X-Fire 2025 – 1st International Workshop on Extreme Wildfire Events: Book of Abstracts

COST Action CA22164 · X-Fire 2025 1st International Workshop on Extreme Wildfire Events X-Fire 2025 1st International Workshop on Extreme Wildfire Events Nicosia, Cyprus · 23–25 September 2025 Book of Abstracts This publication is based upon work from COST Action NERO, CA22164, supported by COST (European Cooperation in Science and Technology). COST (European Cooperation in Science and Technology) is a funding agency for research and innovation networks. Our Actions help connect research initiatives across Europe and enable scientists to grow their ideas by sharing them with their peers. This boosts their research, career, and innovation. www.cost.eu © COST Action CA22164 (NERO), https://nero-network.eu TABLE OF CONTENTS PREFACE................................................................................................................................ 4 WORKSHOP OVERVIEW .......................................................................................................... 5 BACKGROUND AND OBJECTIVES ......................................................................................... 5 STRUCTURE AND FORMAT ................................................................................................... 6 PARTICIPATION ................................................................................................................... 6 ACKNOWLEDGEMENTS ....................................................................................................... 6 WILDFIRE MODELING BY AN ADVECTION-DISPERSION-REACTION (ADR) APPROACH ENRICHED WITH EMPIRICAL KNOWLEDGE ................................................................................................ 9 FORESTSPHERE: A DIGITAL TWIN APPROACH FOR MONITORING AND MANAGING FORESTS AND WILDFIRE RISK ..................................................................................................................... 14 IFIRE AI: AI-POWERED WILDFIRE SIMULATION AND 3D IMMERSIVE VISUALISATION ................... 20 MODELING THE LARGEST WILDFIRE USING FARSITE IN THE CZECH REPUBLIC: NATIONAL PARK BOHEMIAN SWITZERLAND 2022............................................................................................. 32 MAPPING WILDFIRE PROGRESSIONS AND MEASURING RATE OF SPREAD FROM “LINE SCANS” IN AUSTRALIA ........................................................................................................................... 37 PYROCONVECTION ANALYSIS AND DATA GATHERING FROM EXTREME WILDFIRE EVENTS......... 43 ESTIMATING THE PROBABILITY OF PYROCUMULUS FORMATION IN EXTREME WILDFIRE SCENARIOS: A PRELIMINARY APPROACH ............................................................................... 49 EVOLUTION OF EXTREME WUI WILDFIRE INCIDENTS FORENSIC CHRONO-SPATIAL RECONSTRUCTION METHODOLOGY – VARNAVAS/GREECE 2024 APPLICATION ........................ 55 INTEGRATED MULTI-SOURCE DATASET FOR SPATIOTEMPORAL ANALYSIS OF THE 2024 SKRADIN WILDFIRE, CROATIA .............................................................................................................. 61 CALIBRATION OF A THERMAL TUNNEL TO ANALYSE THE EFFECT OF ATMOSPHERIC STRUCTURE ON FIRE SPREAD .................................................................................................................. 66 HYDROCLIMATIC REBOUND DRIVES EXTREME FIRE IN CALIFORNIA'S NON-FORESTED ECOSYSTEMS ....................................................................................................................... 74 CROSS-COUNTRY CORRELATIONS IN FIRE WEATHER ENHANCE THE DANGER OF EXTREMELY WIDESPREAD FIRES IN EUROPE ............................................................................................. 75 EUROPEAN FIRE SYNCHRONICITY MORE LIKELY DUE TO OVERLAPPING FIRE SEASONALITY AND ATMOSPHERIC BLOCKING .................................................................................................... 79 INCREASING LIKELIHOOD OF FIRE-CONDUCTIVE CONDITIONS IN A WARMING EUROPE ........... 83 ERUPTIVE FIRE EARLY WARNING SYSTEM ............................................................................... 87 Preface The X-Fire 2025 – 1st International Workshop on Extreme Wildfire Events marks an important milestone for the NERO network and for the broader communities of researchers and practitioners working on wildland fire behavior. Extreme wildfire events are not just bigger fires and they are no longer rare anomalies. They are events that challenge our understanding, overwhelm operational capacity, and test the limits of preparedness and response. Addressing them requires not only scientific progress but also genuine collaboration between researchers, practitioners, and decision-makers. With this spirit, the COST Action NERO (CA22164 – european Network on Extreme fiRe behaviOr) launched X-Fire as a series of workshops designed to bridge science and practice, share knowledge across disciplines, and strengthen our collective capacity to anticipate and respond to extreme wildfires. The first edition of X-Fire, held in Nicosia (Cyprus) from 23 to 25 September 2025, brought together experts and practitioners from across Europe and beyond to discuss advances in data and modeling, observations, forecasting, and operational challenges. It also served as platform for honest dialogue on how we can transform scientific knowledge into operational readiness and response. This Book of Abstracts captures the diversity of perspectives and research presented during the workshop. It reflects the shared commitment of our community to move from understanding to preparedness, and to ensure that knowledge continues to serve those who face the fires on the ground. I would like to express my sincere appreciation to the European University of Cyprus, and personally to Prof. Dr. Georgios Boustras, for hosting X-Fire 2025; to the COST (European Cooperation in Science and Technology) Association for financially enabling this event; and, of course, to all contributors that made it a reality. Let this first X-Fire workshop be the foundation for continued dialogue, stronger collaborations, and lasting progress toward safer and more resilient landscapes in the era of extreme wildfires! Dr. Theodore M. Giannaros Chair, COST Action NERO (CA22164) National Observatory of Athens 15.10.2025 Workshop Overview Background and Objectives The X-Fire 2025 – 1st International Workshop on Extreme Wildfire Events was organized under the framework of the COST Action NERO (CA22164 – european Network on Extreme fiRe behaviOr). It aimed to serve as a focused international forum for advancing knowledge on the unique challenges of extreme wildfire events through the exchange of scientific and operational insights. The workshop’s objectives were to: ▪ Showcase emerging tools, data, and modeling approaches for forecasting and analysis. ▪ Identify gaps and priorities for improving preparedness, response, and resilience. ▪ Promote dialogue between research and practice. ▪ Strengthen international collaboration across scientific and operational communities. Structure and Format X-Fire 2025 took place from 23 to 25 September 2025 at the European University of Cyprus (EUC) in Nicosia, Cyprus. The workshop programme included: ▪ Three keynote presentations, addressing forecasting, fundamentals of wildland fire behavior, and observational challenges. ▪ Four thematic sessions, covering topics from modeling and observing extreme wildfires, to studies of their drivers and management challenges. ▪ A panel discussion entitled “From Research to Practice”, exploring pathways to translate scientific advances into operational and policy contexts. ▪ Networking and social activities, including a workshop dinner and informal exchanges among participants. Participation The workshop brought together 39 participants from 15 countries, representing a diverse mix of institutions, including Universities, Research Centers, and Fire Management Agencies. This diversity underscores NERO’s mission to bridge science and operations in addressing the growing challenges of extreme wildfire events. Acknowledgements NERO acknowledges the European University of Cyprus (EUC) for hosting the event and the COST Association for its financial support. Special thanks are extended to the Local Organizing Team (Georgios Boustras, Cleo Varianou Mikellidou, Pierantonios Papazoglou, and Klelia Vasiliou) and the Core Group of NERO for their contributions to the workshop’s scientific and logistical success. X-Fire 2025 Abstracts X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Session 1 Numerical Modeling of Extreme Wildfire Events X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Wildfire Modeling by an Advection-Dispersion-Reaction (ADR) Approach Enriched with Empirical Knowledge Konstantinos Vogiatzoglou1,*, Vasilis Bontozoglou1, Costas Papadimitriou1, Konstantinos Ampountolas1, Sergey Litvinov2, Lucas Amoudruz2, Petros Koumoutsakos2 1 University of Thessaly, Department of Mechanical Engineering, Volos, Greece, [email protected] 2 Harvard University, School of Engineering and Applied Sciences, MA, USA, [email protected]vard.edu * Corresponding (presenting) author Summary Wildfires are severe natural hazards with profound impacts on society and the environment. Accurate forecasting requires adaptive, multi-fidelity simulations that integrate weather, fuel, and terrain variability. This work presents a data-and physics-based wildfire spread model that incorporates key thermal and chemical mechanisms alongside empirical knowledge and CFD-inspired insights. The model employs a thermal energy equation with dispersion, advection, reaction, and environmental losses. Simulated cases align with benchmarks, capturing fuel heterogeneity, slope-driven spread, and wind-driven dynamics. The framework is designed as the core of a real-time, data-informed decision support system for wildfire prediction and management. Keywords: Physics-Based Modeling, Fuel Heterogeneity, Slope Effects, Byram Number, Rate of Spread 1. Introduction The frequency of wildfire-disasters has escalated in recent years, a trend expected to persist due to climate change and human interventions. Wildfires are fast-moving fires that spread across vegetation-rich areas, with significant social, environmental, and economic consequences [1]. Between March 2023 and February 2024, wildfires consumed approximately 8,400 km2 across Europe, resulting in the loss of at least 44 lives directly attributed to these events. Wildland fires are among the most complex environmental phenomena, combining the physics of fluid flow with the chemistry of combustion across a wide range of spatial and temporal scales [2]. Broadly, three categories of models aim to predict key aspects of fire evolution. At the simplest level, empirical models, such as Rothermel’s model [3] use functional correlations between the rate of spread (ROS) and variables like wind speed and humidity, leveraging observations and experiments to forecast fire behavior. At the other X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Vegetation Forest Sphere Consortium UAV Lidar/Multispectral 5 to 10 cm On demand Labeled Point Cloud Ground-level vegetation Canopy base height UGV and ground level sensors Lidar/Multispectral 1 to 5 cm On demand Labeled Point Cloud Fire Burned Areas ICNF*5 Satellite; Ground validation 1 ha Yearly Vector EFFIS*6 Satellite 30 ha Daily Vector Ignitions ANEPC*7 Airborne, Multispectral N/A On demand Text (coordinates) Firefighting means on site Ground sensors Discrete (sensor location) On demand Text (coordinates) *1 Instituto Português do Mar e da Atmosfera (https://api.ipma.pt/) *2 Copernicus Atmosphere Monitoring Service (CAMS) Portal (https://atmosphere.copernicus.eu) *3 Copernicus Climate Change Service (C3S) Portal (https://climate.copernicus.eu) *4 Direção-Geral do Território Portal (https://www.dgterritorio.gov.pt/cartografia/cartografia-topografica/modelos-digitais?language=en) *5 Instituto da Conservação da Natureza e das Florestas (https://geocatalogo.icnf.pt/catalogo.html) *6 European Forest Fire Information System - EFFIS (https://forest-fire.emergency.copernicus.eu/apps/data.request.form) *7 Autoridade Nacional de Emergência e Proteção Civil Portal (https://prociv-portal.geomai.mai.gov.pt/) The data, processed and shown through interactive user interfaces, designed in close collaboration with main stakeholders and civil protection end-users, which will be used to obtain valuable insights on the forest structure, health, and the wildfire risk exposure of the forest and the valuable assets located in it, such as the electrical, gas, telecommunication and transport infrastructures. Such data can also be generated or leveraged by intelligent means for forest management, including advanced fire behaviour prediction models and autonomous robotic mulchers or firefighting robots [11]. Figure 1 depicts the proposed ForestSphere architecture. 3. Results and Discussion Challenges such as the computational costs associated with high-dimensional dynamical systems require the development of ad-hoc methods and the employment of HPC resources to support the proposed development. Temporal and spatial synchronization of data, updates and feedback is another challenge for the integration and fusion of heterogeneous data obtained from different sensor modalities and observation sources. Strategies to overcome this include data Resampling and interpolation, use of convolutional neural networks (CNNs) capable of handling multi-resolution inputs, implementing fusion frameworks (e.g., spatial-temporal fusion models) designed for heterogeneous data sources, downscale coarse data using statistical models or machine learning to generate finer-resolution estimates where needed and validation of fused datasets with ground truth measurements. A compromise between accuracy and operability should be found, X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) leveraging soft-synchronization and decision-level fusion approaches. Furthermore, realworld testing shall be required to ensure the platform interoperability, AI-driven models' accuracy and adaptability and final system validation by main stakeholders. Figure 1. ForestSphere Digital Twin architecture. 4. Conclusions Research to develop technological solutions that contribute to forest resilience and protection of human lives and assets is required and has the potential to be quickly assimilated, as we witness an increase in wildfire frequency and extent, exacerbated by the effects of climate change and rural land abandonment. ForestSphere shall contribute both to the resilience of the territory and the forestry industry, with clear socio-economic benefits. Real-time data on vegetation structure and location can be leveraged by the authorities and infrastructure owners to assess wildfire risk exposure. Forestry enterprises can leverage this information for thinning and harvesting activities. Companies responsible for large critical infrastructures such as power distribution, can monitor the growth of trees in the vicinity of powerlines and predict the need for pruning or tree removal actions in advance, facilitating the planning of operations. Civil Protection Authorities will have a digital platform aggregating current and historical multimodal data concerning forests and X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) wildfires. Historical data on wildfire occurrences can support fire preparedness and prevention actions, such as fuel break locations, firefighting means allocations, fuel management strategies and more. Advanced fire spread simulators applied to the interactive DT environment can be leveraged for the training of civil protection teams. During fire occurrences, data can be used to perceive the wildfire behavior and estimate its potential spread, with the help of dedicated models and AI algorithms. Additionally, the platform will facilitate collaboration between firefighting teams, civil protection authorities, and infrastructure managers, ensuring coordinated responses to wildfire emergencies. Acknowledgements This research was funded by the Fundação para a Ciência e a Tecnologia, I.P. (FCT, https://ror.org/00snfqn58) through the project base funding (reference: UIDB/50022/2020, DOI: 54499/UIDB/50022/2020) and under the Protocol between FCT and the Agency for Administrative Modernization, I.P. (AMA), through the project ForestSphere, ref. 2025.01496.DT4ST. It was further supported by COST Action NERO, CA22164, supported by COST (European Cooperation in Science and Technology). David Portugal was supported by the Scientific Employment Stimulus 5th Edition, contract reference 2022.05726.CEECIND funded by FCT. The support given by Agenda TransForm (PRR 02/C05-i01/2022), namely through the project ‘CENTRODEC’ (Centro de Apoio à Decisãocom Dados Multisensoriais para a Proteção da Floresta), with the contract of Carlos Ribeiro, is gratefully acknowledged. The support given by Nuno Luís and João Carvalho in performing the laboratory experiments is gratefully acknowledged. References [1] A. Fuller, Z. Fan, C. Day, and C. Barlow, “Digital Twin: Enabling Technologies, Challenges and Open Research,” IEEE Access, vol. 8, pp. 108952–108971, 2020, doi: 10.1109/ACCESS.2020.2998358. [2] “Digital Twin Consortium Home.” Accessed: Mar. 23, 2025. [Online]. Available: https://www.digitaltwinconsortium.org/ [3] P. Bauer, B. Stevens, and W. Hazeleger, “A digital twin of Earth for the green transition,” Nature Climate Change 2021 11:2, vol. 11, no. 2, pp. 80–83, Feb. 2021, doi: 10.1038/s41558021-00986-y. [4] M. Attaran and B. G. Celik, “Digital Twin: Benefits, use cases, challenges, and opportunities,” Decision Analytics Journal, vol. 6, p. 100165, Mar. 2023, doi: 10.1016/J.DAJOUR.2023.100165. [5] A. C. Tagarakis, L. Benos, G. Kyriakarakos, S. Pearson, C. G. Sørensen, and D. Bochtis, “Digital Twins in Agriculture and Forestry: A Review,” Sensors 2024, Vol. 24, Page 3117, vol. 24, no. 10, p. 3117, May 2024, doi: 10.3390/S24103117. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) [6] J. Döllner, R. De Amicis, J. M. Burmeister, and R. Richter, “Forests in the Digital Age: Concepts and Technologies for Designing and Deploying Forest Digital Twins,” Proceedings - Web3D 2023: 28th International Conference on Web3D Technology, p. 12, Oct. 2023, doi: 10.1145/3611314.3616067. [7] G. Sanchez-Guzman, W. Velasquez, and M. S. Alvarez-Alvarado, “Modeling a simulated Forest to get Burning Times of Tree Species using a Digital Twin,” 2022 IEEE 12th Annual Computing and Communication Workshop and Conference, CCWC 2022, pp. 639–643, 2022, doi: 10.1109/CCWC54503.2022.9720768. [8] C. Zhong, S. Cheng, M. Kasoar, and R. Arcucci, “Reduced-order digital twin and latent data assimilation for global wildfire prediction,” Natural Hazards and Earth System Sciences, vol. 23, no. 5, pp. 1755–1768, May 2023, doi: 10.5194/NHESS-23-1755-2023. [9] M. J. Best et al., “The Joint UK Land Environment Simulator (JULES), model description – Part 1: Energy and water fluxes,” Geosci Model Dev, vol. 4, no. 3, pp. 677–699, Sep. 2011, doi: 10.5194/GMD-4-677-2011. [10] J. Pereira, J. Mendes, J. S. S. Júnior, C. Viegas, and J. R. Paulo, “Metaheuristic algorithms for calibration of two-dimensional wildfire spread prediction model,” Eng Appl Artif Intell, vol. 136, p. 108928, Oct. 2024, doi: 10.1016/J.ENGAPPAI.2024.108928. [11] T. Gameiro, T. Pereira, C. Viegas, F. Di Giorgio, and N. F. Ferreira, “Robots for Forest Maintenance,” Forests, vol. 15, no. 2, p. 381, Feb. 2024, doi: 10.3390/f15020381. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) iFire AI: AI-powered Wildfire Simulation and 3D Immersive Visualisation Renhao Huang*,1, Ali Asadipour2, Dennis Del Favero1, Yang Song1 1iCinema UNSW, Sydney, Australia, {renhao.huang, d.delfavero, yang.song1}@unsw.edu.au 2Computer Science Research Centre, Royal College of Art, United Kingdom, ali.as[email protected] *Corresponding (presenting) author Summary Wildfires, especially extreme wildfires in the urban/rural interface, cause irreversible damage to ecosystems, human lives and economies globally. To reduce such losses, understanding wildfires is crucial for effective preparedness and response by first responders. We introduce iFire AI, aimed at developing the world's leading wildfire visualisation system by integrating AI with wildfire simulation and 3D immersive visualisation for first responders. We propose a deep learning-based wildfire simulation model, providing high-fidelity extreme wildfire behaviour data for our visualisation system. Such data will then be reconstructed using the Unreal Engine 5 and rendered on 3D immersive visualisation platforms, providing dynamically evolving hyper-realistic extreme fire landscapes. By offering life-like visualisations for dynamic wildfire scenarios at the urban/rural interface, we hope iFire AI can enhance first responder risk perception, situational awareness and collaborative decision-making, and thereby reduce risks due to extreme wildfires and promote strategic readiness. Keywords: Wildfire Visualisation, Wildfire simulation, Deep Learning, Emergency preparedness Figure 1 Overview of iFire AI System Figure 2 Pipeline of Training a DL Model X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) 1. Introduction Wildfires are unplanned, unpredictable, and erratic fires that affect vast open areas and urban interfaces [1]. Extreme wildfires at the urban/rural interface (URI) cause irreversible damage to both natural and urban environments and pose risks to life and property [2]. During 2019-20, Australia experienced its largest bushfires, which burnt 19 million hectares, killed at least 34 people and three billion wild animals, affected 450 people from smoke inhalation, and caused up to 100 billion US dollars in economic losses. To mitigate such losses, it is essential to understand wildfire behaviour by offering education and training sessions for firefighters, enhancing their risk perception, situational awareness and collaborative decision-making against wildfires [2]. With the development of computer technologies, researchers have been exploring enhanced wildfire simulation and immersive wildfire visualisation [3], enabling users to safely experience diverse wildfire scenarios without exposing them to real-life dangers. In this study, we propose iFire AI, an AI-powered wildfire simulation and 3D immersive visualisation system. iFire AI is built upon the iCinema Research Centre’s current system iFire, which translates the 2D wildfire scenarios into a first-person perspective using Unreal Engine 5 (UE5) and visualises them in our design-registered 3D immersive visualisation system AVIE and AVIE-SC [4], along with any type of smart screen (Figure 1). In particular, iFire AI has its AI-powered wildfire simulation, which uses advanced deep learning (DL) techniques to simulate the fire spread given the terrain information, ignition and weather conditions. We also propose the FARSITE-8K, a simulation dataset used to compare the capability of feature extraction from the landscape data. Compared to existing 2D and 3D visualisation systems, iFire AI allows a more intuitive and visceral engagement with wildfires by allowing full immersion inside a fire ground with a better sense of how extreme fires interact with different landscapes and environmental conditions [5]. Unlike other immersive visualisation platforms such as head-mounted virtual reality systems [5], iFire AI allows users to physically interact and collaborate with each other in wildfire scenarios, reflecting the real-world conditions of first responder team interaction. Finally, by learning from a large amount of wildfire data, our deep learningbased wildfire simulation model can generate higher-fidelity wildfire scenarios at the urban/rural interface than those produced by mathematical models. 2. Data and Methods The role of wildfire simulation in iFire AI is to provide realistic wildfire scenarios to our X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) visualisation system. Existing fire simulators [6], [7] are usually conducted via mathematical models, which solely rely on handcrafted physical rules and are thus often limited by their lack of scalability for complex landscapes involving interaction between ground and atmospheric processes. With the rapid advancement of artificial intelligence (AI) and, particularly, deep learning (DL), it has become increasingly feasible to incorporate AI methods for extreme event visualisation [8]. Our DL model follows a similar architecture to image segmentation models [6], [7], where the inputs and the outputs are 2D maps. Specifically, the inputs of the model are the landscape data 1 , including eight maps describing the terrain, tree canopy, and surface fuel. These maps are cropped into a constant size (e.g., 512x512) with the ignition as the centre and sent into the DL model to generate the 2D fire arrival map, with each pixel representing the time (in minutes) of the fire arrival at the location. To simplify the model training, we use a Sigmoid function to normalise the outputs, so that each pixel has a float value from 0 to 1. The model is optimised using an L1 loss function with an ADAM optimiser. The integration of the DL model in iFire AI follows the pipeline shown in Figure 1. A tablet interface is designed to allow users to select the scenarios and adjust ignitions. Then, it sends the request to the database to obtain the required input data and uses the DL model to generate the fire arrival maps. These inputs and outputs are further reconstructed into a 3D space using UE5, the most advanced 3D computer graphics game engine. The terrain information is built using pre-designed materials in UE5 based on the landscapes. Meanwhile, the renderings of flames, smoke and embers are built upon the particle effect with hard-coding to simulate their behaviours. Finally, we visualise the wildfire scenarios using the immersive simulation. As a system capable of taking users into a full-body 1:1 scale on a virtual experience of a wildfire in an exact geo-location, iFire AI involves using our 3D cinema AVIE and AVIE-SC, a 360-degree and 130-degree 3D immersive and interactive cinema (Figure 1: Top centre and top left). These systems exhibit wildfire scenarios on panoramic screens at 4K resolution. Users wear 3D glasses to enable a tangible experience of the fire ground. They support up to 30 users to interact simultaneously and directly with each other and the virtual environment, with no occlusions of the views. It is a promising visualisation system that has previously been successfully deployed for multiple first responder projects [9], [10]. Therefore, AVIE and AVIE-SC can be ideal visualisation systems to optimally visualise and experience wildfires. FARSITE-8K. To build a reliable wildfire simulation model, it is essential to evaluate the capability of extracting features from the landscape data among different model architectures. However, it is currently impractical to obtain such data from real-world sources to support our task. To mitigate this problem, we propose the FARSITE-8K dataset. We first sample 8,000 different regions across the US, each having an area of 15.36×15.36𝑘𝑚2. Then, for each region, we download the landscape data from the LANDFIRE (https://landfire.gov/). Each map contains 512×512 pixels with a spatial 1 https://landfire.gov/fuel/landscape X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) resolution of 30 meters. To obtain the fire arrival maps, we use FARSITE to simulate fire spread for 144 hours. The simulated fire spread starts from ignition at the centre of the region and follows constant weather conditions with a temporal resolution of one hour as parameters. This setting allows the DL model to consider only the landscape features once trained on this dataset. 3. Results and Discussion Evaluation Metrics. In our method, we select Mean Intersection of Union (𝑀𝐼𝑜𝑈) to evaluate our model, which calculates the average of the ratio between the intersection and the union of predicted 𝑃 and ground truth 𝐺𝑇 arrival time maps lying in short time slots 𝑀𝐼𝑜𝑈= 1 12∑(𝑡𝑖≤𝑃≤𝑡𝑖+1)∩ (𝑡𝑖<𝐺𝑇≤𝑡𝑖+1) (𝑡𝑖<𝑃≤𝑡𝑖+1)∪(𝑡𝑖<𝐺𝑇≤𝑡𝑖+1) where 𝑡𝑖∈{0,12,24,…,144}. We also calculate 𝐼𝑜𝑈𝑎𝑙𝑙 = (𝑃≤144)∩ (𝐺𝑇≤144) (𝑃≤144)∪(𝐺𝑇≤144) , evaluating the overall burning areas. Results. In this part, we test the capability of existing segmentation models on the FARSITE8K Dataset. Table 1 Performance of models on the FARSITE-8K dataset UNet++ [11] PSPNet [12] DeepLabv3+ [13] SegFormer [14] SegFormer-MiT [14] 𝑀𝐼𝑜𝑈 19.47% 18.00% 17.94% 18.90% 19.26% 𝐼𝑜𝑈𝑎𝑙𝑙 64.43% 62.59% 60.44% 60.12% 60.56% DL models are trained using 80% the dataset following the pipeline in Figure 2 and validated on 5%. Finally, they are tested using the remaining 15% of the dataset. In Table 1, all models achieve above 60% of 𝐼𝑜𝑈𝑎𝑙𝑙 scores. Also, UNet++ performs best among all models, followed by PSPNet. Surprisingly, SegFormer only performs slightly better than DeepLabv3, indicating that MLP may not outperform CNN for decoders on wildfire simulation. Finally, the Mixed Transformer (MiT) encoder outperforms ResNet34 on SegFormer, indicating that Transformer-based decoders are potentially better than CNNs. Finally, all models achieve around 18% of 𝑀𝐼𝑜𝑈, indicating that generated fire arrivals do not well match the ground truth, suggesting future improvements. For the immersive visualisation system, we have demonstrated our prototype version of iFire, using a hypothetical Australian pine plantation fire, a grasslands fire in the Australian state of Victoria in 2022 and the Bridger Foothills Fire in Montana, USA in 2021. We received X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) positive comments and feedback from our stakeholders. Currently, the iFire visualisation system is being deployed by FRNSW for its Emergency Services Academy to train incident commanders and fire station commanders and the Australian Broadcasting Corporation (ABC) for its online news divisions to educate its 12M+ audience. More details can be found on our iFire project page. 4. Conclusions and Future Work This study introduces iFire AI, aimed at developing the world's leading wildfire visualisation system by combining DL-based wildfire simulation and 3D immersive visualisation. The DL model provides realistic wildfire scenarios for our visualisation system. These scenarios are then reconstructed into 3D space using UE5 and are finally rendered on 3D immersive visualisation systems AVIE and AVIE-SC, providing dynamically evolving hyper-realistic URI landscapes in extreme fires. We also design a FARSITE-8K dataset, with collected landscape maps from LANDFIRE and simulated fire arrival maps with constant weather conditions, to explore the capability of the feature extraction on landscapes among different DL models. By offering life-like visualisations for dynamic URI wildfire scenarios, we hope iFire AI can enhance first responder risk perception, situational awareness and collaborative decisionmaking and thereby reduce vulnerabilities due to extreme wildfires and promote long-term readiness. At the current stage, our DL model is successfully built upon the simulation dataset FARSITE-8K, illustrating its capability to learn from the data. To maximise the advantage of deep learning, future studies will explore improving the model by using real-world data, e.g., model fine-tuning. We will also integrate dynamic weather conditions to enable users to see how the same fire may evolve under future climate conditions. Furthermore, the DL model can generate extra outputs, such as smoke and flame behaviour, to enhance 3D modelling. References [1] F. Tedim, V. Leone, M. Amraoui, C. Bouillon, M. R. Coughlan, G. M. Delogu, and Others, “Defining extreme wildfire events: difficulties, challenges, and impacts,” Fire, pp. 1– 9, 2018. [2] G. Drummond, D. Habibi, and M. Cattani, A handbook of wildfire engineering. Bushfire & Natural Hazard Cooperative Research Centre, Victoria, 2020. [3] C. A. T. Cortes, S. Thurow, A. Ong, J. J. Sharples, T. Bednarz, G. Stevens, and D. Del Favero, “Analysis of wildfire visualization systems for research and training: are they up for the challenge of the current state of wildfires?,” IEEE Trans. Vis. Comput. Graph., pp. 4285– 4303, 2023. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) [4] M. McGinity, J. Shaw, V. Kuchelmeister, A. Hardjono, and D. D. Favero, “AVIE: a versatile multi-user stereo 360° interactive VR theatre,” in Proceedings of the Workshop on Emerging Displays Technologies: Images and Beyond: the Future of Displays and Interacton, 2007, pp. 2–5. [5] R. V. Hoang, M. R. Sgambati, T. J. Brown, D. S. Coming, and F. C. Harris Jr, “VFire: immersive wildfire simulation and visualization,” Comput. Graph., pp. 655–664, 2010. [6] M. A. Finney, FARSITE, fire area simulator–model development and evaluation. US Department of Agriculture, Forest Service, Rocky Mountain Research Station, 1998. [7] C. Miller, J. Hilton, A. Sullivan, and M. Prakash, “SPARK–a bushfire spread prediction tool,” in Environmental Software Systems. Infrastructures, Services and Applications, 2015, pp. 262–271. [8] Y. Song, M. Pagnucco, F. Wu, A. Asadipour, and M. J. Ostwald, “Intelligent architectures for extreme event visualisation,” in Climate Disaster Preparedness: Reimagining Extreme Events Through Art and Technology, Springer, 2024, pp. 37–48. [9] D. Del Favero, Y. Song, K. Moinuddin, C. Green, J. Sharples, A. Ong, and N. Brohier, “Penumbra2.0,” in SIGGRAPH Asia 2023 Art Gallery, 2023. [10] D. Del Favero, N. Brown, J. Shaw, and P. Weibel, “T_Visionarium: the aesthetic transcription of televisual databases,” in Present Continuous Past(s): Media Art. Strategies of Presentation, Mediation and Dissemination, Springer, 2005, pp. 132–141. [11] Z. Zhou, M. M. Rahman Siddiquee, N. Tajbakhsh, and J. Liang, “UNet++: a nested Unet architecture for medical image segmentation,” in Deep Learning in Medical Image Analysis and Multimodal Learning for Clinical Decision Support, 2018, pp. 3–11. [12] H. Zhao, J. Shi, X. Qi, X. Wang, and J. Jia, “Pyramid scene parsing network,” in IEEE Conference on Computer Vision and Pattern Recognition, 2017, pp. 6230–6239. [13] L.-C. Chen, Y. Zhu, G. Papandreou, F. Schroff, and H. Adam, “Encoder-decoder with atrous separable convolution for semantic image segmentation,” in European Conference on Computer Vision, vol. 11211, Cham, 2018, pp. 833–851. [14] E. Xie, W. Wang, Z. Yu, A. Anandkumar, J. M. Alvarez, and P. Luo, “SegFormer: simple and efficient design for semantic segmentation with transformers,” in Advances in Neural Information Processing Systems, 2021, vol. 34, pp. 12077–12090. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Modeling the largest wildfire using FARSITE in the Czech Republic: National Park Bohemian Switzerland 2022 Lucie Kudláčková1,2*; Markéta Poděbradská1,2; Monika Hlavsová1,2; Emil Cienciala1,3; Jana Beranová3; Charles McHugh4; Mark Finney4; Jan Novotný1; Pavel Zahradníček1,5; Petr Štěpánek1,5; Rostislav Linda6; Miroslav Pikl1; Dana Vébrová7; Martin Možný1,5; Peter Surový8; Zdeněk Žalud1,2; Miroslav Trnka1,2 1Global Change Research Institute of the Czech Academy of Sciences, Czech Republic, [email protected]; [email protected], [email protected], [email protected], [email protected], [email protected], pikl[email protected], [email protected] 2Mendel University in Brno, Czech Republic, [email protected] 3IFER – Institute of Forest Ecosystem Research, Ltd., Czech Republic, [email protected], [email protected] 4Missoula Fire Sciences Laboratory, US Forest Service, USDA, Missoula, USA, [email protected], mar[email protected] 5Czech Hydrometeorological Institute, Brno, Czech Republic; [email protected] 6Forestry and Game Management Research Institute, Czech Republic, rostislav.lind[email protected] 7Bohemian Switzerland National Park, Czech Republic, [email protected] 8Czech University of Life Sciences, Czech Republic, [email protected] *Corresponding (presenting) author Summary In July 2022, Bohemian Switzerland National Park experienced Czechia's largest recorded wildfire, burning over 1,060 hectares. We used FlamMap (respective FARSITE) to simulate fire behavior under various scenarios combining fuel types and weather conditions. Results indicated that the fire could have reached a similar extent under alternate settings, though healthy, closed-canopy forests would have notably limited its spread. This study enabled the first FARSITE calibration in Central Europe, confirming its relevance for regional wildfire modeling. The findings support improved forest management, highlight the role of vegetation structure, and inform future wildfire preparedness under climate change. Keywords: Czech Republic, modeling, FARSITE, FlamMap, wildfire 1. Introduction The wildfire in Bohemian Switzerland National Park in the summer of 2022 was the largest recorded in the modern history of the Czech Republic, burning over 1,060 hectares on the Czech side and approximately 113 hectares on the German side. The event, which occurred in a landscape heavily affected by bark beetle-induced spruce dieback, introduced new challenges for fire risk assessment and fire modeling in Central Europe [1]. While fire X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) behavior models are commonly used in North America, the Mediterranean, and other parts of the world, their applicability under Central European conditions has remained limited. This study aimed to simulate the 2022 event using the FARSITE model [2] (part of the FlamMap tool [3–4]) and to evaluate various scenarios combining alternative vegetation structures and climatic-meteorological conditions, including variables such as fuel moisture, wind, drought, and forest composition. The study provides insight into wildfire dynamics in temperate European climates, where not only fuel structure but also weather and climate variability play crucial roles. Moreover, it offers valuable knowledge to improve fire management and preparedness strategies. Importantly, this exceptional event provided a unique opportunity to verify and calibrate the development of high-resolution (5 m) input layers for fire behavior modeling, which are typically not available for the territory of the Czech Republic. 2. Data and Methods The simulation framework was based on a combination of high-resolution spatial data and advanced modeling tools for predicting fire behavior. Terrain was represented using a 5 m resolution digital elevation model, supplemented with forest structure derived from airborne LiDAR data, including stand height, canopy bulk density, canopy base height and canopy cover. Fuel characteristics were classified according to standardized models [5], while meteorological inputs were taken from observations during the actual wildfire event in July 2022. To account for local wind flow in complex terrain, the WindNinja model was applied [6]. Fire behavior was simulated using the FlamMap system (FARSITE model). In addition to the real-case scenario, a set of seven hypothetical scenarios was developed, reflecting different combinations of vegetation structure (e.g., bark beetle-affected dead stands, healthy spruce forests, natural mixed forests, clearings) and environmental variables (e.g., reduced temperature, varying wind intensity, different levels of drought and fuel moisture). This scenario-based approach allowed for the evaluation of the relative influence of individual factors on fire spread and intensity under Central European conditions. It also provided a unique opportunity for the first calibration of the FARSITE model based on a real wildfire event in the Czech Republic and Central Europe, increasing its relevance and transferability to similar regions. 3. Results and Discussion The FlamMap model proved to be an effective tool for simulating wildfire spread under Central European conditions. The simulations accurately captured the spatial dynamics of X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) the fire and highlighted the key role of forest structure, fuel condition, and meteorological factors—particularly wind [1]. Scenario analyses revealed that natural mixed forests or healthy spruce stands significantly limited fire spread compared to dead spruce monocultures. In contrast, removing deadwood led to faster, though less intense, fire propagation. These findings underscore the importance of forest species composition and suggest that fuel reduction alone may not be sufficient to limit fire extent [1]. The model also confirmed that inaccurate input data—especially wind measurements—can strongly affect the reliability of simulations. Overall, FlamMap demonstrates strong potential as a decision-support tool for planning prevention and response strategies to future extreme wildfire events in a changing climate. Figure 1. Simulated rate of spread (m/min) for Scenario 1 (Reality) – the July 2022 wildfire in Bohemian Switzerland National Park, Czech Republic (FARSITE model) [1]. 4. Conclusions This study demonstrated the effectiveness of the FARSITE model (part of FlamMap) for simulating wildfire behavior in Central Europe [1]. It represents the first calibration of the FARSITE model based on a real fire event in the Czech Republic. The findings highlight the importance of forest species composition in mitigating fire spread and confirm the potential of scenario-based modeling for forest management planning. Future research could focus on expanding validation datasets, testing additional model parameters, and gradually integrating modeling into decision support for forest protection and crisis management— serving, for example, as a complementary tool for fire and rescue services or protected area authorities. Acknowledgements This work was supported by the projects by the Ministry of Education, Youth and Sports of the Czech Republic (grant AdAgriF—Advanced methods of greenhouse gases emission reduction and sequestration in agriculture and forest landscape for climate change X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) mitigation; CZ.02.01.01/00/22_008/0004635), Interreg Central Europe and the European Union in the framework of the project Clim4Cast (study about forecasting drought, heatwave, and fire weather in central Europe— Clim4Cast; CE0100059), PERUN TAČR Project (SS02030040) and program Transadapt. References 1. Kudláčková, L.; Poděbradská, M.; Bláhová, M.; Cienciala, E.; Beranová, J.; McHugh, C.; Finney, M.; Novotný, J.; Zahradníček, P.; Štěpánek, P.; et al. Using FlamMap to assess wildfire behavior in Bohemian Switzerland National Park. Nat. Hazards 2024, 120, 3943– 3977. https://doi.org/10.1007/s11069-023-06361-8 2. Finney, M.A. FARSITE: Fire Area Simulator—Model Development and Evaluation. U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station, 1998. https://doi.org/10.2737/rmrs-rp-4 3. Finney, M. An Overview of FlamMap Fire Modeling Capabilities. In Fuels Management— How to Measure Success: Conference Proceedings, Andrews, P.L., Butler, B.W., Comps.; U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station: Fort Collins, CO, USA, 2006; Proceedings RMRS-P-41, pp. 213–220. 4. Finney, M.A.; Brittain, S.; Seli, R.C.; McHugh, C.W.; Gangi, L. FlamMap: Fire Mapping and Analysis System (Version 6.0); U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station: Fort Collins, CO, 2019. [Software] 5. Scott, J.H.; Burgan, R.E. Standard Fire Behavior Fuel Models: A Comprehensive Set for Use with Rothermel’s Surface Fire Spread Model. U.S. Department of Agriculture, Forest Service, Rocky Mountain Research Station: Fort Collins, CO, USA, 2005; RMRS-GTR-153. 6. Forthofer, J.M.; Butler, B.W.; Wagenbrenner, N.S. A Comparison of Three Approaches for Simulating Fine-Scale Surface Winds in Support of Wildland Fire Management. Part I. Model Formulation and Comparison Against Measurements. Int. J. Wildland Fire 2014, 23, 969–981. https://doi.org/10.1071/WF12089. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Session 2 Observing Extreme Wildfire Events X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Mapping wildfire progressions and measuring rate of spread from “line scans” in Australia Dr Michael Storey1,*; Assoc. Prof Owen Price2 1University of Wollongong, Australia, [email protected] 2 University of Wollongong, Australia *Corresponding (presenting) author Summary We have developed a prototype database containing over 9,000 fire progressions and 727 individual rate of spread (ROS) measurements from Australian wildfires. 237 spread faster than 1 km/h, and 66 spread > 3 km/h, with a maximum of 18.4 kmh. Progressions are mapped from aerial line scans and satellite imagery. While created for an ongoing Bayesian ROS modelling project, the database has the potential to become a widely accessible resource for wildfire researchers. It currently has the largest set of wildfire spread observations currently available in Australia. Keywords: line scan, wildfire spread, database 1. Introduction Wildfires can spread too rapidly and intensely to allow for detailed, systematic observations of their progression. In Australia, the most accurate way to map these fires is through the use of multi-spectral scanners mounted on aircraft, which capture geo-referenced images of active fires [1]. These images—known as “line scans”—and the resulting GIS progression polygons have a wide range of applications, including model development and validation, identifying extreme fire behaviours, analysis of suppression effectiveness, and post-fire reviews by fire agencies. Much of this data is held by fire agencies, but it is not always stored consistently, making it difficult to access and process for research purposes. Improving access to this data would significantly support research to better understand wildfire behaviour. As part of a project to develop a Bayesian rate of spread (ROS) model, we have collected and processed more than 9,000 line scan and satellite images of wildfires, including extreme fires, across Australia. We have mapped fire progressions from this data (some progressions were provided by fire agencies) and measured ROS for many fires. Additionally, we are developing a prototype ROS database to store the data in a structured format and enable efficient searching. We have also created an initial R package [2] that automates X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) tasks such as database queries, satellite data retrieval, fire progression mapping, and matching ROS measurements to weather data. The next part of our project will use this database to develop the Bayesian wildfire rate of spread model [3,4]. This model will generate predictions that explicitly incorporate and communicate uncertainty in wildfire spread. However, a major goal of the overall project is to continue to explore the establishment of a publicly accessible, consistently maintained database of wildfire spread in Australia. 2. Data and Methods The main source of fire observations in our database is from aerially acquired “line scans”. These scans are produced by sensors mounted on aircraft flown over active fires. In most cases, the resulting images clearly show the active fire boundary and burnt areas. However, cloud cover can obstruct the view, and burnt areas may be difficult to distinguish depending on vegetation type. While the scanners are capable of capturing multiple spectral bands, typically only a subset is provided to fire agencies—either a single thermal infrared band (with high values artificially coloured red in the final image) or a combination of three bands (one blue and two shortwave infrared), where active fire appears as yellow-orange (Figure 1). Figure 1. Example of a 3-band line scan image from a fire in 2019-2020 in south-eastern Australia. Most intense fire in yellow, recently intense fire in orange. Dark green vegetation is forest and light green is grass. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) To generate GIS fire progression polygons from these images, the process usually involves manual mapping in ArcGIS by fire operations staff. The accuracy of these manually created progressions can vary depending on the time available to the mapper—who may be under pressure during major fire events—and whether any post-fire analysis was conducted. Detailed post-event mapping is usually more accurate. For our project, we were provided with both line scans and some mapped progression polygons. Where progressions were missing, we manually mapped them ourselves. We also reviewed and corrected polygons provided by fire agencies, including DEECA (Victoria), RFS (NSW), and DFES (WA). For most scans, we mapped fire progressions where line scan pairs were available between 20 minutes and 5 hours apart. This time window allowed us to calculate rate of spread (ROS) during periods of likely continuous fire activity, avoiding stretches where fire behaviour may have paused or slowed significantly. For the 2019–2020 wildfires in southeastern Australia, we mapped fire progressions from all available scans as part of work conducted for the NSW Bushfire Inquiry. In some cases where no line scan data existed, we used supplementary satellite data (VIIRS, MODIS, HIMAWARI, Landsat, and Sentinel-2) to map fire progression. The choice of satellite source depended on factors such as timing relative to line scans, fire size, image resolution, and cloud cover. We developed a prototype Postgres database to store fire progression polygons. Each entry in the database includes geometry, progression time, a link to the AWS-hosted line scans used, fire type (main or spot), and fire name. We also developed an associated R package that includes functions to automate ROS measurement from two progression polygons (Figure 2), query the database, download line scan and satellite data, sample local reanalysis weather data and extract fire areas from line scans using Meta’s Segment Anything algorithm [5] for efficient polygon mapping. Both the database and the R package are still under development, but we aim gain support to develop them into a widely accessible and regularly updated resource for fire behaviour research. 3. Results and Discussion Our project is still in progress; however, there are already some initial results in terms of data and database development. Our database currently contains over 9,000 unique fire progression times. Our automated approach to measuring rate of spread (implemented as a function in our R package) has so far produced 727 unique ROS measurements (Figure 3). Most of these measurements are from fires in New South Wales and Victoria. Of the total, 33% exceed 1 km/h, while 9% exceed 3 km/h. The maximum recorded ROS was 18.3 km/h, observed in a fire in Western Australia. Many images reveal extreme fire behaviours, including mass spotting, which could be coded in the database and extracted for future research. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Figure 2. Example of a rate of spread (ROS) measurement using two line scans (day left, night right). Blue polygons are the fire edges mapped from both scans (overlaid on both images). Pink line is an automatically generated line of maximum spread from an R package function, red square is end of line. Spread line = 2.6 km, spread time = 41 minutes, ROS = 3.8 km\h Spot-fires not included in ROS measurement. We are continuing to develop our R package, but the functions added so far work well and integrate effectively with the Postgres database. The database serves as a straightforward prototype that demonstrates the advantages of structured data storage. However, further development of the database is needed in the future, and funding for this work is still needed. We have conducted some initial testing of the "Segment Anything" approach in R for mapping polygons. The early results are promising—many progressions can be mapped and saved to a GeoPackage from a scan in just one to two minutes. In contrast, the previous approach, which relied entirely on manual mapping in ArcGIS, takes at least five to ten minutes per scan, as each scan must be manually loaded, mapped, and saved. Manual mapping time can increase significantly depending on the fire's complexity and size. The Segment Anything algorithm captures burning edges extremely well but can struggle with previously burnt edges that are not clear to the user, or with dense smoke or cloud cover. Our process includes a step for manual oversight and correction of the automated polygons to ensure accuracy. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Figure 3. Histogram of rate of spread (ROS km/h) measurements extracted from fire progression polygons. 4. Conclusions To date, we have produced a large set of wildfire progressions and rate of spread measurements. We have demonstrated a structured storage approach for this data through our prototype database. We have also shown how data can be produced more easily and quickly using our R package and image segmentation. In addition to contributing to our Bayesian ROS model, our aim is to further develop the database and tools to provide a method for generating and storing ROS data, and to allow easy access. This will help remove a significant roadblock in Australia (finding spread data) which can hinder fire behaviour research. Acknowledgements Our research is funded by Natural Hazards Research Australia. References 1. McRae, R. Remotely Mapping Fires. Aust. J. Emerg. Manag. 2022, 37, 45–51, doi:10.3316/informit.455216453245402. 2. R Core Team R: A Language and Environment for Statistical Computing 2020. 3. Kruschke, J. Doing Bayesian Data Analysis: A Tutorial with R, JAGS, and Stan; Academic Press, 2014; ISBN 978-0-12-405916-0. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) References [1] Castellnou, M., et al. FIRE-RES Transfer of Lessons Learned on Extreme Wildfire Events. FIRE-RES project. 2022. [2] Castellnou, M., et al. Protocols for data collection on Extreme Wildfire Events. EWED Project. 2024. [3] Castellnou, M., et al. Pyroconvection classification based on atmospheric vertical profiling correlation with extreme fire spread observations. JGR: Atmospheres, 127, e2022JD036920. [4] Tedim, F., et al. Defining extreme wildfire events: Difficulties, challenges, and impacts. Fire, 1(1), 9. 2018. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Estimating the Probability of Pyrocumulus Formation in Extreme Wildfire Scenarios: A Preliminary Approach Néstor Bravo1; Carlos Carrillo *, 2; Javier Panadero 2; Ana Cortés 2 and Tomàs Margalef 2 1Universitat Autònoma de Barcelona, Spain, [email protected] 2Universitat Autònoma de Barcelona, Spain, {carles.carrillo; javier.panadero; ana.cortes; Tomas.margalef} @uab.cat *Corresponding (presenting) author Summary This study presents a methodology to predict pyrocumulus formation based on the thermal power released by wildfires and atmospheric conditions. By integrating fire behavior simulations with the Pyrocumulonimbus Firepower Threshold (PFT), we evaluate the 2022 El Pont de Vilomara wildfire. Results show that fire power briefly exceeded the PFT, aligning with observed pyrocumulus development, but was insufficient to sustain it. The findings highlight the potential of the proposed framework and emphasize the need to incorporate convective energy losses to improve predictive accuracy. Keywords: Extreme Wildfire, Pyrocumulus, Wildfire energy, Pyrocumulonimbus Firepower Threshold 1. Introduction Extreme Wildfire Events (EWEs) represent a growing and critical challenge for societies across the globe. Although they account for a small fraction of all wildfires, their destructive potential and impact on human lives and ecosystems are often disproportionate [1]. These fires exceed the current control capacity, even in wellprepared regions and areas with extensive firefighting experience. Due to climate change, the number and frequency of fires have increased significantly. This has been reflected in devastating fires in Portugal, Greece, and the USA, as well as in large fires in Catalonia, such as Torre de l'Espanyol (2019, 4,281 hectares). These fires are usually related to extreme phenomena such as the formation of pyrocumulus clouds (clouds generated due to atmospheric conditions and the presence of a large fire), which can generate electrical storms, winds, and sudden changes in the direction of the fire, or other phenomena such as the formation of fire tornadoes and the projection of burning fuel over long distances. Some studies highlight that at this point, the behavior of the fire cannot be simulated by the current fire spread models [2]. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Despite the growing recognition of these phenomena, there remains no operational methodology capable of determining in real time whether a fire will remain within expected behavior limits or escalate into an EWE. In particular, the identification of conditions conducive to pyrocumulus or pyrocumulonimbus formation remains a key gap in predictive wildfire science. By combining wildfire thermal power estimates derived from fire spread simulations with atmospheric instability metrics, the method aims to identify critical time intervals during which pyroconvective transitions become thermodynamically favorable. Such capability would support earlier detection of regime shifts in fire behavior, thereby enhancing strategic wildfire management and improving the effectiveness of emergency response protocols. 2. Data and Methods The formation of pyrocumulonimbus clouds requires specific atmospheric conditions combined with sufficient thermal energy release from wildfire. The methodology proposed in this study aims to quantify this relationship through two complementary techniques: • To estimate the wildfire energy release through the utilization of a forest fire spread simulator, enabling direct comparison with the calculated PFT values. • Application of the Pyrocumulonimbus Firepower Threshold (PFT) [3], which establishes the minimum power release required for pyrocumulonimbus development within a given atmospheric condition. 2.1. Estimation of Wildfire Energy Release The thermal power released by an active wildfire is estimated using an equation derived from Catchpole et al. [4]. formulated an expression that relates wildfire power to the rate of change in burned area over time: 𝑃 = ∫𝐼 𝑑𝑆 = 𝐻 𝜔 𝑑𝐴 𝑑𝑇 Where 𝑃 represents the instantaneous wildfire power in watts (W), 𝐼 denotes the fireline intensity (W/m), 𝑆 is the perimeter length (m), 𝐻 signifies the heat released per unit area (J/𝑚2), 𝜔 is the mass flow rate (kg/(𝑚2/s)), and 𝑑𝐴 𝑑𝑡 is the rate of area change (𝑚2/s). 𝐻 and 𝜔 can be obtained in lcp data files. And 𝑑𝐴 𝑑𝑡 can be computed using a wildfire simulator by knowing the total area burned and the time elapsed. 2.2. Pyrocumulonimbus Firepower Threshold (PFT) X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) The Pyrocumulonimbus Firepower Threshold (PFT) defines the minimum wildfire energy needed to trigger pyrocumulonimbus formation under given atmospheric conditions, based on three key variables. PFT = 0.3 ⋅𝑧fc ⋅𝑈ML ⋅Δ𝜃fc Where 𝑧fc represents the free-convection height (m), 𝑈ML denotes the mixed-layer wind speed (m/s), and Δ𝜃fc signifies the potential temperature difference (K) between the mixedlayer and the free-convection level. These instances correspond to conditions where pyrocumulus formation becomes thermodynamically favorable. Conversely, if the energy released remains below the PFT, the development of pyrocumulus clouds is unlikely, even in the presence of supportive atmospheric profiles. 2.3. Coupled Analysis of Fire Power and Atmospheric Favorability To predict the risk of pyrocumulus formation, both components, Estimation of Wildfire Energy Release and PFT are compared. When a simulated wildfire power exceeds the calculated PFT for a sustained period, pyrocumulonimbus formation becomes thermodynamically favorable. Conversely, when fire power remains below this threshold, cloud formation is unlikely despite otherwise conducive atmospheric conditions. The combination of these two components provides a simple and fast methodology for assessing the risk of pyrocumulus formation and identifying transitions between fire behavior to extreme fire behavior. The case study focuses on the wildfire that occurred in El Pont de Vilomara (Catalonia) between 17 and 18 July 2022, with an affected total area of 1,422 hectares. This event is of particular interest because the fire released sufficient thermal energy to initiate pyrocumulus development, but not enough to sustain it over an extended period. Atmospheric data for this study is derived from ERA5. 3. Results and Discussion The simulation was carried out using the FARSITE fire simulator [5]. Using a temporal discretization of 35 minutes and 35 meters, the results were compared with the PFT values calculated. Figure 1 shows that the power released by the fire exceeded the PFT values during the first hours of the fire, which coincides with the reported formation of pyrocumulus by Bombers de Catalunya [6]. However, as it is noticed in the report, the pyrocumulus clouds disappeared a few hours after their formation, which indicates that the power released by the fire was not high enough to maintain them for a long time. This is consistent with the results obtained from the simulation, which show that, although the power released by the X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) fire was still high, PFT values increased significantly, making it unlikely for pyrocumulus clouds to form again. Figure 1: Power released by the fire and PFT values during the fire in El Pont de Vilomara (Catalonia) in 2022. Figure 1 should be interpreted carefully, as convection losses were not considered when computing the fire power, making the results not directly comparable with the PFT values. Nevertheless, the results demonstrate that the power released was higher during a short period when the PFT was lower; however, this relationship did not persist long enough to sustain pyrocumulus development. During the observed pyrocumulus formation, during the afternoon, our simulation shows higher power values and lower PFT values, while when no pyrocumulus was observed, during the night, fire power decreased as PFT values increased. These findings suggest that the methodology proposed in this study provides a promising foundation for predicting pyrocumulus formation, though further refinements are needed to improve predictive accuracy. 4. Conclusions The integration of meteorological indices with fire behavior modeling offers a promising approach to predicting pyrocumulonimbus formation. The findings of this study remark on the importance of integrating meteorological indices and fire behavior modeling to enhance X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) predictive capabilities regarding pyrocumulonimbus formation. The 2022 El Pont de Vilomara wildfire demonstrates the effectiveness of the Pyrocumulonimbus Firepower Threshold (PFT) in linking wildfire energy release to atmospheric conditions. While fire power initially exceeded the PFT values during the initial hours of the event, aligning with observed pyrocumulus cloud formation, sustained formation was not achieved due to insufficient energy input over time. These results underscore the importance of accounting for convective energy losses and highlight the need for further research to refine predictive tools and deepen our understanding of fire–atmosphere interactions. To enhance the accuracy of wildfire power estimates and improve pyrocumulonimbus prediction capabilities, future research should focus on implementing coupled fireatmosphere models that would provide more realistic simulations. Additionally, incorporating high-resolution meteorological data and real-time fire detection systems would further enhance the operational value of the proposed methodology. The combination of improved power calculations with advanced PFT assessments could lead to the development of an early warning system for extreme wildfire behavior, potentially providing critical lead time for emergency response actions during high-risk fire events. Acknowledgements This work has been granted by the Ministerio de Ciencia e Innovación MCIN AEI/10.13039/501100011033 under contract PID2020-113614RB-C21 and PID2023146193OB-I00, and CPP2021-008762 by the European Union-NextGenerationEU/PRTR. It has also been supported by the Catalan Government under grant 2021-SGR-574. The authors also acknowledge the contributions of Kevin J. Tory on Pyrocumulonimbus Firepower Threshold as key aspect of this study. References 1. F. Tedim, V. Leone, M. Coughlan, C. Bouillon, G. Xanthopoulos, D. Royé, F. J. Correia, and C. Ferreira. 1 - extreme wildfire events: The definition. In F. Tedim, V. Leone, and T. K. McGee, editors, Extreme Wildfire Events and Disasters, pages 3–29. Elsevier, 2020. ISBN 978-0-12-815721-3. URL https://www.sciencedirect.com/science/article/pii/B9780128157213000011. 2. Castellnou, M., Bachfischer, M., Miralles, M., Ruiz, B., Stoof, C. R., & Vilà-Guerau de Arellano, J. (2022). Pyroconvection classification based on atmospheric vertical profiling correlation with extreme fire spread observations. Journal of Geophysical Research: Atmospheres, 127, e2022JD036920. https://doi. org/10.1029/2022JD036920 3. K. J. Tory, W. Thurston, and J. D. Kepert. Thermodynamics of pyrocumulus: A conceptual study. Monthly Weather Review, 146(8):2579 – 2598, 2018. doi:10.1175/MWR-D-17- X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) 0377.1. URL https://journals.ametsoc.org/view/journals/mwre/146/8/mwr-d-170377.1.xml. 4. Catchpole,E. A. and Mestre,N. J. de and Gill,A. M., 19830684522, English, Journal article, Australia, 0004-914X, 12, (1), Canberra, Australian Forest Research, (47–54), CSIRO, Forestry and Timber Bureau, Intensity of fire at its perimeter., (1982) 5. Mr. A. Finney. FARSITE: Fire area simulator - model development and evaluation. Research Paper RMRS-RP-4 Revised, USDA Forest Service, Rocky Mountain Research Station, Fort Collins, 2004. 6. Departament d’Interior de la Generalitat de Catalunya. Informes d’incendis forestals. https://interior.gencat.cat/web/.content/home/030_arees_dactuacio/bombers/foc_for estal/consulta_incendis/ (2022). X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Evolution of Extreme WUI Wildfire Incidents Forensic ChronoSpatial Reconstruction Methodology – Varnavas/Greece 2024 application Nikolaos Kamakiotis*,1, Georgios Papavasileiou2, Theodoros M. Giannaros2 1 Flevo Ltd., Cyprus, [email protected] 2 National Observatory of Greece/Institute for Environmental Research and Sustainable Development, Greece Summary This work presents the evolution of a forensic digital reconstruction methodology, applied to the August 2024 Varnavas wildfire (Attica, Greece), a high-impact Wildland–Urban Interface event. Integrating satellite imagery, field observations, meteorological data, and geotagged user-generated content, the spatial and temporal progression of fire fronts to be reconstructed. Key outcomes include Rate of Spread (RoS) estimation, burn rate per cultivated plot by crop type and condition, and impact assessment on the built environment. The study supports tactical response analysis, resilience planning, public awareness, and the enhancement of AI wildfire simulation models through reliable data-driven learning inputs. Keywords: WUI Wildfire, Reconstruction, Fire spread Analysis, Public Safety 1. Introduction Many extreme Wildland–Urban Interface (WUI) wildfire incidents have been recorded, driving research and policy initiatives at European Union and national levels to adapt accordingly towards mitigating devastating impacts [1]. The August 2024 Varnavas wildfire (Attica, Greece) exemplifies this growing threat: ignited under adverse fire weather conditions, the fire rapidly spread through rural area, breached the urban grid, and resulted in extensive destruction of residential and commercial properties, and ultimately, one fatality. Since response time remains a critical constraint [2], high-resolution spatiotemporal reconstruction at short intervals (10–30 minutes) was deemed essential. This forensic-level reconstruction aimed to identify fire behavior patterns, assess exposure of the built environment, and analyze the influence of agricultural assets on fire front propagation. The key objective is to demonstrate how digital reconstruction can be leveraged to raise public awareness by providing intuitive visualizations that convey the speed and intensity of X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) WUI wildfires, which are projected to increase in frequency under climate change scenarios [3]. Furthermore, the study serves to enhance an evolving innovative methodology that identifies temporal fire spread milestones and geospatial pathways using diverse sources of evidence, including satellite data, user-generated content (e.g., social media images and videos), conventional media, leased firefighting aircraft flight paths, and firsthand interviews with incident participants (e.g. residents and voluntary firefighters). 2. Data and Methods 2.1 Method The methodology used in this study was developed by N. Kamakiotis to analyze WildlandUrban Interface (WUI) wildfire incidents with a focus on public safety. The process aims to map detailed fire front spread patterns. It includes three main stages: data collection, processing, and the design of isochronous fire front curves [4]. The first step involves identifying fire front milestones—specific times and locations where the fire front was observed, including starting point. Once these are determined, they are manually drawn on recent high-resolution satellite imagery, representing fire fronts as curves corresponding to specific time intervals. After identifying fire front milestones, intermediate isochronous curves are drawn between them, considering influencing parameters such as wind speed and direction, topography, vegetation types and density, continuity of fuels and physical barriers [5]. Each parameter’s local influence is assessed to determine the density and orientation of the isochronous curves. The exact area within which the isochronous curves are designed is defined via the detailed design of the actual fire scar in reference to multiple satellite pictures after and before each incident. The final step is creating a digital animation that links static curves with their respective timestamps, enabling a visual reconstruction of fire spread over time. The method had already been applied on Arakapas wildfire incident (Cyprus/July 2021) and its outcome has been included in relevant NERO Network’s database (Figure 1). 2.2 Data Sources Fire front milestones are identified using diverse data sources that provide spatial and temporal clues on fire behaviour. These include: - Site inspections - Videos and images published by conventional and social media users documenting the fire's progress. - Interviews with individuals involved in firefighting or who directly observed the incident. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) - Satellite imagery showing fire progression and burn scars. - Web applications (e.g., FlightRadar24) tracking firefighting aircraft paths to infer water drop zones based on changes in altitude and speed. Each media file undergoes detailed processing to determine the capture location, the visible fire front location, and the date/time of capture. If metadata is missing, investigators estimate locations by comparing visible elements (e.g., terrain, structures) with satellite or Street View imagery. Time is cross verified by comparing sun position, shadows, and other recordings. When necessary, publishers may be contacted for confirmation. Wind data is sourced from automated weather stations or national services, and if unavailable, archived forecast data may be used. Topography data is gathered using tools like Google Earth, EO Browser, or national land survey services. Vegetation types, fuel continuity and density are inferred from satellite images and site visits. Burn scars are mapped with precision using multiple satellite products, forming the reconstruction’s base layer. These ensure an accurate depiction of fire progression and key barriers that affected its spread. Figure 1. Isochronous curves designed on reconstruction of Arakapas wildfire (Cyprus/2021). 3. Results and Discussion The Varnavas wildfire incident’s reconstruction is in progress, and it is expected to be completed by the end of July 2025. 3.1 Data Collection Information and visual evidence about the location and time that fire fronts appeared during the incident had already been collected via multiple site visits, surveys across all main X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) organization within the HDF5 file enabled seamless integration and accessibility of all components for analysis. 3. Results and Discussion Our initial exploration of the Skradin wildfire dataset highlights the power of a multi-source, spatiotemporally aligned approach in revealing key dynamics of wildfire behavior. The integration of satellite fire detections and meteorological reanalysis into a unified structure enables a more coherent understanding of how environmental factors shape fire progression. Analysis of active fire detections shows a rapid dual-direction spread from a central ignition zone. By the time of the first satellite detection at 10:00 AM on July 30th, the fire had already advanced significantly, suggesting early ignition and an intense initial burn phase. The progression of hotspots closely follows dominant wind directions retrieved from ERA5 reanalysis, underlining the strong influence of wind on fire trajectory. Topographic influence is also evident in the fire’s behavior. Initially confined to a lowerelevation basin, the fire gradually expanded upslope, with spread patterns suggesting terrain channeling and slope - driven propagation. These interactions, while often hard to isolate in disconnected datasets, are more readily discernible through integrated analysis within a common spatial and temporal frame. Fire spread can be seen on figure 1. The impact of vegetation and fuel types, represented by the downscaled FirEUrisk-based fuel map, has yet to yield clearly identifiable patterns in this early phase of analysis. Future work will focus on quantifying these relationships more rigorously, potentially using simulation or classification-based modeling techniques. This early case study already demonstrates the practical strength of combining disparate geospatial, meteorological, and thematic layers into a single, accessible HDF5 dataset. It provides not only a foundation for historical fire inspection but also foundation for simulations and scenario analysis if coupled with robust models. 4. Conclusions This extended abstract presents format and use of a unified, multi-source dataset capturing the 2024 Skradin wildfire in Dalmatia, Croatia. This fire represents a complex, multi-day event shaped by topography and meteorological conditions. By integrating satellite-based fire detections (VIIRS, MODIS), ERA5 reanalysis, high-resolution elevation data, and detailed fuel map into a cohesive HDF5 structure, this work delivers a robust foundation for spatiotemporal analysis of a wildfire. Multi facet visualization of data reveal clear wind-driven spread patterns and terrain-driven behavior, demonstrating how aligned spatiotemporal datasets can uncover environmental X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) drivers not easily observable in isolation. While fuel-type influences remain under investigation, and the coarse resolution of meteorological input presents limitations, the dataset already supports meaningful interpretation of fire-environment interactions. This work underscores the importance of merging satellite and atmospheric observations into a spatially and temporally coherent format. The HDF5-based approach not only enables efficient storage of multidimensional data but also promotes accessibility and scalability for future case studies. Next steps include enhanced visualization of wind vectors, interoperability with fire simulation software and application of this framework to additional regional fires References 1. John S Schreck et al. “Machine Learning and VIIRS Satellite Retrievals for Skillful Fuel Moisture Content Monitoring in Wildfire Management”. In: Remote Sensing 15.13 (2023), p. 3372. 2. Claudia Vitolo et al. “ERA5-based global meteorological wildfire danger maps”. In: Scientific data 7.1 (2020), 3. p. 216. 4. The HDF Group. Hierarchical Data Format, version 5 (HDF5). Available from: https://www.hdfgroup.org/solutions/hdf5/. 5. Dominik Traxl. FireTracks Scientific Dataset. Available from: https://github.com/dominiktraxl/firetracks. 2024. 6. G.R. van der Werf et al. Global Fire Emissions Database (GFED). Available from: https://www.geo.vu.nl/~gwerf/GFED/GFED4/. 2017. 7. M. Bugarić et al. “Calculation and validation new high resolution fuel map of Croatia”. In: Proceedings for the 7th International Fire Behavior and Fuels Conference. April 15-19. Boise, Idaho, USA – Tralee, Ireland – Canberra, Australia: International Association of Wildland Fire, Apr. 2024. 8. D. Stipaničev et al. FirEUrisk modeli i karte goriva za područje Republike Hrvatske. Dissemination Report. 9. FirEUrisk Dissemination Report, 2024. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Calibration of a Thermal Tunnel to Analyse the Effect of Atmospheric Structure on Fire Spread Carlos Ribeiro*,1; Domingos Viegas1; Thiago Fernandes Barbosa1; Tiago Rodrigues1 1 Univ Coimbra, ADAI, Department of Mechanical Engineering, Rua Luís Reis Santos, Pólo II, 3030788 Coimbra, Portugal, [email protected]t *Corresponding (presenting) author Summary Forest fires are influenced by complex interactions between fire and atmospheric conditions, particularly by temperature and wind velocity vertical profiles. These "fireatmosphere interactions" affect fire behaviour, especially the vertical air movement driven by heat. Atmospheric stability, characterized by vertical temperature gradient, can either suppress or intensify fire spread. To study this problem, CEIF | ADAI at the University of Coimbra developed a Thermal Tunnel capable of simulating different air temperature and velocity vertical profiles. This work focuses on calibrating the Thermal Tunnel to examine how atmospheric conditions that affect the rate of fire spread can be replicated in the laboratory. Keywords: Fire Behaviour, Forest Fires, Dynamic Fire Behaviour, Fire and Atmosphere Interaction, Fire atmosphere coupling, Physical Modelling. 1. Introduction Forest fires often exhibit complex and dynamic fire behaviour resulting from interactions between the fire and the surrounding environment. These interactions, known as "fireatmosphere interactions" or “fire atmosphere coupling,” can cause instantaneous increases or decreases in the rate of fire spread (ROS), creating critical safety issues [1,2]. The vertical structure of the atmosphere, specifically temperature and wind velocity profiles, plays a significant role in fire dynamics. The heat generated by a wildfire primarily causes vertical air movement, the intensity of which is affected by air density in the lowest layer of the atmosphere (the troposphere) [3-6]. Atmospheric stability, characterized by the vertical temperature gradient, determines whether vertical mixing is suppressed or intensified:(i) Stable Atmosphere (Temperature Inversion): Occurs when temperature increases with height. This structure resists vertical movement, inhibiting vertical mixing, and potentially trapping smoke near the surface, which may lead to higher ground-level temperatures that affect fire behaviour; (ii) Unstable Atmosphere: Occurs when X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) temperature decreases rapidly with height. This condition promotes vertical mixing, allowing hot gases to rise more easily, which can lead to rapid plume development, increased fire intensity, and erratic fire behaviour (such as that observed in the large Portuguese wildfires of 2017) [7, 8]. To isolate and analyse these specific effects, experimental studies in controlled environments like thermal tunnels are essential. This work focuses on calibrating the Thermal Tunnel created by CEIF | ADAI to investigate the properties of the flow inside its test chamber and analyse how atmospheric conditions affecting the rate of fire spread can be reliably replicated in a laboratory setting. 2. Data and Methods 2.1 Thermal Tunnel The Thermal Tunnel is located at the Forest Fire Research Laboratory of the University of Coimbra in Lousã, Portugal, see Figure 10. The tunnel has a working area of 8x2m2 with a cross-section of 1.5x2m2 and is open on the top to avoid flow stratification and smoke accumulation in the combustion area. The tunnel is open on the top to prevent flow stratification and smoke accumulation in the combustion area. Tempered glass side walls facilitate observation within the test chamber. the wind flow is produced by three independent channels or ducts of rectangular cross-section. The flow velocity in each channel can be controlled independently. a) b) Figure 10: Thermal Tunnel of the Forest Fire Research Laboratory of the University of Coimbra in Lousã (Portugal). a) Schematic view; b) View of the combustion thermal tunnel in the laboratory. Two high-power heat pumps produce water (ranging between 6 °C and 45 °C) to either heat or cool the air in the upper and lower channels. The mid channel works with air at ambient U1, T1 U2, T2 U3, T3 Túnel Térmico Cofinanciado por: CENTRO-01-0246-FEDER-000038 Apoiado por: Fabricado por: Novembro 2022 X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) temperature. This design allows the facility to blow air in three distinct layers at different temperatures and velocities, crucial for simulating various atmospheric stability regimes. 2.2 Air velocity and temperature measurements The calibration aimed to investigate the flow properties within the test chamber. The HT-400 series system, including a hot-sphere-type thermal anemometer probe (HT-41x), a transducer unit (HT-42x), and a multichannel power supply (HT-430), was used for air velocity and temperature measurements. The probe includes an omnidirectional velocity sensor (enamelled copper wire, 3mm sphere) and a temperature sensor (thin nickel wire), both covered with a special AL coating to reduce the effect of thermal radiation Figure 11. During the calibration of the Thermal Tunnel, 10 sensors (𝑆𝑖 where 𝑖 range between 1 to 10) were used to measure the air flow velocity and temperature. For each air duct 3 sensors (𝑆1=(0,100,125)𝑐𝑚, 𝑆2=(0,100,75)𝑐𝑚, and 𝑆3=(0,100,25)𝑐𝑚) were fixed in their geometric centre. The other seven sensors (𝑆4=(𝑑𝑖𝑠,50,125)𝑐𝑚, 𝑆5=(𝑑𝑖𝑠,50,25)𝑐𝑚, 𝑆6=(𝑑𝑖𝑠,100,125)𝑐𝑚, 𝑆7=(𝑑𝑖𝑠,100,75)𝑐𝑚, 𝑆8=(𝑑𝑖𝑠,100,25)𝑐𝑚, 𝑆9= (𝑑𝑖𝑠,150,125)𝑐𝑚 and 𝑆10 =(𝑑𝑖𝑠,150,25)𝑐𝑚) were placed on a structure to calibrate the thermal tunnel at various distances. Thus, the following distances were considered 𝑑𝑖𝑠= 100,300,500 and 700cm. For each distance 4 different flow velocities were considered, and the potential difference was 1.5V, 2V, 3V and 4V. The Figure 3 represent the schematic view for the position of each sensor. Figure 11: The probe is schematically shown. The dimensions of the velocity sensor and the temperature sensor is presented in this figure. Figure 12: Schematic view of the sensors to measure the air flow velocity and temperature. 3. Results and Discussion X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) 3.1 Air flow rate as a function of potential difference The calibration results characterized the tunnel’s capabilities regarding airflow control linearity, sensitivity, flow uniformity, and the successful generation of stability regimes. The analysis evaluated the linearity and sensitivity of the fan systems (UTA1, UTA2, UTA3) in response to changes in input voltage, Figure 13. All three datasets demonstrated a strong linear correlation between voltage and airflow rate, as indicated by high coefficients of determination (𝑅2>0.99). Figure 13: Relationship between airflow rate (Q, in m³/s) and the potential difference (DV, in Volts) for three different air ducts under test (UTA1, UTA2, and UTA3) within the Thermal Tunnel. 3.2 Wind Flow Velocity Profiles Understanding the wind flow velocity profile is essential for characterizing the environment’s impact on fire behaviour. Figure 14 presents the wind flow velocity profiles for four different set velocities, measured at various distances (𝑑𝑖𝑠=1,3,5 and 7m) along the tunnel. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Figure 14: Wind Flow Velocity Profiles. The profiles were measured at various distances (1m, 3m, 5m, and 7m). While the profiles change slightly as the flow moves down the tunnel due to factors like wall friction and turbulence, the uniformity of the flow is essentially maintained across the entire test chamber. The tunnel demonstrated its ability to generate a range of flow conditions with great uniformity, as the flow velocity profile tends to remain almost constant throughout the test chamber. 3.3 Vertical Temperature Profiles The capability to establish controlled atmospheric conditions is crucial for accurately assessing the impact of atmospheric structure on the fire rate of spread. The Thermal Tunnel successfully achieved three distinct stability regimes: Figure 15a Ambient Temperature Profile (Neutral Stability): Represents a baseline condition where the temperature is relatively uniform throughout the height of the tunnel. This simulates a neutrally stable atmosphere where vertical mixing is neither enhanced nor suppressed, serving as a control for comparison; Figure 15b Temperature Inversion (Stable Atmosphere): Established by having temperature increase with height. This simulates a stable atmospheric condition that inhibits vertical mixing; and Figure 15c Unstable Atmosphere: Established by having temperature decrease with height. This promotes vertical mixing, potentially leading to rapid plume development and increased fire intensity. These analyses demonstrate the Thermal Tunnel's capability of generating a range of atmospheric stability conditions. By systematically varying the temperature profiles, it becomes possible to investigate the specific influence of atmospheric stability on various aspects of fire behaviour. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) a) b) c) Figure 15: Vertical Temperature Profiles: a) Ambient Temperature Profile; b) Temperature Inversion and c) Unstable Atmosphere. 4. Conclusions This study confirmed the thermal tunnel's suitability for conducting controlled experiments on fire behaviour under varying atmospheric conditions. The calibration successfully demonstrated the tunnel's ability to generate controlled and repeatable vertical temperature profiles, simulating stable, neutral, and unstable atmospheric conditions. Furthermore, the characterization of the fan systems showed excellent linearity and provided a comprehensive understanding of the uniform flow dynamics within the tunnel. By allowing researchers to manipulate both temperature and wind velocity profiles, the Thermal Tunnel enables the isolation and analysis of specific fire-atmosphere interactions, addressing a critical need in fire research. Future work will utilize this calibrated environment to investigate the rate of spread of head fires under stable and unstable atmospheric conditions to contribute to improved fire behavior prediction and mitigation strategies. Acknowledgements This research was funded in part by the Fundação para a Ciência e a Tecnologia, I.P. (FCT, https://ror.org/00snfqn5816) under Grant UIDB/50022/2020 (https://doi.org/10.54499/UIDB/50022/2020); UIDP/50022/2020 (https://doi.org/10.54499/UIDP/50022/2020); LA/P/0079/2020 (https://doi.org/10.54499/LA/P/0079/2020). For the purpose of Open Access, the author has applied a CC-BY public copyright license to any Author's Accepted Manuscript (AAM) version arising from this submission. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) This research work was funded and carried out within the scope of the following project Large Fires funded by FCT — Foundation for Science and Technology with reference COMPETE2030-FEDER-00878700. References 1. Viegas, X., Raposo, J., Ribeiro, C., Reis, L., Abouali, A., & Viegas, C. (2021). On the nonmonotonic behaviour of fire spread. International Journal of Wildland Fire, 30(9), 702– 719. https://doi.org/10.1071/WF21016 2. Viegas, X., Raposo, J., Ribeiro, C., Reis, L., Abouali, A., Ribeiro, L. M., & Viegas, C. (2022). On the intermittent nature of forest fire spread – Part 2. International Journal of Wildland Fire, 31(10), 967–981. https://doi.org/10.1071/WF21098 3. Wagner, C. E. van. (1977). Conditions for the start and spread of crown fire. Canadian Journal of Forest Research, 7(1), 23–34. https://doi.org/10.1139/x77-004 4. Sharples, J. J., McRae, R. H. D., & Wilkes, S. R. (2012). Wind - terrain effects on the propagation of wildfires in rugged terrain: fire channelling. International Journal of Wildland Fire, 21(3), 282. https://doi.org/10.1071/WF10055 5. Lareau, N. P., & Clements, C. B. (2015). Cold Smoke: smoke-induced density currents cause unexpected smoke transport near large wildfires. Atmospheric Chemistry and Physics, 15(20), 11513–11520. https://doi.org/10.5194/acp-15-11513-2015 6. Potter, B. E. (2012). Atmospheric interactions with wildland fire behaviour - I. Basic surface interactions, vertical profiles and synoptic structures. International Journal of Wildland Fire, 21(7), 779. https://doi.org/10.1071/WF11128 7. Rosenberg, D., Pouquet, A., & Marino, R. (2021). Correlation between Buoyancy Flux, Dissipation and Potential Vorticity in Rotating Stratified Turbulence. Atmosphere, 12(2), 157. https://doi.org/10.3390/atmos12020157 8. Heilman, W. E. (2023). Atmospheric turbulence and wildland fires: a review. International Journal of Wildland Fire, 32(4), 476–495. https://doi.org/10.1071/WF22053 X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Session 3 Drivers, Trends, and Future Projections of Extreme Wildfire Events X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) 2. Data and Methods We extract fire observations from the satellite-derived fire event dataset FRY_v2.0_FireCCI51_6D [1]. This dataset aggregates fire patches based on burned area and active fire detections, including start date, size, and shape information for each fire patch. To reduce the anthropogenic influence of agricultural fires, we filter these fire events based on land cover information to focus on wildfires occurring in near natural vegetation. A fire polygon should have least 80% of natural vegetation coverage to be considered in our analysis. It should be noted that prescribed burning, if occurred in natural vegetation we define, may still be included this analysis. For climate data, we use: (1) a year-round daily weather regime classification scheme in the European-Atlantic sector [2–4]; (2) daily outputs of temperature, relative humidity and wind speed from the climate reanalysis CERRA [5]. To detect fire synchronicity, we extract daily fire time series from 2001 to 2020 for each of the ten European regions and apply a statistical framework of event synchronicity [6] to identify significant fire synchronicity between these regions in each season. To link synchronous fire events to weather regimes, we calculate the relative conditional probability [7] of synchronous wildfires co-occurring with weather regimes in spring, summer and fall. We further analyze the relationship between the spatial scale of synchronous wildfires and weather regimes by defining five levels of synchronicity – no, regional (wildfire(s) occurred within a region), low (2-3 regions co-experienced wildfires), medium (4-5 regions co-experienced wildfires) and high synchronicity (>5 regions coexperienced wildfires). 3. Results and Discussion We detect significant fire synchronicity between multiple European regions with seasonal variations. Regions with overlapping fire seasons are more likely to co-experience fires within these seasons. For example, regions with spring fires (British Isles, Middle Europe, and Northeastern Europe) show significant fire synchronicity in spring, whereas southern regions with summer fire activity show significant fire synchronicity in summer. These findings indicate that European regions with strongly overlapping fire seasons have reduced opportunities to share firefighting resources in case of synchronous wildfire emergency. Further, we compare atmospheric conditions on regional fire days versus synchronous fire days and find that these two types of days are featured by warm and dry weather conditions. Synchronous fire days co-occur with even warmer and drier conditions than regional fire days. The frequency of compound warm and dry events is likely to increase in coming decades, suggesting higher synchronous fire danger under climate warming. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Figure 1. A schematic overview of two driving processes of synchronous wildfires. Synchronous wildfires in Europe are significantly modulated by weather regimes. Weather regimes featured by a high-pressure center above the European continent, such as European Blocking and Scandinavian Blocking, increase the likelihood of synchronous wildfires by driving warm and dry weather conditions. Further, extreme synchronous wildfires with more than five regions involved are the most likely to co-occur with European Blocking. Conversely, when a low-pressure center develops above the continent, synchronous wildfires are suppressed. As weather regimes are predictable on a subseasonal timescale, these findings may help to develop a synchronous wildfire early warning system for Europe. Conclusions We detect seasonally varying patterns of significant fire synchronicity between different European regions. We reveal that fire seasonality and persistent weather regimes coregulate the occurrence of synchronous wildfires. Compound warm and dry weather conditions co-occurring in several regions have the potential to trigger synchronous wildfires at regional to continental scales. European blocking can trigger synchronous wildfires over large spatial extents by promoting warm and dry conditions. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) References 1. Laurent, P., Mouillot, F., Yue, C., Ciais, P., Moreno, M.V., Nogueira, J.M.P. FRY, a global database of fire patch functional traits derived from space-borne burned area products. Sci Data 2018, 5, 180132. 2. Grams, C.M., Beerli, R., Pfenninger, S., Staffell, I., Wernli, H. Balancing Europe’s windpower output through spatial deployment informed by weather regimes. Nature Clim Change 2017, 7, 557–562. 3. Büeler, D., Ferranti, L., Magnusson, L., Quinting, J.F., Grams, C.M. Year‐round sub‐ seasonal forecast skill for Atlantic–European weather regimes. Quart J Royal Meteoro Soc 2021, 147, 4283–4309. 4. Hauser, S., Teubler, F., Riemer, M., Knippertz, P., Grams, C.M. Life cycle dynamics of Greenland blocking from a potential vorticity perspective. Weather Clim. Dynam. 2024, 5, 633–658. 5. Ridal, M., Bazile, E., Le Moigne, P., Randriamampianina, R., Schimanke, S., Andrae, U., et al. CERRA, the Copernicus European Regional Reanalysis system. Quart J Royal Meteoro Soc 2024, 150, 3385–3411. 6. Boers, N., Goswami, B., Rheinwalt, A., Bookhagen, B., Hoskins, B., Kurths, J. Complex networks reveal global pattern of extreme-rainfall teleconnections. Nature 2019, 566, 373–377. 7. Brunner, L., Schaller, N., Anstey, J., Sillmann, J., Steiner, A.K. Dependence of Present and Future European Temperature Extremes on the Location of Atmospheric Blocking. Geophysical Research Letters 2018, 45, 6311–6320. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Increasing likelihood of fire-conductive conditions in a warming Europe Julia Miller*,123, Xinhang Li123, Danielle Touma4, Manuela I. Brunner123 1Institute for Atmospheric and Climate Science, ETH Zurich, 8006 Zurich, Switzerland 2WSL Institute for Snow and Avalanche Research SLF, 7260 Davos Dorf, Switzerland1 3Climate Change, Extremes and Natural Hazards in Alpine Regions Research Center CERC 4Jackson School of Geosciences, Institute for Geophysics, University of Austin, 78758 Austin TX, United States. *Corresponding (presenting) author *[email protected] Summary Recent extreme wildfire seasons in Europe, particularly in 2017 and 2018, were driven by compounding fuel, drought, and fire weather conditions. By using the CESM2-Large Ensemble, we assess how the likelihood of such wildfire months—characterized by FWI, SPEI, and GPP—has changed from pre-industrial times to present climate and future warming scenarios. Fire-promoting conditions become substantially more likely in the future, with return periods of observed events dropping below three years in Southern Europe and reaching decadal levels in Central and Northern Europe. These results highlight increasing wildfire risks and the need for coordinated climate-adaptive fire management across Europe. Keywords: climate change, wildfires, compound events 1. Introduction In recent years, various regions in Europe experienced unprecedented wildfire extremes, including the Iberian Peninsula and the Mediterranean in 2017 and Central Europe, the British Isles, and Scandinavia in 2018. These wildfire extremes consisted of multiple large wildfires that occurred in close temporal proximity in the case study regions and led to the most severe months ever recorded in terms of burned area and damage to ecosystems, livelihoods, and infrastructure. All case study months are characterized by pronounced drought conditions, which promote fuel aridity and high fire danger levels. While climate change likely promoted the extreme conditions that led to the extreme fire events in the Iberian Peninsula in October 2017, the Mediterranean in July 2017, the British Isles and Scandinavia in July 2018, and Central Europe in September 2018, the degree to which drought conditions, fire weather, and fuel availability have changed compared to preindustrial levels is not yet fully understood. Here, we use the Community Earth System X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Model version 2 Large Ensemble (CESM2-LE)2 to investigate how the likelihood of extreme wildfires and their drivers in different regions of Europe compare to pre-industrial levels and explore how these likelihoods change in a Europe that is 2° or 3° warmer. 2. Data and Methods We first identify case studies of extreme wildfire seasons between 2001 and 2020 in different climate regions of Europe, i. e. Iberian Peninsula, Mediterranean, British Isles, Scandinavia and Central Europe. Extreme wildfire seasons are defined as the top three years with the largest burned area observed between June and November in each region. Within these extreme seasons, we identify the month with the largest burned area (derived from FireCCI1) and evaluate how extreme these months were in terms of burned area (derived from FireCCI1), FWI and SPEI-3M (derived from CERRA reanalysis3) and GPP (derived from MODIS4). To illustrate the likelihood of these events in observed climate, we fit a GEV distribution to burned area and FWI and a log Gamma distribution to SPEI and GPP summer and fall monthly mean values in the 20-year observation period. Next, we map the likelihoods of the observed extremes to respective likelihoods in pre-industrial climate, present climate (2001-2020) and future climate, i. e. 2°- and 3°-degree warming levels for Europe, which we derive from the CESM2-LE. We map the likelihoods of the observed events in terms of FWI, SPEI and GPP to assess how the preconditions of extreme wildfire months change across climates. Burned area information form CESM2-LE does not compare consistently across seasons and regions with observational records and is therefore omitted from the climate change analysis. 3. Results and Discussion We identified 2017 and 2018 as the most severe wildfire seasons in multiple regions. For example, 2017 was the worst fire season in terms of burned area in the Iberian Peninsula and the third worst in the Mediterranean. Meanwhile, 2018 was the worst fire season in terms of burned area in the British Isles, Scandinavia, and Central Europe. Within these seasons, the peak monthly burned area was observed in October 2017 in the Iberian Peninsula and in August 2017 in the Mediterranean. In 2018, July was the worst month in the British Isles and Scandinavia, and September was the worst month in Central Europe. The wildfires in the British Isles and Scandinavia (Northern Europe) in 2018 co-occurred with the highest observed FWI and the strongest observed drought (SPEI) indicator values. In these regions, GPP was higher than average. In Central Europe, FWI conditions were less extreme, while GPP conditions were below average under severe drought conditions. The FWI conditions were average for the Iberian Peninsula wildfires in October 2017. In the Mediterranean, the FWI conditions were among the five highest observed values in 2017. In both regions (further Southern Europe), drought conditions in the SPEI represent moderate X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) drought, while GPP is lower than average in the Iberian Peninsula and average in the Mediterranean. We map the probabilities of these conditions under pre-industrial, present, 2, and 3-degree warmer climate conditions within the CESM2-LE and find that, in comparison to the preindustrial climate, the observed FWI and SPEI extremes became at least twice as likely in all regions under present climate conditions. Increasing warming will make FWI extremes, such as those observed during our identified extreme wildfire months, four times more likely in a 2° warmer Europe and at least five times more likely in a 3° warmer climate. SPEI extremes are even more likely to occur. For example, in Southern Europe, SPEI conditions will be three times more likely under 2°C warming and five times more likely under 3°C warming. In Northern and Central Europe, extreme drought conditions will be ten times more likely under 2°C warming and more than fifteen times more likely under 3°C warming. Our findings imply that the preconditions for wildfires will worsen significantly across different European climate regions. The likelihood of such wildfire-promoting extreme conditions will rise especially in Central and Northern Europe, suggesting that these regions will become more fire-prone in the future. Though these extremes will increase significantly in Northern and Central Europe, their return levels will remain above a 10-year return period under different warming levels. In the Mediterranean, however, the conditions of the observed extreme months will become much more frequent, i.e., occur every two to three years. 4. Conclusions Our study shows that the likelihood of preconditions, i.e., FWI and SPEI-3M, for severe wildfire months increases drastically under climate change conditions in all European regions. As extreme drought and SPEI conditions approach annual and decadal return levels, respectively, in Southern, and Northern and Central Europe, it is crucial to emphasize the need for Europe-wide wildfire management and preparedness that addresses the pressure that climate change puts on fire-prone regions and those that will become fireprone in the future. References 1. Lizundia-Loiola, J., Otón, G., Ramo, R. & Chuvieco, E. A spatio-temporal active-fire clustering approach for global burned area mapping at 250 m from MODIS data. Remote Sens. Environ. 236, 111493 (2020). 2. Danabasoglu, G. et al. The Community Earth System Model Version 2 (CESM2). J. Adv. Model. Earth Syst. 12, e2019MS001916 (2020). 3. Schimanke, S. et al. CERRA sub-daily regional reanalysis data for Europe on single levels from 1984 to present. https://doi.org/10.24381/cds.622a565a (2021). X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) 4. Running, S., Mu, Q. & Zhao, M. MOD17A2H MODIS/Terra Gross Primary Productivity 8Day L4 Global 500m SIN Grid V006. NASA EOSDIS Land Processes Distributed Active Archive Center https://doi.org/10.5067/MODIS/MOD17A2H.006 (2015). X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) Eruptive Fire Early Warning System Darko Stipaničev1,*; Marin Bugarić2; Ljiljana Šerić3; Damir Krstinić4 1 FESB University of Split, Croatia, [email protected] 2 FESB University of Split, Croatia, marin.buga[email protected] 3 FESB University of Split, Croatia, ljiljana.s[email protected] 4 FESB University of Split, Croatia, [email protected] *Corresponding (presenting) author Summary One of the most dangerous forms of extreme wildfire behavior is undoubtedly eruptive fire spread. Eruptive fire spread is a frequent cause of wildfire accidents in which numerous firefighters have been injured or killed. For this reason, we believe that prediction of possible eruptive fire spread should be included in wildfire risk assessments as a specific indicator of propagation potential, referred to as the Potential Eruptive Fire Spread Indicator. The system described in this paper was implemented successfully in Croatian advanced wildfire surveillance system used by Croatian firefighters in daily practice enhancing the firefighter’s safety. Keywords: eruptive fire, early warning, wildfire surveillance 1. Introduction One of the most dangerous forms of extreme wildfire behavior is undoubtedly eruptive fire spread. Fire does not spread uniformly as assumed in the essential Rothermel model, but rather accelerates - this acceleration depends on wind speed and terrain slope. This phenomenon is known as Eruptive Fire or Blow-Up Fire. Eruptive fire spread is a frequent cause of wildfire accidents in which numerous firefighters have been injured or killed. The worst firefighting tragedy in Croatia occurred in 2007 on the island of Kornat, in the Šipnate canyon, where 12 firefighters lost their lives. According to many indicators, the tragedy was triggered by eruptive fire spread [1]. For this reason, we believe that prediction of possible eruptive fire spread should be included in wildfire risk assessments as a specific indicator of propagation potential, referred to as the Potential Eruptive Fire Spread Indicator. In Croatia, firefighters use the advanced STRIBOR OiV Fire Detect AI wildfire surveillance system [2] in daily operations. In addition to early fire detection, the system integrates various tools that support firefighting efforts, such as micro-location fire risk prediction and simulation of potential fire spread from a specific ignition point [3]. One of the embedded tools is the prediction of potential X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) eruptive fire spread included as an Eruptive Fire Early Warning System, which is the focus of this paper. In the following sections, we will describe the methodology for determining this indicator, based on mathematical models of eruptive fire behavior, along with the results obtained. We hope that this contribution will help improve firefighter safety in Croatia. Index is calculated dynamically based on actual weather data, as well as on weather prediction for the next 12 hours using Croatian official weather prediction model ALADIN. Therefore, when the firefighter clicks on map showing Potential Eruptive Fire Spread Indicator two times a day (12:00 and 24:00) it shows indicator based on real weather data and between them prediction based on ALADIN model in time resolution of 2 hours. 2. Related Work Any wildfire could experience accelerated, eruptive spread if conditions related to slope and terrain shape, as well as wind direction and speed, are met. Therefore, it may be more accurate to use the term eruptive fire spread effect. Rothermel’s equation assumes that a grassland fire reaching a slope of 14%, with wind blowing upslope (in the direction of fire spread) at a speed of 5.5 m/s at 10 meters height, will spread upslope at approximately 20 meters per minute. According to Rothermel, the speed would remain the same at both the bottom and top of the slope, regardless of its length. In other words, the rate of spread determined by the equation depends solely on three factors: terrain slope, vegetation, and meteorological conditions. The eruptive effect is essentially an extension of Rothermel’s base fire spread model, incorporating the effect of fire acceleration when terrain and wind conditions are favorable. In the previously mentioned example, when the fire reaches a sloped incline, its speed will not be the same at the beginning and end of the slope. As it moves uphill, the speed will accelerate and can become ten times greater at the top compared to the base. This behavior has been observed in both experiments and real fires, making the eruptive effect a more accurate representation of how fires spread. According to the FirEUrisk Extreme Fire Handbook [4], eruptive fire behavior is associated with a very rapid acceleration of the fire, which can be compared to an eruption - hence the name. The destructive force of such eruptions is significant and can catch people by surprise. Most fatal wildfire accidents are linked to this type of fire behavior. While closely associated with canyons and steep slopes, the underlying physical processes can also occur in more typical situations. In the literature, such events are sometimes referred to as “blow-up fires”, although the term “explosive” can be misleading, as there is no actual detonation. The increase in spread rate is a continuous process, not a sudden flare-up, which is why the term eruptive or accelerated fire spread is more appropriate. X-Fire 2025 –– Book of Abstracts | COST Action NERO (CA22164) This type of extreme fire behavior was first systematically studied by Viegas et al. in 2002 [5], who found that eruptive behavior is linked to fire-induced airflow, amplified by the concave shape of canyon terrain. In laboratory experiments, a tabletop canyon model composed of two sloped planes was used to systematically analyze this behavior. The shape of the canyon was defined by the angles of inclination of the two slopes and the overall base angle. Higher values of these angles result in greater fire acceleration. There are two types of eruptive fire behavior: slope-driven eruptive fire and wind-driven eruptive fire. Since slope-driven eruptive fire occurs more frequently, the term eruptive fire is commonly used to refer specifically to this type of accelerated fire spread caused by terrain slope. However, it is important to note that the eruptive effect can also be triggered by wind. Viegas [6] demonstrated this duality and the possibility of treating the influence of both wind and slope in a unified manner when assessing the potential for eruptive fire development. The fundamental equation of eruptive fire spread, as proposed by Viegas and his co-authors [6,7,8], is a semi-empirical equation that relates the increase in the rate of fire front spread to the speed at which the fire would spread if the eruptive fire effect were not present. It should be noted from the outset that these equations were derived for fire spread on inclined planes, not for canyon-shaped terrain. In canyons, all these effects would be even more pronounced and the spread rates even higher - something that has also been confirmed experimentally [5,6]. 3. Data, Methods, Results and Discussion Our research on the Potential Eruptive Fire Spread Indicator starts during the EU IPA Adriatic Holistic project [9]. Here, we present an improved version developed during work on H2020 FirEUrisk project [10]. The results are mostly based on the work of Viegas from 2004 [6] and the study by Charleton et al. from 2015 [11]. The main difference compared to the earlier version is the introduction of one additional class (see Figure 1). In this improved version, we propose a total of five classes: Slope-driven Eruptive Fire: • Class I – Potential for eruptive fire spread caused by terrain slope. Class I is for those regions where slope S ≥ 250, regardless of wind. Based on [11, p.1356] and [4, Fig.10]. Windand Slope-driven Eruptive Fire: • Class II – Potential for eruptive fire spread caused by wind and slope. Class II is for those regions where corrected midflame wind speed considering slope and aspect is MFWS_S_A ≥ 2 m/s [4].