A multimodal bidirectional spatiotemporal fusion framework for fire spread prediction and mutation identification

Abstract

The process of forest fire spread is often accompanied by unsteady-state fire behavior. Once a fire becomes uncontrollable, the difficulty of fire suppression and rescue operations increases substantially. To address this issue, this study focuses on fire perimeter prediction and mutation identification and proposes a multimodal bidirectional spatiotemporal fusion framework. The model employs a dynamic feature-extraction module to capture historical spatiotemporal fire features, and a static spatial–meteorological joint module to extract features related to terrain, vegetation, and weather. These features are then integrated through a multimodal data fusion and prediction module to achieve fire perimeter prediction. On this basis, a center-of-mass offset method is used to explicitly quantify the angular and distance mutations in fire spread. The proposed approach was validated through simulation experiments and field burning experiments. The simulation results demonstrate that the model achieved an IoU of 82% for fire perimeter prediction, with the identification accuracies of angular mutation and distance mutation reaching 86.2% and 82.4%, respectively. When the predicted angular and distance variations increased significantly, the probabilities of actual mutation occurrence were 85% and 96%, respectively. The model achieved an IoU of 78% for fire perimeter prediction under real fire conditions, with angular mutation and distance mutation identification accuracies of 65.2% and 71.4%, respectively. Furthermore, the parameter sensitivity analysis results indicate that wind direction has a more pronounced effect on angular mutation, wind speed plays a dominant role in distance mutation, and landscape heterogeneity exerts strong influences on both types of mutations. The proposed method enables fire spread prediction and quantitative characterization of mutation behavior, providing references for fire spread risk assessment and firefighting decision-making in complex environments.

Keywords

Wildfire spread / Fire behavior mutation / Multimodal fusion / Spatiotemporal modeling / Perimeter prediction

References

[1] Andrianarivony HS, Akhloufi MA. Machine learning and deep learning for wildfire spread prediction: a review. Fire, 2024, 712132.

[2] Andrianarivony HS, Akhloufi MA. Linu-Mamba: visual Mamba U-Net with linear attention to predict wildfire spread. Remote Sens, 2025, 17152715

[3] Brys C, Martínez DLL, Marinelli M. Machine learning methods for wildfire risk assessment. Earth Sci Inform, 2025.

[4] Burge J, Bonanni MR, Hu RL, Ihme M. Recurrent convolutional deep neural networks for modeling time-resolved wildfire spread behavior. Fire Technol, 2023, 59633273354.

[5] Corbea-Pérez A, Merino A, Navarro A, García-Ortega E. Integrating physical drivers for wildfire hazard modelling in support of disaster risk reduction: a case study in Castile and León. Spain Int J Disaster Risk Reduct, 2026, 135. 106047

[6] Costa-Saura JM, Spano D, Sirca C, Bacciu V. Contrasting patterns and interpretations between a fire spread simulator and a machine learning model when mapping burn probabilities: a case study for Mediterranean areas. Environ Model Softw, 2023, 163. 105685

[7] Cruz MG, Alexander ME. The 10% wind speed rule of thumb for estimating a wildfire's forward rate of spread in forests and shrublands. Ann for Sci, 2019.

[8] Cruz MG, Kilinc M, Gould JS, Anderson WR. Observations of wildfire spread dynamics in southern Australian grasslands. Int J Wildland Fire, 2024.

[9] Dold J, Zinoviev A, Leslie E. Intensity accumulation in unsteady firelines: a simple model for vegetation engagement. Fire Saf J, 2011, 46126369.

[10] Fu ZX, Gong A, Wan JH, Ba WR, Wang HH, Zhang JM. Forest fire risk assessment model optimized by stochastic average gradient descent. Ecol Indic, 2025, 170. 113006

[11] Ghali R, Akhloufi MA. Deep learning approaches for wildland fires using satellite remote sensing data: detection, mapping, and prediction. Fire, 2023.

[12] Haq B, Jamshed MA, Kasi B, Kasi MK. Predicting wildfires with advanced deep learning models on wireless sensor data. IEEE Sens J, 2026, 26227972804.

[13] Hayajneh SM, Alzghoul MI, Naser J. Wind and slope effects on wildland fire spread: a review of experimental, empirical, mathematical, and physics-based models. Fire, 2026.

[14] Holsinger L, Parks SA, Miller C. Weather, fuels, and topography impede wildland fire spread in western US landscapes. For Ecol Manag, 2016, 380: 59-69.

[15] Huot F, Hu RL, Goyal N, Sankar T, Ihme M, Chen YF. Next day wildfire spread: a machine learning dataset to predict wildfire spreading from remote-sensing data. IEEE Trans Geosci Remote Sens, 2022.

[16] Jahdi R, Salis M, Darvishsefat AA, Mostafavi MA, Alcasena F, Etemad V, Lozano O, Spano D. Calibration of FARSITE simulator in northern Iranian forests. Nat Hazards Earth Syst Sci, 2015, 153443459.

[17] Jia YM, Opabola EA. Interpretable machine learning insights into wildfire damage drivers in California, USA. Int J Disaster Risk Reduct, 2025, 126. 105610

[18] Jiang WY, Qiao YM, Su GF, Li X, Meng QX, Wu HY, Quan W, Wang J, Wang F. WFNet: a hierarchical convolutional neural network for wildfire spread prediction. Environ Model Softw, 2023, 170. 105841

[19] Khan SM, Shafi I, Butt WH, Diez ID, Flores MAL, Galan JC, Ashraf I. A systematic review of disaster management systems: approaches, challenges, and future directions. Land (Basel), 2023.

[20] Kim J, Hu YJ, Elhami-Khorasani N, Sun K, Zhou RZ. Assessment of deep learning models integrated with weather and environmental variables for wildfire spread prediction and a case study of the 2023 Maui fires. Nat Hazards, 2025.

[21] Kirkland M, Atkinson PW, Pearce-Higgins JW, de Jong MC, Dowling TPF, Grummo D, Critchley M, Ashton-Butt A. Landscape fires disproportionally affect high conservation value temperate peatlands, meadows, and deciduous forests, but only under low moisture conditions. Sci Total Environ, 2023, 884. 163849

[22] Liu YY, Huang YF, Liggio J, Hayden K, Mihele C, Wentzell J, Wheeler M, Leithead A, Moussa S, Xie C, Yang Y, Zhang Y, Han T, Li SM. A newly developed Lagrangian chemical transport scheme: part 1. Simulation of a boreal forest fire plume. Sci Total Environ, 2023, 880. 163232

[23] McNorton JR, Di Giuseppe F, Pinnington E, Chantry M, Barnard C. A global probability-of-fire (PoF) forecast. Geophys Res Lett, 2024.

[24] Morvan D. Wind effects, unsteady behaviors, and regimes of propagation of surface fires in open field. Combust Sci Technol, 2014, 1867869888.

[25] Or D, Furtak-Cole E, Berli M, Shillito R, Ebrahimian H, Vahdat-Aboueshagh H, McKenna SA. Review of wildfire modeling considering effects on land surfaces. Earth Sci Rev, 2023, 245. 104569

[26] Ritter SM, Hoffman CM, Battaglia MA, Linn R, Mell WE. Vertical and horizontal crown fuel continuity influences group-scale ignition and fuel consumption. Fire (Basel), 2023.

[27] Rodrigues M, de Dios VR, Sil A, Camprubi AC, Fernandes PM. VPD-based models of dead fine fuel moisture provide best estimates in a global dataset. Agric for Meteorol, 2024, 346. 109868

[28] Rossa CG, Davim DA, Sil A, Fernandes PM. Field-based generic empirical flame length-fireline intensity relationships for wildland surface fires. Int J Wildland Fire, 2024.

[29] Shadrin D, Illarionova S, Gubanov F, Evteeva K, Mironenko M, Levchunets I, Belousov R, Burnaev E. Wildfire spreading prediction using multimodal data and deep neural network approach. Sci Rep, 2024.

[30] Shmuel A, Heifetz E. Re-examining the assumption of dominant regional wind and fire spread directions. Int J Wildland Fire, 2022, 315480491.

[31] Wang Q, Ihme M, Linn RR, Chen YF, Yang V, Sha F, Clements C, McDanold JS, Anderson J. A high-resolution large-eddy simulation framework for wildland fire predictions using TensorFlow. Int J Wildland Fire, 2023, 321217111725.

[32] Xu ZS, Li J, Cheng SB, Rui X, Zhao Y, He HJ, Guan HY, Sharma A, Erxleben M, Chang R, Xu LL. Deep learning for wildfire risk prediction: integrating remote sensing and environmental data. ISPRS J Photogramm Remote Sens, 2025, 227: 632-677.

[33] Zahura FT, Bisht G, Li Z, McKnight S, Chen XY. Impact of topography and climate on post-fire vegetation recovery across different burn severity and land cover types through random forest. Ecol Inform, 2024, 82. 102757

Funding

National Natural Science Foundation of China(32471857)

Heilongjiang Provincial Natural Science Foundation(ZL2025F001)

Centeal Finance Foresty scence and Tecinoloev PrmoionDemonstration Proiect(Hei[2024]TG11)

Rights & permissions

Northeast Forestry University