Adapting to climate change: exploring the efficacy of transformer models in forecasting wildfires in Australia

Abstract

Wildfires pose a significant threat to human communities and the environment, necessitating accurate and timely forecasting to inform mitigation and preparedness strategies. Traditional methods for wildfire forecasting struggle with capturing complex temporal dependencies and spatial variations, motivating the need for more advanced approaches. This study addresses these challenges by exploring the capabilities of transformer-based deep learning models for forecasting fire confidence hotspots. Specifically, we investigate five distinct architectures namely TimesNet, Crossformer, PatchTST, iTransformer, and DLinear, chosen for their ability to model temporal dependencies and capture long-term trends in time series data. We assess the performance of these models on historical wildfire and weather data across seven Australian regions, comparing their ability to capture temporal dependencies, spatial variations, and long-term trends. Our approach leverages the key features of transformers such as the attention mechanism and positional encoding to prioritize efficiency and interpretability while maintaining flexibility for capturing diverse aspects of the data. Our results showed that both PatchTST and Crossformer achieved competitive performance across all seven regions of Australia. PatchTST had an overall average mean squared error (MSE) ranging from 0.386 to 0.988 and mean absolute error (MAE) from 0.383 to 0.712, while Crossformer achieved an overall average MSE ranging from 0.350 to 1.013 and MAE from 0.381 to 0.747. PatchTST performed particularly well for shorter forecast horizons, with South Australia (SA), ViCctoria (VI), and TaSsmania (TA) showing the lowest errors. Crossformer also showed strong performance, with Northern Territory (NT) and QueeDnsland (QL) exhibiting the lowest errors. To further understand the models’ predictions, we employed SHapley Additive exPlanations (SHAP) analysis, which revealed that temperature, solar radiation, and relative humidity significantly influence wildfire danger. This interpretability adds valuable insight into the underlying drivers of wildfire risk, providing practical benefits for enhancing wildfire management strategies. The findings offer significant insights into developing effective forecasting models, with implications for enhancing wildfire response strategies.

Keywords

Wildfire / Transformer / Forest fire / Forecasting / Deep learning

References

[1] Abdollahi A, Pradhan B. Explainable artificial intelligence (XAI) for interpreting the contributing factors feed into the wildfire susceptibility prediction model. Sci Total Environ, 2023, 879 163004

[2] Abujayyab SKM, Kassem MM, Khan AA, Wazirali R, Coşkun M, Taşoğlu E, Öztürk A, Toprak F. Wildfire susceptibility mapping using five boosting machine learning algorithms: the case study of the Mediterranean region of Turkey. Adv Civ Eng, 2022, 2022(1 3959150

[3] Anderegg WRL, Wu C, Acil N, Carvalhais N, Pugh TAM, Sadler JP, Seidl R. A climate risk analysis of Earth’s forests in the 21st century. Science, 2022, 377(66101099-1103

[4] Bhadoria RS, Pandey MK, Kundu P. RVFR: random vector forest regression model for integrated & enhanced approach in forest fires predictions. Ecol Inform, 2021, 66 101471

[5] Charizanos G, Demirhan H. Bayesian prediction of wildfire event probability using normalized difference vegetation index data from an Australian forest. Ecol Inform, 2023, 73 101899

[6] Chen ZY, Zhang C, Li WD, Gao LY, Liu LM, Fang L, Zhang CS. Fire danger forecasting using machine learning-based models and meteorological observation: a case study in Northeastern China. Multimed Tools Appl, 2024, 83(22): 61861-61881

[7] Cho K, van Merrienboer B, Gulcehre C, Bahdanau D, Bougares F, Schwenk H, Bengio Y (2014) Learning phrase representations using RNN encoder-decoder for statistical machine translation. arXiv:1406.1078. https://doi.org/10.48550/arXiv.1406.1078

[8] Coffield SR, Graff CA, Chen Y, Smyth P, Foufoula-Georgiou E, Randerson JT. Machine learning to predict final fire size at the time of ignition. Int J Wildland Fire, 2019, 28(11): 861-873

[9] Collini E, Ipsaro Palesi LA, Nesi P, Pantaleo G, Nocentini N, Rosi A. Predicting and understanding landslide events with explainable AI. IEEE Access, 2022, 10: 31175-31189

[10] de Santana RO, Delgado RC, Schiavetti A. Modeling susceptibility to forest fires in the Central Corridor of the Atlantic Forest using the frequency ratio method. J Environ Manag, 2021, 296 113343

[11] Derot J, Sugiura N, Kim S, Kouketsu S. Improved climate time series forecasts by machine learning and statistical models coupled with signature method: a case study with El Niño. Ecol Inform, 2024, 79 102437

[12] Dillis C, Butsic V, Moanga D, Parker-Shames P, Wartenberg A, Grantham TE. The threat of wildfire is unique to Cannabis among agricultural sectors in California. Ecosphere, 2022, 13(9 e4205

[13] dos Santos AC, da Rocha MS, Ferreira MC, Barradas ACS, Schmidt IB. Managing fires in a changing world: fuel and weather determine fire behavior and safety in the neotropical savannas. J Environ Manag, 2021, 289 112508

[14] Fedirka A (2020). The geopolitical consequences of Australia’s wildfires. https://geopoliticalfutures.com. Accessed on 30 July 2024

[15] Filkov AI, Ngo T, Matthews S, Telfer S, Penman TD. Impact of Australia’s catastrophic 2019/20 bushfire season on communities and environment. Retrospective analysis and current trends. J Saf Sci Resil, 2020, 1(1): 44-56

[16] Gao C, Lin HL, Hu HQ. Forest-fire-risk prediction based on random forest and backpropagation neural network of Heihe area in Heilongjiang Province, China. Forests, 2023, 14(2 170

[17] Gaudreau J, Perez L, Drapeau P. BorealFireSim: a GIS-based cellular automata model of wildfires for the boreal forest of Quebec in a climate change paradigm. Ecol Inform, 2016, 32: 12-27

[18] Hochreiter S, Schmidhuber J. Long short-term memory. Neural Comput, 1997, 9(8): 1735-1780

[19] Hong HY, Tsangaratos P, Ilia I, Liu JZ, Zhu AX, Xu C. Applying genetic algorithms to set the optimal combination of forest fire related variables and model forest fire susceptibility based on data mining models. The case of Dayu County, China. Sci Total Environ, 2018, 630: 1044-1056

[20] Iban MC, Sekertekin A. Machine learning based wildfire susceptibility mapping using remotely sensed fire data and GIS: a case study of Adana and Mersin provinces, Turkey. Ecol Inform, 2022, 69 101647

[21] Jaafari A, Gholami DM, Zenner EK. A Bayesian modeling of wildfire probability in the Zagros Mountains, Iran. Ecol Inform, 2017, 39: 32-44

[22] Jiang H, Hu H, Zhong RH, Xu JF, Xu JL, Huang JF, Wang SW, Ying YB, Lin T. A deep learning approach to conflating heterogeneous geospatial data for corn yield estimation: a case study of the US Corn Belt at the county level. Glob Chang Biol, 2020, 26(3): 1754-1766

[23] Jones MW, Abatzoglou JT, Veraverbeke S, Andela N, Lasslop G, Forkel M, Smith AJP, Burton C, Betts RA, van der Werf GR, Sitch S, Canadell JG, Santín C, Kolden C, Doerr SH, Le Quéré C. Global and regional trends and drivers of fire under climate change. Rev Geophys, 2022, 60(3 e2020RG000726

[24] Kannangara KKPM, Zhou WH, Ding Z, Hong ZH. Investigation of feature contribution to shield tunneling-induced settlement using Shapley additive explanations method. J Rock Mech Geotech Eng, 2022, 14(4): 1052-1063

[25] Lall S, Mathibela B (2016) The application of artificial neural networks for wildfire risk prediction. In: 2016 International Conference on Robotics and Automation for Humanitarian Applications (RAHA), December 2016, Amritapuri, India. IEEE: pp 1–6. https://doi.org/10.1109/RAHA.2016.7931880

[26] Li YD, Feng ZK, Chen SL, Zhao ZY. Wang FG (2020) Application of the artificial neural network and support vector machines in forest fire prediction in the Guangxi autonomous region, China. Discrete Dyn Nat Soc, 2020, 1: 5612650

[27] Li T, Cui LZ, Liu LL, Chen Y, Liu HD, Song XF, Xu ZH. Advances in the study of global forest wildfires. J Soils Sediments, 2023, 23(7): 2654-2668

[28] Liu Y, Hu T, Zhang H, Wu H, Wang S, Ma L, Long M (2023). Itransformer: inverted transformers are effective for time series forecasting. arXiv:2310.06625. https://doi.org/10.48550/arXiv.2310.06625

[29] Lundberg SM, Lee SI (2017). A unified approach to interpreting model predictions. In: Advances in Neural Information Processing Systems, December 2017. Curran Associates, Inc. https://proceedings.neurips.cc/paper_files/paper/2017/hash/8a20a8621978632d76c43dfd28b67767-Abstract.html. Accessed on 30 July 2024

[30] McColl-Gausden SC, Bennett LT, Clarke HG, Ababei DA, Penman TD. The fuel–climate–fire conundrum: how will fire regimes change in temperate eucalypt forests under climate change?. Glob Change Biol, 2022, 28(17): 5211-5226

[31] Miao XY, Li J, Mu YJ, He C, Ma YF, Chen J, Wei WT, Gao DM. Time series forest fire prediction based on improved transformer. Forests, 2023, 14(8): 1596

[32] Muhammad K, Ahmad J, Baik SW. Early fire detection using convolutional neural networks during surveillance for effective disaster management. Neurocomputing, 2018, 288: 30-42

[33] Nie YQ, Nguyen NH, Sinthong P, Kalagnanam J (2022) A time series is worth 64 words: long-term forecasting with transformers. arXiv:2211.14730. https://doi.org/10.48550/arXiv.2211.14730

[34] Pang YQ, Li YD, Feng ZK, Feng ZM, Zhao ZY, Chen SL, Zhang HY. Forest fire occurrence prediction in China based on machine learning methods. Remote Sens, 2022, 14(21): 5546

[35] Pérez-Sánchez J, Jimeno-Sáez P, Senent-Aparicio J, Díaz-Palmero JM, de Dios Cabezas-Cerezo J. Evolution of burned area in forest fires under climate change conditions in southern Spain using ANN. Appl Sci, 2019, 9(19 4155

[36] Pham VT, Do TAT, Tran HD, Do ANT. Classifying forest cover and mapping forest fire susceptibility in Dak Nong province, Vietnam utilizing remote sensing and machine learning. Ecol Inform, 2024, 79 102392

[37] Qiao YM, Jiang WY, Wang F, Su GF, Li X, Jiang JC. FireFormer: an efficient Transformer to identify forest fire from surveillance cameras. Int J Wildland Fire, 2023, 32(9): 1364-1380

[38] Ribeiro M, Singh S, Guestrin C (2016) “Why should I trust you?”: explaining the predictions of any classifier. In: Proceedings of the 2016 Conference of the North American Chapter of the Association for Computational Linguistics: demonstrations, June 2016, San Diego, California. Association for Computational Linguistics: pp 97–101. https://doi.org/10.18653/v1/n16-3020

[39] Rodrigues M, de la Riva J, Fotheringham S. Modeling the spatial variation of the explanatory factors of human-caused wildfires in Spain using geographically weighted logistic regression. Appl Geogr, 2014, 48: 52-63

[40] Sivrikaya F, Küçük Ö. Modeling forest fire risk based on GIS-based analytical hierarchy process and statistical analysis in Mediterranean region. Ecol Inform, 2022, 68 101537

[41] Sivrikaya F, Günlü A, Küçük Ö, Ürker O. Forest fire risk mapping with Landsat 8 OLI images: evaluation of the potential use of vegetation indices. Ecol Inform, 2024, 79 102461

[42] Succarie A, Xu ZH, Wang WJ. The variation and trends of nitrogen cycling and nitrogen isotope composition in tree rings: the potential for fingerprinting climate extremes and bushfires. J Soils Sediments, 2022, 22(9): 2343-2353

[43] Surbhi Singh S, Jeganathan C. Using ensemble machine learning algorithm to predict forest fire occurrence probability in Madhya Pradesh and Chhattisgarh, India. Adv Space Res, 2024, 73(6): 2969-2987

[44] Tien Bui D, Van Le H, Hoang ND. GIS-based spatial prediction of tropical forest fire danger using a new hybrid machine learning method. Ecol Inform, 2018, 48: 104-116

[45] Tuyen TT, Jaafari A, Yen HPH, Nguyen-Thoi T, Phong TV, Nguyen HD, Van Le H, Phuong TTM, Nguyen SH, Prakash I, Pham BT. Mapping forest fire susceptibility using spatially explicit ensemble models based on the locally weighted learning algorithm. Ecol Inform, 2021, 63 101292

[46] United Nations Environment Programme (2022). Spreading like wildfire: The rising threat of extraordinary landscape fires. https://eprints.qut.edu.au/240827/. Accessed on 30 July 2024

[47] Van LN, Tran VN, Nguyen GV, Yeon M, Do MT, Lee G. Enhancing wildfire mapping accuracy using mono-temporal Sentinel-2 data: a novel approach through qualitative and quantitative feature selection with explainable AI. Ecol Inform, 2024, 81 102601

[48] Vaswani A, Shazeer N, Parmar N, Uszkoreit J, Jones L, Gomez AN, Kaiser Ł, Polosukhin I (2017). Attention is all you need. In: Advances in Neural Information Processing Systems, December 2017. Curran Associates, Inc. https://proceedings.neurips.cc/paper/2017/hash/3f5ee243547dee91fbd053c1c4a845aa-Abstract.html. Accessed on 30 July 2024

[49] Wang WQ, Zhao FJ, Wang YX, Huang XY, Ye JX. Seasonal differences in the spatial patterns of wildfire drivers and susceptibility in the southwest mountains of China. Sci Total Environ, 2023, 869 161782

[50] Wu HX, Hu TG, Liu Y, Zhou H, Wang JM, Long MS (2022) TimesNet: temporal 2D-variation modeling for general time series analysis. arXiv:2210.02186. https://doi.org/10.48550/arXiv.2210.02186

[51] Zeng AL, Chen MX, Zhang L, Xu Q. Are transformers effective for time series forecasting?. Proc AAAI Conf Artif Intell, 2023, 37(9): 11121-11128

[52] Zerveas G, Jayaraman S, Patel D, Bhamidipaty A, Eickhoff C (2021) A transformer-based framework for multivariate time series representation learning. In: Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining. virtual event Singapore. ACM, 2114–2124. https://doi.org/10.1145/3447548.3467401

[53] Zhang Y, Yan J (2023) Crossformer: transformer utilizing cross-dimension dependency for multivariate time series forecasting. In: 11th international conference on learning representations. https://openreview.net/forum?id=vSVLM2j9eie. Accessed on 30 July 2024

Rights & permissions

Northeast Forestry University