野火是世界範圍內普遍存在的現象，對環境、生物的多樣性產生重大影響，並威脅人類的健康和財產安全。對於野火管理員、消防員、研究人員來說，了解野火的蔓延行為非常重要，因為這可以幫助他們獲取野火蔓延的實時關鍵信息，幫助 他們進行野火風險評估，並決定如何配資源用於野火控制和撲滅。野火蔓延是一種複雜的物理現象，很多野火蔓延模擬方法都是基於物理學的方法，這種方法不方便使用者理解野火的蔓延行為。

在這項工作中，我們提出了一個野火蔓延預測的可視化分析系統，該系統基於經過深度訓練的卷積逆向圖形網絡模型(DCIGN)，此模型可以用於野火蔓延預測。基於神經網絡模型的可解釋性和可解釋性，此模型還可以幫助我們模擬和揭示野火蔓延的複雜行為。

我們的可視化分析系統提供了一些視圖來協助野火管理人員、消防員、研究人 員分析野火蔓延的特徵。實例視圖可以通過選擇輸入地形和天氣參數來可視化野火 蔓延預測的區域。該系統還利用深度Shapely additive explanations(SHAP)解釋工具提供多種輸入參數敏感性分析視圖。這些敏感性分析視圖包括所有野火預測區域不同空間區域的全局平均敏感性和針對用戶選擇的特定區域進行的局部參數敏感性分析。在局部參數敏感性分析基礎上，有時需要了解多個天氣參數如何共同影響野火蔓延的路徑我們的系統包含的參數相關性分析可以幫助用戶了解所選的天氣參數如何影響野火的蔓延。

Dropout層是解決模型訓練時過擬合的有效手段，同時會導致模型預測結果的不確定性，系統提供的不確定性視圖可以對野火預測的不確定性進行分析。此外，我們的可視化系統提供了參數優化視圖，當從地圖上選擇一個區域時，該視圖可以幫助用戶分析出可能導致野火蔓延到該區域的天氣參數。該視圖可用於分析可能導致野火蔓延到地圖上選定區域的天氣參數，這在實際野火分析中具有重要意義。

Wildfires are a widespread phenomenon across the world, with significant impacts on the environment, biodiversity and threats to human health and property safety. Understanding the spreading behavior of wildfires is important for wildfire managers, firefighters, and researchers. This understanding can help them obtain real-time crit- ical information, conduct wildfire risk assessments, and make decisions about where to allocate resources for wildfire control and suppression. However, the spread of wildfires is a complex physical phenomenon, and there are many physics-based wild- fire spread simulation models. These models can be computationally expensive and difficult to use, which makes it inconvenient to understand the spread behavior of wildfires.

In this work, we propose a visual analysis system for wildfire spread predic- tion. This system is based on a trained deep convolutional inverse graphics network (DCIGN) model for wildfire spread prediction. The interpretability and explainabil- ity of neural network-based models allow us to reveal the complex behavior of wildfire spread simulation.

The visual analysis system provides a number of views to assist wildfire managers, firefighters, and researchers in analyzing the characteristics of wildfire spread. The instance view visualizes the wildfire spread prediction simulation output by choosing the input terrain and weather parameters. The system also provides a variety of input parameter sensitivity analysis views by utilizing the deep Shapely additive explanations(SHAP) tool. These sensitivity analysis views include the global average sensitivity of different spatial regions across all the wildfire prediction areas and local parameter sensitivity analysis for a specific area selected by the users. Furthermore, it is imperative to comprehend how multiple weather parameters jointly influence the trajectory of wildfire spread. The system also includes a parameter correlation analysis to help users understand how the selected weather parameters affect wildfire spread.

The dropout layer is an effective means to solve overfitting during model training, but it can also lead to uncertainty in the model prediction results. We address this by providing an uncertainty view that allows users to analyze the uncertainty of wildfire prediction. Our visualization system also provides a parameter optimization view. This view can be used to analyze the weather parameters that could cause wildfires to spread to a selected area on the map. This is of great significance in actual wildfire analysis.

Elder C. Hill K. Lumley T. Davies A. Brown, A. and McNamara T. Cambridge Advanced Learner’s Dictionary. Cambridge University, 2008.
WildfireHealthCenter. Wildfires damage. https://www.who.int/ health-topics/wildfires.
Md Kamrul Haque, Md Abul Kalam Azad, Md Yeamin Hossain, Tareq Ahmed, Minhaz Uddin, and Md Mukhtar Hossain. Wildfire in australia during 2019-2020, its impact on health, biodiversity and environment with some proposals for risk management: a review. Journal of Environmental Protection, 12(6):391–414, 2021.
A Park Williams, John T Abatzoglou, Alexander Gershunov, Janin Guzman- Morales, Daniel A Bishop, Jennifer K Balch, and Dennis P Lettenmaier. Ob- served impacts of anthropogenic climate change on wildfire in california. Earth’s Future, 7(8):892–910, 2019.
National Interagency Fire Center. Total wildland fires and acres (1983-2022). https://www.nifc.gov/fire-information/statistics/wildfires.
National Wildfire Coordinating Group. Nwcg report on wildland firefighter fa- talities in the united states: 2007–2016., 2017.
Claire Miller, James Hilton, Andrew Sullivan, and Mahesh Prakash. Spark–a bushfire spread prediction tool. In Environmental Software Systems. Infrastruc- tures, Services and Applications: 11th IFIP WG 5.11 International Symposium, ISESS 2015, Melbourne, VIC, Australia, March 25-27, 2015. Proceedings 11, pages 262–271. Springer, 2015.
JJ Sharples, R Richards, J Hilton, S Ferguson, R Cohen, and M Thatcher. Dynamic simulation of the cape barren island fire using the spark framework. In MODSIM2017, 22nd International Congress on Modelling and Simulation, eds G. Syme, D. Hatton MacDonald, B. Fulton, and J. Piantadosi (Hobart, TAS: Modelling and Simulation Society of Australia and New Zealand), pages 1111– 1117, 2017.
Ujjwal Kc, Jagannath Aryal, James Hilton, and Saurabh Garg. A surrogate model for rapidly assessing the size of a wildfire over time. Fire, 4(2):20, 2021.
Scott M Lundberg and Su-In Lee. A unified approach to interpreting model predictions. Advances in neural information processing systems, 30, 2017.
Miguel G Cruz, N Phillip Cheney, James S Gould, W Lachlan McCaw, Musa Kilinc, and Andrew L Sullivan. An empirical-based model for predicting the forward spread rate of wildfires in eucalypt forests. International journal of wildland fire, 31(1):81–95, 2021.
Carlos G Rossa and Paulo M Fernandes. An empirical model for the effect of wind on fire spread rate. Fire, 1(2):31, 2018.
Carlos G Rossa and Paulo M Fernandes. Empirical modeling of fire spread rate in no-wind and no-slope conditions. Forest Science, 64(4):358–370, 2018.
AL Sullivan. A review of wildland fire spread modelling, 1990-present 2: Empir- ical and quasi-empirical models. arXiv preprint arXiv:0706.4128, 2007.
Jan Mandel, Jonathan D Beezley, and Adam K Kochanski. Coupled atmosphere- wildland fire modeling with wrf-fire version 3.3. Geoscientific Model Development Discussions, 4(1):497–545, 2011.
J Mandel, S Amram, JD Beezley, G Kelman, AK Kochanski, VY Kondratenko, BH Lynn, B Regev, and M Vejmelka. Recent advances and applications of wrf– sfire. Natural Hazards and Earth System Sciences, 14(10):2829–2845, 2014.
Mark A Finney. FARSITE, Fire Area Simulator–model development and evalua- tion. Number 4. US Department of Agriculture, Forest Service, Rocky Mountain Research Station, 1998.
William Mell, Mary Ann Jenkins, Jim Gould, and Phil Cheney. A physics-based approach to modelling grassland fires. International Journal of Wildland Fire, 16(1):1–22, 2007.
Michael A Storey, Michael Bedward, Owen F Price, Ross A Bradstock, and Jason J Sharples. Derivation of a bayesian fire spread model using large-scale wildfire observations. Environmental Modelling & Software, 144:105127, 2021.
Bernoussi A. Mˆaatouk M. Amharref M. Jellouli, O. Forest fire modelling using cellular automata: application to the watershed oued laou. Mathematical and computer modelling of dynamical systems, 22:493–507, 2016.
Fadoua Khennou, Jade Ghaoui, and Moulay A Akhloufi. Forest fire spread prediction using deep learning. In Geospatial Informatics XI, volume 11733, pages 106–117. SPIE, 2021.
Jonathan L Hodges and Brian Y Lattimer. Wildland fire spread modeling using convolutional neural networks. Fire technology, 55:2115–2142, 2019.
VI Kozik, ES Nezhevenko, and AS Feoktistov. Studying the method of adap- tive prediction of forest fire evolution on the basis of recurrent neural networks. Optoelectronics, Instrumentation and Data Processing, 50:395–401, 2014.
John Burge, Matthew Bonanni, Matthias Ihme, and Lily Hu. Convolu- tional lstm neural networks for modeling wildland fire dynamics. arXiv e- prints,arXiv:2012.06679, 2020.
Sigfrido Waidelich, Karina Laneri, and M´onica M Denham. Fire propagation visualization in real time. Journal of Computer Science & Technology, 18:238– 247, 2018.
Monica M Denham, Sigfrido Waidelich, and Karina Laneri. Visualization and modeling of forest fire propagation in patagonia. Environmental Modelling & Software, 158:105526, 2022.
Micael Makenzius and Jonas Bylerius. A web application for wildfire spread prediction and visualisation in sweden using geospatial data and technology. Master’s thesis, KTH Royal Institute of Technology, 2022.
Rui Wu, Chao Chen, Sajjad Ahmad, John M Volk, Cristina Luca, Frederick C Harris, and Sergiu M Dascalu. A real-time web-based wildfire simulation sys- tem. In IECON 2016-42nd Annual Conference of the IEEE Industrial Electronics Society, pages 4964–4969. IEEE, 2016.
Sadegh Khanmohammadi, Mehrdad Arashpour, Emadaldin Mohammadi Golaf- shani, Miguel G Cruz, Abbas Rajabifard, and Yu Bai. Prediction of wildfire rate of spread in grasslands using machine learning methods. Environmental Modelling & Software, 156:105507, 2022.
Frederic Allaire, Vivien Mallet, and Jean-Baptiste Filippi. Emulation of wildland fire spread simulation using deep learning. Neural networks, 141:184–198, 2021.
JE Hilton, C Miller, A Sullivan, and C Rucinski. Incorporation of variation into wildfire spread models using a level set approach. Environmental Modelling amp; Software, 67:118–127, 2015.
Diederik P Kingma and Max Welling. Auto-encoding variational bayes. arXiv preprint arXiv:1312.6114, 2013.
Tejas D Kulkarni, William F Whitney, Pushmeet Kohli, and Josh Tenenbaum. Deep convolutional inverse graphics network. Advances in neural information processing systems, 28, 2015.
Anders Boesen Lindbo Larsen, Søren Kaae Sønderby, Hugo Larochelle, and Ole Winther. Autoencoding beyond pixels using a learned similarity metric. In International conference on machine learning, pages 1558–1566. PMLR, 2016.
Irina Higgins, Loic Matthey, Arka Pal, Christopher Burgess, Xavier Glorot, Matthew Botvinick, Shakir Mohamed, and Alexander Lerchner. beta-vae: Learn- ing basic visual concepts with a constrained variational framework. In Interna- tional conference on learning representations, 2017.
Jonathan Lee Hodges. Predicting large domain multi-physics fire behavior using artificial neural networks. PhD thesis, Virginia Tech, 2018.
Avanti Shrikumar, Peyton Greenside, and Anshul Kundaje. Learning important features through propagating activation differences. In International conference on machine learning, pages 3145–3153. PMLR, 2017.
Dumitru Erhan, Yoshua Bengio, Aaron Courville, and Pascal Vincent. Visual- izing higher-layer features of a deep network. University of Montreal, 1341(3):1, 2009.
ESRI. Geography coordinate. https://support.esri.com/en/ technical-article/000011853, 2020.
James Hilton, William Swedosh, Lachlan Hetherton, Andrew Sullivan, and Ma- hesh Prakash. Spark user guide. 2017.
Fausto Milletari, Nassir Navab, and Seyed-Ahmad Ahmadi. V-net: Fully con- volutional neural networks for volumetric medical image segmentation. In 2016 fourth international conference on 3D vision (3DV), pages 565–571. IEEE, 2016.
Shruti Jadon. A survey of loss functions for semantic segmentation. In 2020 IEEE conference on computational intelligence in bioinformatics and computational biology (CIBCB), pages 1–7. IEEE, 2020.
Seyed Sadegh Mohseni Salehi, Deniz Erdogmus, and Ali Gholipour. Tversky loss function for image segmentation using 3d fully convolutional deep networks. arXiv e-prints, pages arXiv–1706, 2017.
Nabila Abraham and Naimul Mefraz Khan. A novel focal tversky loss function with improved attention u-net for lesion segmentation. In 2019 IEEE 16th inter- national symposium on biomedical imaging (ISBI 2019), pages 683–687. IEEE, 2019.
Vasyl Pihur, Susmita Datta, and Somnath Datta. Weighted rank aggregation of cluster validation measures: a monte carlo cross-entropy approach. Bioinfor- matics, 23(13):1607–1615, 2007.
Yarin Gal and Zoubin Ghahramani. Dropout as a bayesian approximation: Rep- resenting model uncertainty in deep learning. In international conference on machine learning, pages 1050–1059. PMLR, 2016.