1', Jean-Baptiste Filippi, 08 Mar 2026 reply
  
  Dear Akli,
  
  Thanks a lot for your helpful comments that will help to greatly improve manuscript.
  
  We are waiting for other reviewers comment before we can make complete answers but just one note: test case is in Australia, thus why we do not use CLC, and basically, most important was to get distribution of spotting fuel vs non spotting fuels.
  
  Regards
  
  Reply
  
  Citation: https://doi.org/10.5194/egusphere-2025-4855-CC1
RC2:
'Comment on egusphere-2025-4855', Anonymous Referee #2, 23 Mar 2026 reply
The paper describes a 3D Eulerian method for firebrand transport used in a coupled fire-atmosphere framework. The Eulerian method is implemented in MesoNH-ForeFire. The model is evaluated against a Lagrangian version of firebrand transport developed by the same authors on two idealized configurations and compared with real observations from an Australian wildfire.
The paper presents a promising framework that appears more efficient than Lagrangian transport while maintaining acceptable accuracy. However, the current version of the manuscript requires some major modifications.
Major comments
I am not convinced that the current modelling approach could qualify as a “spotting model”. A complete spotting model requires four components: generation of firebrands, transport, deposition/accumulation, and ignition. This work provides a transport method but relies on oversimplified generation, lacks details on deposition characteristics, and lacks ignition capabilities. I suggest clarifying this choice in the paper title, specifying that this work focuses on a firebrand transport model rather than a complete spotting model.

21-22: “supporting its potential for operational wildfire modelling and decision making”. This statement oversells the paper's impact, as many components are currently missing for it to be considered a complete spotting model. The results are encouraging but not yet exploitable in an operational context.

Model equations are not very clear. L.107 describes a “diffusion-advection conservation equation,” but Eq. 1 does not show any diffusion term; it is a pure advection equation. L.116 claims that “the source term includes the turbulent diffusion”, can you explicitly explain how? L.559-560 says “diffusivity assumed equal to that of the air flow”, where is the term in the equation? Also, you need to specify the units of each term of the equation in the text. If U is the wind flow and appears in the equation, where does the terminal velocity arise in the equation?

A mass conservation study is missing. This study is mandatory for the verification of every transport scheme. The authors need to verify that the mass of firebrands you inject equals the sum of the mass of firebrands in the atmosphere and the mass deposited (and account for mass flux at lateral boundaries) at all times. If the model is not perfectly conservative, one will need to quantify the degree of non-conservativity and assess whether this could be a limitation.

If I understand correctly, the Eulerian model does not include aerodynamic interactions, and the terminal velocity is only a function of firebrands' density. On the other hand, the Lagrangian model uses a drag model and should account for the particle's aspect ratio. How did you account for this extra degree of freedom in the validation dataset? How did you select firebrands' properties for the Lagrangian model? If the Eulerian model is supposed to represent an average firebrand, shouldn’t your validation dataset contain multiple runs of the Lagrangian model with various aerodynamic properties to account for the diversity of particles and compare the average of this ensemble to the Eulerian results?

The injection parameters seem a bit too arbitrary and unrealistic. A flux of 1 kg/m^2/s for 60s is 60 kg/m^2 of firebrands injected. This way over any wildland fuel load. For comparison, considering your heat flux of 60 kW/m^2 for 250s and a combustion constant of 17.433 MJ/kg (usual for vegetation, value used in WRF-SFIRE), the fuel load you use for the real case is 0.86 kg/m^2, which is 70 times lower than the firebrand load you inject in the atmosphere. Using an unrealistic value can inform the deposition pattern, but the authors cannot reasonably examine the absolute deposited mass (L.456 is hardly defensible to me); one can only study normalized deposition. Moreover, to apply this model to an arbitrary injection value, the authors would need to demonstrate whether the transport is sensitive to the injection rate (how do the deposition pattern/values change at 0.5 kg/m^2/s versus 2 kg/m^2/s?).

130: What element in Fig.7 of Filkov et al. 2017 justifies the delay of 30s? An increase in wind speed is not necessarily perfectly located at the fire front. Why is this delay important for the transport model? Does it change the firebrand landing location?

There is no section about deposition. How do you compute the deposition flux? 227 is not very explicit.

235: Normalized to what quantity? The total mass injected, the mass injected into one cell? This is critical information that needs to be explicitly described.

1. Ticks must be revised as the top right panel is not exploitable due to only one tick being present. Orography of the hill must be shown as contours in 1b. I am not sure to understand how the authors compute “occurrence” with the Eulerian model. L. 238, the distribution may look similar, but one seems to have an order of magnitude difference for the mid-range (5 to 8 km). Can the authors explain it? The Eulerian result exhibits significant dissymmetry. Is it something to be expected from this ignition/spread pattern?

I am not sure about the scientific merit of Section 3.2.4, which compares the computational cost of fire vs. no-fire simulations. The expected computation cost study compares 3 simulations: coupled without firebrand transport (reference), coupled with Lagrangian transport, and coupled with new Eulerian transport. This can be done for idealized and/or real configurations.

Minor comments
17: “assess wind influence”: This claim is misleading as the paper does not examine the effect of ambient wind speed in the idealized configuration. I think the right sentence is “assess orography influence”.

53-55: Cite examples of physical phenomena

118: Thomas et al. 2020 is the reference containing the terminal velocity, not 2017.

120: Add reference for adjustment factor value justification

146: “The wind is assumed to be purely horizontal”, this statement is misleading. The wind flow is not purely horizontal in the plume and over the hill. The inflow boundary condition provides the U and V components, but within the domain, the wind is free to have a vertical component.

174: How did you choose the heat flux intensity and duration values?

195: Are the authors describing the rate of spread (L/T) or the fire growth rate (L^2/T)?

220: “longitudinal and horizontal”; this is unclear. Use consistent naming in the paper; later, you refer to longitudinal and lateral, which seems better.

2: Add a second x-axis to show the correspondence between the number of cells and distance.

3 is difficult to read. I think the authors can make it more readable.

I am not convinced of the interest of Figure 4. It is hard to read (image quality, Y-axis tiny, imposed deduction of the Y-direction with red and blue inverted relative to common use, and white color not centered at 0). I would rather see the fuel map showing where the firebrand generation occurs. I would rather see more global evolution of the fire, as you can plot the fire area evolution over time for both the model and the observations from the perimeters, and overlay the emission period and intensity.

Figure 9: What is the surface used to integrate the firebrand load deposited? Is it the 80x80m cell? To extend the analysis in Fig. 9, I would like to see statistics on the deposition value at each observed spot.

524: Plume advection is not a process of spotting; it is the mode of advection linked to the process of firebrand transport. For me, deposition should be added, as the deposition and accumulation of firebrands can be complex and depend on the landing surface.

Common practice is to order citations chronologically when multiple citations are present (e.g., l. 41-42).

Make sure to have a consistent notation for unit (either kg/m^2 or kg m^-2) throughout the manuscript.

Reply
Citation: https://doi.org/10.5194/egusphere-2025-4855-RC2

Alberto Alonso-Pinar, Jean-Baptiste Filippi, Adrien Guyot, Nicholas McCarthy, Pierre Tulet, and Alexander Filkov

Data sets

Simulation files for Eulerian modelling of spotting using a coupled Fire-Atmosphere approach Alberto Alonso-Pinar et al. https://doi.org/10.5281/zenodo.17241970

Alberto Alonso-Pinar, Jean-Baptiste Filippi, Adrien Guyot, Nicholas McCarthy, Pierre Tulet, and Alexander Filkov

Viewed

Total article views: 506 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	BibTeX	EndNote
315	161	30	506	33	37

HTML: 315
PDF: 161
XML: 30
Total: 506
BibTeX: 33
EndNote: 37

Views and downloads (calculated since 17 Dec 2025)

Month	HTML	PDF	XML	Total
Dec 2025	114	31	10	155
Jan 2026	90	40	3	133
Feb 2026	46	42	9	97
Mar 2026	65	48	8	121

Cumulative views and downloads (calculated since 17 Dec 2025)

Month	HTML	PDF	XML	Total
Dec 2025	114	31	10	155
Jan 2026	90	40	3	133
Feb 2026	46	42	9	97
Mar 2026	65	48	8	121

Viewed (geographical distribution)

Total article views: 488 (including HTML, PDF, and XML) Thereof 488 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 24 Mar 2026

Short summary

Wildfires can spread quickly when firebrands are carried by wind to ignite new fires, a process known as spotting. Predicting this process is vital for safety and firefighting, but it is difficult to model. This paper introduces a new, faster way to simulate spotting using a coupled fire–atmosphere model. Tested on real and ideal cases, our approach reproduced realistic fire spread and timing, showing promise for faster, more reliable forecasts to support wildfire management and decision-making.


Total:	0
HTML:	0
PDF:	0
XML:	0