pith. sign in

arxiv: 2606.13633 · v1 · pith:MYWAP5Y6new · submitted 2026-06-11 · 📡 eess.SY · cs.LG· cs.SY

Aerial Wildfire Suppression Planning with a Hybrid CNN-Cellular Automata Fire Model

Ion Matei , Maksym Zhenirovskyy , Takuya Kurihana , Rohit Vupala , Anthony Wong This is my paper

Pith reviewed 2026-06-27 05:33 UTC · model grok-4.3

classification 📡 eess.SY cs.LGcs.SY
keywords aerial wildfire suppressionhybrid CNN-cellular automata modelfire spread predictiongradient-based optimizationuncertainty quantificationintervention strategiesBear Fire case study
0
0 comments X

The pith

A hybrid neural-cellular automaton fire model paired with gradient-based optimization designs aerial drop schedules that reduce total burned area under uncertainty.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper introduces a modeling and optimization framework that links a hybrid CNN-cellular automata predictor of wildfire spread to an intervention module that selects aerial drop locations and orientations. The predictor takes terrain, fuel, and wind inputs to forecast spatially varying fire behavior, while the optimizer treats drops of water and retardant as binary actions with continuous parameters that map onto the simulation grid. Water produces an immediate drop in active burning and retardant produces a lasting barrier to future spread. Uncertainty is handled by Monte Carlo sampling of daily states for aleatoric variation and by adding spatially correlated perturbations for epistemic error. A case study on the 2020 Bear Fire demonstrates that the resulting plans produce coherent schedules that shrink the final fire-affected area while permitting explicit uncertainty analysis.

Core claim

The framework combines a hybrid neural-cellular automaton wildfire model that predicts spatially varying spread behavior from terrain, fuel, and wind data with gradient-based design of targeted aerial drops. Binary drop actions with continuous-valued location and orientation parameters are mapped to the simulation grid. Water and retardant receive distinct suppression effects, corresponding to immediate reduction of active burning and persistent reduction of future spread. Robustness is quantified through Monte Carlo sampling of daily fire-state realizations for aleatoric uncertainty and through spatially correlated prediction-error perturbations for epistemic uncertainty. The 2020 Bear Fire

What carries the argument

The hybrid CNN-cellular automata wildfire model that forecasts spread together with the intervention module that maps binary drop actions and continuous location-orientation parameters onto the grid.

If this is right

  • The framework generates coherent aerial suppression schedules.
  • These schedules reduce total fire-affected area.
  • The method supports uncertainty-aware analysis through Monte Carlo sampling of fire states and spatially correlated prediction-error perturbations.
  • Water drops produce immediate suppression while retardant drops produce persistent barriers to spread.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same hybrid predictor-optimizer structure could be tested on other fire events or landscapes to check transferability.
  • Real-time satellite or sensor feeds could be added to update drop plans during an ongoing fire.
  • The uncertainty outputs might be used to compare alternative resource-allocation policies before a fire season.

Load-bearing premise

The hybrid CNN-cellular automata wildfire model accurately predicts spatially varying spread behavior from terrain, fuel, and wind data.

What would settle it

Running the Bear Fire case study with the optimized drop schedule and finding no statistically significant reduction in total fire-affected area compared with a no-drop baseline would falsify the claim that the framework produces effective suppression plans.

read the original abstract

Aerial wildfire suppression requires not only predicting fire spread, but also designing effective intervention strategies under operational and environmental uncertainty. We present a modeling and optimization framework for aerial wildfire suppression that combines a hybrid neural-cellular automaton wildfire model with gradient-based design of targeted aerial drops. The wildfire model predicts spatially varying spread behavior from terrain, fuel, and wind data, while the intervention module determines binary drop actions with continuous-valued location and orientation parameters mapped to the simulation grid. Water and retardant are represented with distinct suppression effects, corresponding to immediate reduction of active burning and persistent reduction of future spread. To evaluate the robustness of the resulting suppression plans, we quantify both aleatoric uncertainty through Monte Carlo sampling of daily fire-state realizations and epistemic uncertainty through spatially correlated prediction-error perturbations. A case study based on the 2020 Bear Fire shows that the framework can generate coherent aerial suppression schedules for reducing total fire-affected area and can support uncertainty-aware analysis of wildfire intervention strategies.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 3 minor

Summary. The manuscript presents a hybrid CNN-cellular automata wildfire spread model combined with a differentiable parameterization of aerial water/retardant drops. Gradient-based optimization is used to select binary drop actions and continuous location/orientation parameters. Robustness is assessed via Monte Carlo sampling of aleatoric uncertainty and spatially correlated perturbations for epistemic uncertainty. A case study on the 2020 Bear Fire is used to illustrate generation of suppression schedules that reduce total fire-affected area.

Significance. If the hybrid model's predictive fidelity and the differentiability of the drop mapping hold, the framework offers a concrete route to uncertainty-aware aerial suppression planning. The explicit use of real-event data, Monte Carlo aleatoric sampling, and spatially correlated epistemic perturbations, together with reported case-study outcomes, constitute a substantive engineering contribution in the eess.SY domain.

major comments (1)
  1. [§4, Table 2] §4 (Case Study), Table 2: the reported reduction in final burned area relative to the no-intervention baseline is given without an accompanying comparison to a non-gradient heuristic (e.g., uniform or wind-aligned drop placement); this comparison is required to establish that the gradient-based schedules are meaningfully superior rather than merely feasible.
minor comments (3)
  1. [Eq. (7)] Eq. (7): the mapping from continuous drop parameters to the discrete grid cells is described only at a high level; an explicit formula or pseudocode block would remove ambiguity about how partial-cell coverage is handled.
  2. [Figure 5] Figure 5: the color scale for the epistemic perturbation field is not labeled with units or range; this makes it difficult to judge the magnitude of the added spatial correlation.
  3. [§3.2] §3.2: the statement that the CNN component is trained on 'historical fire data' should cite the exact dataset and split used, even if only in a footnote.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the positive assessment and the constructive suggestion regarding additional baselines. We address the major comment below and will incorporate the requested comparisons in the revised manuscript.

read point-by-point responses

  1. Referee: [§4, Table 2] §4 (Case Study), Table 2: the reported reduction in final burned area relative to the no-intervention baseline is given without an accompanying comparison to a non-gradient heuristic (e.g., uniform or wind-aligned drop placement); this comparison is required to establish that the gradient-based schedules are meaningfully superior rather than merely feasible.

    Authors: We agree that direct comparisons to non-gradient heuristics are necessary to substantiate that the optimized schedules outperform simpler strategies. In the revised manuscript we will augment Table 2 with results for uniform drop placement and wind-aligned drop placement (both using the same total drop budget and uncertainty quantification protocol), thereby quantifying the incremental benefit of the gradient-based parameterization. revision: yes

Circularity Check

0 steps flagged

No significant circularity

full rationale

The paper presents a hybrid CNN-cellular automata model for fire spread prediction combined with gradient-based optimization for aerial drop scheduling. All load-bearing elements (model predictions, drop parameterization differentiability, uncertainty quantification via Monte Carlo and perturbations) are validated through explicit case-study experiments on the 2020 Bear Fire rather than reducing to fitted inputs or self-citations by construction. No self-definitional loops, fitted quantities renamed as predictions, or uniqueness theorems imported from prior author work appear in the derivation chain. The framework is self-contained against the reported external benchmarks and data.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract-only review yields no explicit free parameters, axioms, or invented entities. The hybrid model implicitly relies on learned CNN weights and standard cellular-automata update rules whose accuracy is assumed but not detailed.

pith-pipeline@v0.9.1-grok · 5716 in / 1048 out tokens · 35172 ms · 2026-06-27T05:33:23.064715+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

25 extracted references · 14 canonical work pages

  1. [1]

    Research Paper INT-115, USDA Forest Service, Intermountain Forest and Range Experiment Station (1972)

    Rothermel, R.C.: A mathematical model for predicting fire spread in wildland fuels. Research Paper INT-115, USDA Forest Service, Intermountain Forest and Range Experiment Station (1972)

    1972

  2. [2]

    doi:10.1016/j.amc.2008.06.046

    Alexandridis, A., Vakalis, D., Siettos, C.I., Bafas, G.V.: A cellular automata model for forest fire spread prediction: The case of the wildfire that swept through Spetses island in 1990. Applied Mathematics and Computation204(1), 191–201 (2008) https://doi.org/10.1016/j.amc.2008.06.046

    work page doi:10.1016/j.amc.2008.06.046 1990

  3. [3]

    Natural Hazards and Earth System Sciences19(1), 169–179 (2019) https://doi.org/10.5194/nhess-19-169-2019

    Freire, J.G., DaCamara, C.C.: Using cellular automata to simulate wildfire prop- agation and to assist in fire management. Natural Hazards and Earth System Sciences19(1), 169–179 (2019) https://doi.org/10.5194/nhess-19-169-2019

    work page doi:10.5194/nhess-19-169-2019 2019

  4. [4]

    Eco- logical Modelling 348, 33–43

    Zheng, Z., Huang, W., Li, S., Zeng, Y.: Forest fire spread simulating model using cellular automaton with extreme learning machine. Ecological Modelling348, 33–43 (2017) https://doi.org/10.1016/j.ecolmodel.2016.12.022

    work page doi:10.1016/j.ecolmodel.2016.12.022 2017

  5. [5]

    Environmental Reviews28(4), 478–505 (2020) https://doi.org/10.1139/ er-2020-0019

    Jain, P., Coogan, S.C.P., Subramanian, S.G., Crowley, M., Taylor, S., Flannigan, M.D.: A review of machine learning applications in wildfire science and manage- ment. Environmental Reviews28(4), 478–505 (2020) https://doi.org/10.1139/ er-2020-0019

    2020

  6. [6]

    Forest Science42(3), 267–281 (1996) https://doi.org/10.1093/forestscience/42.3

    Fried, J.S., Fried, B.D.: Simulating wildfire containment with realistic tactics. Forest Science42(3), 267–281 (1996) https://doi.org/10.1093/forestscience/42.3. 267

    work page doi:10.1093/forestscience/42.3 1996

  7. [7]

    European Journal of Operational Research 316, 488–502 (2024) https://doi.org/10.1016/j.ejor.2024.03.005

    Avci, M.G., Avci, M., Battarra, M., Erdo˘ gan, G.: The wildfire suppression prob- lem with multiple types of resources. European Journal of Operational Research 316, 488–502 (2024) https://doi.org/10.1016/j.ejor.2024.03.005

    work page doi:10.1016/j.ejor.2024.03.005 2024

  8. [8]

    Canadian Journal of Forest Research45, 384–393 (2015) https://doi.org/10.1139/cjfr-2014-0252

    Belval, E.J., Wei, Y., Bevers, M.: A mixed integer program to model spatial wildfire behavior and suppression placement decisions. Canadian Journal of Forest Research45, 384–393 (2015) https://doi.org/10.1139/cjfr-2014-0252

    work page doi:10.1139/cjfr-2014-0252 2015

  9. [9]

    Fire6(2) (2023) https://doi.org/10.3390/ fire6020037

    Granda, B., Le´ on, J., Vitoriano, B., Hearne, J.: Decision support models and methodologies for fire suppression. Fire6(2) (2023) https://doi.org/10.3390/ fire6020037

    2023

  10. [10]

    Forest Science59(1), 105–117 (2013) https://doi.org/10.5849/ forsci.11-022

    Ntaimo, L., Gallego-Arrubla, J.A., Gan, J., Stripling, C., Young, J., Spencer, T.: A simulation and stochastic integer programming approach to wildfire initial attack planning. Forest Science59(1), 105–117 (2013) https://doi.org/10.5849/ forsci.11-022

    2013

  11. [11]

    Forest Science64(5), 504–514 (2018) https://doi.org/10.1093/forsci/fxy012

    Rodr´ ıguez-Veiga, J., G´ omez-Costa, I., Ginzo-Villamayor, M.J., Casas-M´ endez, B., S´ aiz-D´ ıaz, J.L.: Assignment problems in wildfire suppression: Models for 21 optimization of aerial resource logistics. Forest Science64(5), 504–514 (2018) https://doi.org/10.1093/forsci/fxy012

    work page doi:10.1093/forsci/fxy012 2018

  12. [12]

    International Journal of Wildland Fire22, 871–882 (2013) https://doi.org/10.1071/WF12157

    Houtman, R.M., Montgomery, C.A., Gagnon, A.R., Calkin, D.E., Dietterich, T.G., McGregor, S., Crowley, M.: Allowing a wildfire to burn: Estimating the effect on future fire suppression costs. International Journal of Wildland Fire22, 871–882 (2013) https://doi.org/10.1071/WF12157

    work page doi:10.1071/wf12157 2013

  13. [13]

    Resources7(1) (2018) https://doi.org/10.3390/resources7010004

    Riley, K.L., Thompson, M.P., Scott, J.H., Gilbertson-Day, J.W.: A model-based framework to evaluate alternative wildfire suppression strategies. Resources7(1) (2018) https://doi.org/10.3390/resources7010004

    work page doi:10.3390/resources7010004 2018

  14. [14]

    International Journal of Wildland Fire26, 587–597 (2017) https://doi.org/10.1071/WF16135

    O’Connor, C.D., Calkin, D.E., Thompson, M.P.: An empirical machine learning method for predicting potential fire control locations for pre-fire planning and operational fire management. International Journal of Wildland Fire26, 587–597 (2017) https://doi.org/10.1071/WF16135

    work page doi:10.1071/wf16135 2017

  15. [15]

    International Journal of Wildland Fire34, 24160 (2025) https://doi.org/10.1071/ WF24160

    Cardil, A., Jim´ enez-Ruano, A., Monedero, S., SeLegue, P., Ortega, M., Quilez, R., Fuentes, J., Marshall, G., Clark, R., Chavez, T., de-Miguel, S., Qui˜ nones, T., Bezares, F., Purdy, S., Allison, K., Metzger, T., Waters, C., Mohan, M., Ramirez, J.: Assessing the suppression difficulty of wildland fires for initial attack response. International Journal ...

    2025

  16. [16]

    Forest Science55(3), 249–255 (2009) https://doi.org/10.1093/forestscience/55.3.249

    Finney, M., Grenfell, I.C., McHugh, C.W.: Modeling containment of large wild- fires using generalized linear mixed-model analysis. Forest Science55(3), 249–255 (2009) https://doi.org/10.1093/forestscience/55.3.249

    work page doi:10.1093/forestscience/55.3.249 2009

  17. [17]

    https://arxiv.org/ abs/2606.11676

    Zhenirovskyy, M., Matei, I., Vuppala, R., Kurihana, T., Wonga, H.Y.: Neural- Parameterized Cellular Automata for Wildfire Spread (2026). https://arxiv.org/ abs/2606.11676

    Pith/arXiv arXiv 2026

  18. [18]

    Environmental Modelling & Software 188, 106401

    Xia, Z., Cheng, S.: PyTorchFire: A GPU-accelerated wildfire simulator with dif- ferentiable cellular automata. Environmental Modelling & Software188, 106401 (2025) https://doi.org/10.1016/j.envsoft.2025.106401

    work page doi:10.1016/j.envsoft.2025.106401 2025

  19. [19]

    arXiv preprint arXiv:2512.06102 (2025) arXiv:2512.06102 [cs.LG]

    C ¸ akır, U., Darvariu, V.-A., Lacerda, B., Hawes, N.: JaxWildfire: A GPU- accelerated wildfire simulator for reinforcement learning. arXiv preprint arXiv:2512.06102 (2025) arXiv:2512.06102 [cs.LG]

    arXiv 2025

  20. [20]

    Control Engineering Practice11(7), 733–764 (2003) https://doi.org/10

    Qin, S.J., Badgwell, T.A.: A survey of industrial model predictive control tech- nology. Control Engineering Practice11(7), 733–764 (2003) https://doi.org/10. 1016/S0967-0661(02)00186-7

    2003

  21. [21]

    Annual Reviews in Control57, 100929 (2024) https://doi.org/10.1016/j.arcontrol.2023.100929 22

    K¨ ohler, J., M¨ uller, M.A., Allg¨ ower, F.: Analysis and design of model predic- tive control frameworks for dynamic operation - an overview. Annual Reviews in Control57, 100929 (2024) https://doi.org/10.1016/j.arcontrol.2023.100929 22

    work page doi:10.1016/j.arcontrol.2023.100929 2024

  22. [22]

    MendeleyData, V1

    Xia, Z., Cheng, S.: Data for: ”PyTorchFire: A GPU-Accelerated Wildfire Simula- tor with Differentiable Cellular Automata”. Mendeley Data, V1. Dataset (2024). https://doi.org/10.17632/nx2wsksp9k.1 . https://data.mendeley.com/datasets/ nx2wsksp9k/1

    work page doi:10.17632/nx2wsksp9k.1 2024

  23. [23]

    Technical report, State of California (2021)

    California Department of Forestry and Fire Protection (CAL FIRE): 2020 Fire Siege Report. Technical report, State of California (2021). Detailed analysis of North Complex and Bear Fire operations. https://www.fire.ca.gov/stats-events/

    2020

  24. [24]

    Technical report, National Interagency Fire Center, Boise, ID (2021)

    National Interagency Coordination Center (NICC): 2020 Wildland Fire Summary and Statistics Annual Report. Technical report, National Interagency Fire Center, Boise, ID (2021)

    2020

  25. [25]

    Technical report, Incident Management Team (October 2020)

    North Complex Incident Management Team: North complex incident action plan. Technical report, Incident Management Team (October 2020). See Air Operations Summary, including Chester Air Attack and Mobile Retardant Base at Chester. https://ftp.wildfire.gov/public/incident specific data/calif n/2020 FEDERAL Incidents/gacc support/%212020/CA PNF 001308 North ...

    2020