Fire Behavior Prediction Research Articles

Vegetation phenology as a key driver for fire occurrence in the UK and comparable humid temperate regions

Background Fire activity in the UK and comparable regions of northwest Europe is generally out of phase with peak fire weather conditions. Aims Here, we assess the potential effect of phenology on fire occurrence patterns for the UK. Methods We examined fire occurrence and vegetation phenology in the UK for 2012–2023, mapped onto the main fire-affected vegetation cover types within distinct precipitation regions, allowing the fire occurrence for fuels in different phenological phases to be explored across distinct ‘fuel’ types and regions. Key results The UK’s fire regime is characterised by burning in semi-natural grasslands and evergreen dwarf shrub ecosystems in early spring when vegetation is still dormant. During the high-greenness phase in late spring and summer, fire activity is reduced by a factor of 5–6 despite typically elevated fire weather conditions within that period. Conclusions and implications Semi-natural vegetation in the UK is very resistant to burning during the high-greenness phase. However, this ‘fire barrier’ is diminished during severe drought episodes, which are predicted to become more extreme in the coming decades. Incorporating phenology information into models therefore has great potential for improving future fire danger and behaviour predictions in the UK and comparable humid temperate regions.

A physics-based model of thermodynamically varying fuel moisture content for fire behavior prediction

Fuel moisture content (FMC) is a critical parameter in fire and plume behaviors, showing diurnal and spatial variations influenced by local meteorological conditions, soil characteristics, and fuel properties. In low-intensity fires, small-scale FMC variations intensify, leading to an amplification of their effects on fire physics. In an effort to capture these variations, this paper presents the development of a physics-based model that couples a thermodynamic-based FMC prediction model for dead fuels with the Fire Dynamics Simulator of the National Institute of Standards and Technology. The model accuracy is validated against several existing experimental data, showing improvements over the baseline model which uses the kinetic-based Arrhenius drying approach. A case study of flame propagation in a small fuel bed is also presented, indicating the improved performance of the new model and its novel capabilities in capturing complex processes of fuel drying and moisture flux exchanges between the fuel and ambient atmosphere.

A perspective and survey on the implementation and uptake of tools to support decision-making in Canadian wildland fire management

The level of implementation and uptake of specific tools used to support wildland fire management decision-making has received little attention in Canada. Our aim is to aid the fire research-to-practice discourse in Canada by describing key terms and concepts for characterizing implementation, uptake, and capacity. We also designed and conducted a survey to assess the implementation and uptake of some of the available tools used by Canadian provincial and territorial fire management agencies. We assessed nine tools and found distinct differences in their implementation and uptake, with differing results at national versus provincial and territorial scale. The Canadian Fire Weather Index and Fire Behaviour Prediction Systems had the highest level of both implementation and uptake nationally. The other tools have substantially lower but varying degrees of implementation and uptake across the country. The results encourage further investigation into the factors affecting implementation and uptake of fire management tools, both nationally and in provinces and territories.

Fire behavior and post-fire residual tensile strength prediction of carbon fiber/phthalonitrile composite laminates

Carbon fiber/phthalonitrile (Cf/PN) composite has a high potential for applications in the aerospace sector because of the outstanding heat and flame resistance of phthalonitrile resin. Here we aim to evaluate the fire behavior of Cf/PN laminate under localized one-sided flame heating as well as the residual tensile performance. The residual tensile strength after quasi-isothermal pyrolysis in an inert environment decreases linearly with the increase in the pyrolysis degree. The temperature response and pyrolysis behavior are analyzed through a combination of experiments and finite element simulations, and the damage mode of the laminate structure is concluded through morphological observation after flame exposure and the surface axial strain field using digital image correlation. Eventually, the residual tensile strength of laminates with different thicknesses after different times of flame exposure was effectively predicted based on the numerical simulated pyrolysis degree distribution. This research supplies fundamental test data on the mechanical properties of Cf/PN laminates and is expected to provide guidelines for the engineering application of Cf/PN composites in thermal protection/load-bearing all-in-one structures.

Incorporating burn heterogeneity with fuel load estimates may improve fire behaviour predictions in south-east Australian eucalypt forest

Background Simulations of fire spread are vital for operational fire management and strategic risk planning. Aims To quantify burn heterogeneity effects on post-fire fuel loads, and test whether modifying fuel load estimates based on the fire severity and patchiness of the last fire improves the accuracy of simulations of subsequent fires. Methods We (1) measured fine fuels in eucalypt forests in south-eastern Australia following fires of differing severity; (2) modified post-fire fuel accumulation estimates based on our results; and (3) ran different fire simulations for a case-study area which was subject to a planned hazard reduction burn followed by a wildfire shortly thereafter. Key results Increasing fire severity resulted in increased reduction in bark fuels. In contrast, surface and elevated fuels were reduced by similar amounts following both low-moderate and high-extreme fire severity. Accounting for burn heterogeneity, and fire severity effects on bark, improved the accuracy of fire spread for a case study fire. Conclusions Integration of burn heterogeneity into post-burn fuel load estimates may substantially improve fire behaviour predictions. Implications Without accounting for burn heterogeneity, patchy burns of low severity may mean that risk estimations are incorrect. This has implications for evaluating the cost-effectiveness of planned burn programmes.

A hierarchical, multi‐sensor framework for peatland sub‐class and vegetation mapping throughout the Canadian boreal forest

AbstractPeatlands in the Canadian boreal forest are being negatively impacted by anthropogenic climate change, the effects of which are expected to worsen. Peatland types and sub‐classes vary in their ecohydrological characteristics and are expected to have different responses to climate change. Large‐scale modelling frameworks such as the Canadian Model for Peatlands, the Canadian Fire Behaviour Prediction System and the Canadian Land Data Assimilation System require peatland maps including information on sub‐types and vegetation as critical inputs. Additionally, peatland class and vegetation height are critical variables for wildlife habitat management and are related to the carbon cycle and wildfire fuel loading. This research aimed to create a map of peatland sub‐classes (bog, poor fen, rich fen permafrost peat complex) for the Canadian boreal forest and create an inventory of peatland vegetation height characteristics using ICESat‐2. A three‐stage hierarchical classification framework was developed to map peatland sub‐classes within the Canadian boreal forest circa 2020. Training and validation data consisted of peatland locations derived from various sources (field data, aerial photo interpretation, measurements documented in literature). A combination of multispectral data, L‐band SAR backscatter and C‐Band interferometric SAR coherence, forest structure and ancillary variables was used as model predictors. Ancillary data were used to mask agricultural areas and urban regions and account for regions that may exhibit permafrost. In the first stage of the classification, wetlands, uplands and water were classified with 86.5% accuracy. In the second stage, within the wetland areas only, peatland and mineral wetlands were differentiated with 93.3% accuracy. In the third stage, constrained to only the peatland areas, bogs, rich fens, poor fens and permafrost peat complexes were classified with 71.5% accuracy. Then, ICESat‐2 ATL08 spaceborne lidar data were used to describe regional variations in peatland vegetation height characteristics and regional and class‐wise variations based on a boreal forest wide sample. This research introduced a comprehensive large‐scale peatland sub‐class mapping framework for the Canadian boreal forest, presenting the first moderate resolution map of its kind.

Assessing Chilgoza Pine (Pinus gerardiana) forest fire severity: Remote sensing analysis, correlations, and predictive modeling for enhanced management strategies

Forest fires represent a critical global threat to both humans and ecosystems. This study examines the intensity and impacts of Chilgoza (Pinus gerardiana) Pine Forest fires by using advanced remote sensing techniques comprising Normalized Burn Ratio (NBR) and Difference Normalized Burn Ratio (dNBR) analyses based on Landsat 9 datasets. The study highlights the severe effect of these fires, resulting in noteworthy losses of livestock and private properties and widespread damage to 10,156.53 acres of the Chilgoza Pine Forest. A comprehensive variable correlation analysis is conducted to gain deeper insights into the influencing factors causing forest fires. Spearman's Rank Correlation Coefficient was used to assess the association between burnt and unburnt areas and various independent factors. The analysis reveals compelling evidence of significant correlations with forest fire prevalence. This study found moderate negative (-0.532, p < 0.05) and positive (0.513, p < 0.05) correlations with elevation and Land Surface Temperature (LST), respectively, and a weak positive correlation (0.252, p < 0.05) with a Wind Speed (V). To predict forest fire susceptibility and better understand the contributing factors, three machine learning models, Random Forest (RF), XGBoost, and logistic regression, are applied to assess variable importance scores. Among the considered factors, LST is the most critical variable, with consistently high variable importance scores (100 %, 96 %, and 59 %) across all three models. Wind Speed (V) also proved influential in all models, with variable importance scores of 78 %, 83 %, and 61 % for RF, XGBoost, and logistic regression, respectively. Moreover, elevation significantly influences the frequency of forest fires, as evidenced by variable importance scores ranging from 26 % to 100 %. Comparatively, the Random Forest model outperforms XGBoost and Logistic Regression in predicting forest fire vulnerability. During the training stage, the Random Forest (RF) model achieves an impressive classification accuracy of 99.1 %, followed by XGBoost with 94.5 % and Logistic Regression with 85.6 %. On evaluation with the validation dataset, the accuracies remain promising, with RF at 96.4 %, XGBoost at 91.1 %, and Logistic Regression at 84.6 %. Based on the Random Forest model, the identified high-risk sites offer valuable insights for proactive fire management and prevention strategies. This study provides a robust predictive model and a comprehensive understanding of forest fire severity and impacts. Future research should consider climate change scenarios and account for human activities to enhance fire behavior predictions and risk assessment models.

Fuel types misrepresent forest structure and composition in interior British Columbia: a way forward

BackgroundA clear understanding of the connectivity, structure, and composition of wildland fuels is essential for effective wildfire management. However, fuel typing and mapping are challenging owing to a broad diversity of fuel conditions and their spatial and temporal heterogeneity. In Canada, fuel types and potential fire behavior are characterized using the Fire Behavior Prediction (FBP) System, which uses an association approach to categorize vegetation into 16 fuel types based on stand structure and composition. In British Columbia (BC), provincial and national FBP System fuel type maps are derived from remotely sensed forest inventory data and are widely used for wildfire operations, fuel management, and scientific research. Despite their widespread usage, the accuracy and applicability of these fuel type maps have not been formally assessed. To address this knowledge gap, we quantified the agreement between on-site assessments and provincial and national fuel type maps in interior BC.ResultsWe consistently found poor correspondence between field assessment data and both provincial and national fuel types. Mismatches were particularly frequent for (i) dry interior ecosystems, (ii) mixedwood and deciduous fuel types, and (iii) post-harvesting conditions. For 58% of field plots, there was no suitable match to the extant fuel structure and composition. Mismatches were driven by the accuracy and availability of forest inventory data and low applicability of the Canadian FBP System to interior BC fuels.ConclusionsThe fuel typing mismatches we identified can limit scientific research, but also challenge wildfire operations and fuel management decisions. Improving fuel typing accuracy will require a significant effort in fuel inventory data and system upgrades to adequately represent the diversity of extant fuels. To more effectively link conditions to expected fire behavior outcomes, we recommend a fuel classification approach and emphasis on observed fuels and measured fire behavior data for the systems we seek to represent.

Wildfire spreading prediction using multimodal data and deep neural network approach

Predicting wildfire spread behavior is an extremely important task for many countries. On a small scale, it is possible to ensure constant monitoring of the natural landscape through ground means. However, on the scale of large countries, this becomes practically impossible due to remote and vast forest territories. The most promising source of data in this case that can provide global monitoring is remote sensing data. Currently, the main challenge is the development of an effective pipeline that combines geospatial data collection and the application of advanced machine learning algorithms. Most approaches focus on short-term fire spreading prediction and utilize data from unmanned aerial vehicles (UAVs) for this purpose. In this study, we address the challenge of predicting fire spread on a large scale and consider a forecasting horizon ranging from 1 to 5 days. We train a neural network model based on the MA-Net architecture to predict wildfire spread based on environmental and climate data, taking into account spatial distribution features. Estimating the importance of features is another critical issue in fire behavior prediction, so we analyze their contribution to the model’s results. According to the experimental results, the most significant features are wind direction and land cover parameters. The F1-score for the predicted burned area varies from 0.64 to 0.68 depending on the day of prediction (from 1 to 5 days). The study was conducted in northern Russian regions and shows promise for further transfer and adaptation to other regions. This geospatial data-based artificial intelligence (AI) approach can be beneficial for supporting emergency systems and facilitating rapid decision-making.

A statistical approach to analyze and forecast the dynamics of active fire in the Brazilian legal Amazon

Anthropic fires are hugely responsible for the deforestation of the Brazilian legal Amazon region, one of the world’s most important ecosystems. In this way, policies must be implemented to reduce such a phenomenon. Preventive policies must be manifold, but historical time series analysis and active fire behavior predictions might subside part of the actions. Following this track, the present study aims to understand the spatiotemporal dynamics of active fire, using historical data of active fire comprising all states in the Brazilian legal Amazon region from January 2000 to December 2022. We used information from the National Institute for Space Research to fit appropriate statistical models for counting time series and forecast the monthly behavior of active fire until May 2023. The results showed that approximately 73% of active fires were registered in Pará, Mato Grosso, and Rondônia states, mainly between 2000 and 2010. In Pará, the forecast obtained reflects the phenomenon’s seasonality, with a higher concentration between August and November. In summary, the adopted methodology provided results that facilitated an understanding of the dynamics of active fire in the Brazilian legal Amazon region and can be considered a relevant tool to help authorities formulate public policies for arson prevention.

Prediction of fire spalling behaviour of fibre-reinforced concrete

Fire spalling prediction of fibre-reinforced concrete containing polypropylene (PP) or steel fibre at elevated temperatures is challenging. It is difficult to use conventional methods, such as finite-element modelling and discrete-element modelling, to solve the problem, because of the complicate coupling mechanism of PP and steel fibres in concrete. To this end, two artificial neural network (ANN) models, one (ANN1) based on a concrete mix study and the other (ANN2) based on a compressive strength study, were introduced to assess the resistance of concrete to explosive spalling. Test data (321 and 318 items, respectively) gathered from the literature were utilised to train the proposed models. Twenty-four concrete mixes (96 groups), that is, seven plain concrete mixes, four high performance concrete mixes reinforced with PP fibre, three ultra-high-performance concrete (UHPC) mixes with reinforced PP fibre and ten UHPC mixes reinforced with PP and steel hybrid fibres were designed and tested to validate the accuracy of the models. The study demonstrates that ANN1 and ANN2 can achieve a predictive accuracy of 89.6% and 84.4% for the explosive spalling, respectively, indicating the feasibility of the proposed models for predicting explosive spalling threat of the hybrid fibre-reinforced concrete.

Hybrid fire collapse simulation of reinforced concrete structures under localized fires

The safety of extensive reinforced concrete (RC) buildings under sudden fire scenarios is of particular concern nowadays. Localized fires may bring serious damage and even collapse to the RC bearing structures, causing huge casualties and economic losses. However, due to the limitations of test facilities, technology, and high costs, etc., full-scale and even reduced-scale structural fire tests have been very limited. The recent development of numerical techniques instead of challenging physical tests has provided exciting prospects for refinement of structural fire behavior prediction. This paper is therefore focused on a hybrid fire collapse simulation method through performing damage analysis of the locally fire-heated components and collapse simulation of the remaining major structure. The damage and failure behavior of the fire-exposed substructure is simulated based on an energy-based elastoplastic damage model for concrete at elevated temperatures. While the resulting deformation and collapse of the rest main structure is modeled by conventional collapse analysis. The key to the hybrid simulation is the coordination of internal forces and deformations between the substructure and the main structure. The developed method allows for an integrated simulation of the whole process from high temperature damage of materials to fire destruction of local components and to the collapse of the entire RC structure.

Development of a Fuel Model for Cistus spp. and Testing Its Fire Behavior Prediction Performance

This paper presents the development of a fuel model that can describe fuel situations in areas dominated by Cistus spp. (rockrose) in Greece. In order to obtain the necessary fuel data, thirty (30) 1 m2 plots were destructively sampled in phryganic areas dominated by Cistus creticus, Cistus parviflorus, and Cistus salvifolius in western Greece. To develop the fuel model for Cistus spp., field measurements were supplemented with published parameter values for Cistus salvifolius. The resulting fuel model (with a height of 9.44 cm) is suitable for describing Cistus spp.-dominated phryganic areas of relatively low vegetation height in southeastern Europe. Once developed, the fuel model was inputted into the BehavePlus system to produce surface fire rate of spread predictions (ROSpred, m·min−1), which were then compared to 21 surface fire ROS field observations (ROSobs) obtained from wildfires or prescribed burns in areas covered by Cistus spp. It was found that the ROS for the Cistus spp. fuel model significantly underpredicted the actual ROS. A statistically significant linear regression equation was developed to mathematically describe the relationship between the predicted and observed ROS. This equation can be used to adjust BehavePlus predictions to more accurately reflect the real-world fire behavior for this fuel type.

Differences in Canopy Cover Estimations from ALS Data and Their Effect on Fire Prediction

Canopy cover is a primary attribute used in empirical wildfire models for certain fuel types. Accurate estimation of canopy cover is a key to ensuring accurate prediction of fire spread and behaviour in these fuels. Airborne Laser Scanning (ALS) is a promising active remote sensing technology for estimating canopy cover in natural ecosystems since it can penetrate and measure the vegetation canopy. Various methods have been developed to estimate canopy cover from ALS data. However, little attention has been given to the evaluation of algorithms used to calculate canopy cover and the subsequent influence these algorithms can have on wildfire behaviour models. In this study we evaluate the effect of using different algorithms to calculate canopy cover on the performance of the Australian Mallee-heath fire spread model. ALS data was used to derive five canopy cover models. Fire spread metrics including burned area, unburned area within the fire extent, and extent of fire were compared for different model run times. The results show that these metrics are strongly influenced by choice of algorithm used to calculate canopy cover. The results from this study may provide practical guidance for the optimal selection of estimation methods in canopy cover mapping.

Detecting, Monitoring and Foreseeing Wildland Fire Requires Similar Multiscale Viewpoints as Meteorology and Climatology

Achieving sustainable coexistence with wildfires in the Anthropocene requires skilful integrated fire observations, fire behaviour predictions, forecasts of fire risk, and projections of change to fire climates. The diverse and multiscale approaches used by the atmospheric sciences, to understand geographic patterns, temporal trends and likely trajectories of weather and climate, provide a role model for how multiscale assessments of fire danger can be formulated and delivered to fire managers, emergency responders and at-risk communities. Adaptation to escalating risk of fire disasters requires specialised national agencies, like weather services, that provide to provide a diverse range of products to enable detection and near and longer-range prediction of landscape fire activity.

Indoor Experiments on the Moisture Dynamic Response to Wind Velocity for Fuelbeds with Different Degrees of Compactness

The semiphysical method is presently the most widely used for predicting litter moisture content, but it produces some errors. These are mainly due to the simplification of the water loss process and not accounting for the fuelbed structure, which can have a serious impact on the accuracy of litter moisture content predictions and, consequently, on forest fire management. As such, in this study, we constructed fuelbeds with different degrees of compactness, and the moisture content is saturated at this time. The drying process is recorded every 10 min under different wind velocity, and the experiment is stopped when the moisture content is not changing. Taking the saturated fibers’ moisture content (30%) as the threshold value, the drying process was artificially divided into two stages (from the initial moisture content to 30%, it is a process of free water drying, and from 30% to the equilibrium moisture content, this is the process of drying of bound water), which is called the distinguishing drying process. The whole drying process (from the initial to the equilibrium moisture content) is called the undistinguishing drying process. Drying coefficient and effect factors were calculated by distinguishing and not distinguishing the drying process, respectively. This established a prediction model based on compactness and wind velocity. The results show that the drying coefficients, k2 and k, of the two litter types were significantly different: the k2 of the white oak fuelbed was significantly lower than its k, with a maximum variation difference of 57.10%. The k2 in the Masson pine fuelbed was significantly higher than its k, with a maximum variation difference of 72.76%. Wind velocity and compactness had significant effects on all the drying coefficients of the two litter types, but with changes in the effect factors. The changes in k2 were weaker than those of the other drying coefficients. Compared with the model that did not distinguish the drying process, the MRE of the prediction models for white oak and Masson pine decreased by 27.39% and 2.35%, respectively. The prediction accuracy of the model of the drying coefficient obtained by distinguishing the drying loss process was higher than that of the model that did not distinguish the drying process. This study was an indoor simulation experiment that elucidated the drying mechanism of litter and established a prediction model for the drying coefficient based on effect factors. It is of great significance for further field verification and for improving the accuracy of moisture content predictions based on the semiphysical method and will significantly improve the accuracy of fire risk and fire behavior prediction.

Estimating Mediterranean stand fuel characteristics using handheld mobile laser scanning technology

Background Accurate, timely and easily obtainable information on stand fuel is of great importance in the prediction of fire behaviour. Aims The objective of this study is to measure several stand fuel characteristics with handheld mobile laser scanning (HMLS) in six fuel types for Mediterranean region, and compare the results with traditional field fuel measurements (FFM) in 35 different sampling plots. Methods The measurements involved overstorey (the number of trees, diameter at breast height, crown base height, tree height, maximum tree height, stand crown closure) and understorey (understorey closure, understorey height) fuel characteristics, and ground slope. Correlation analysis and t-test were performed to examine the relationship between FFM and HMLS datasets. In addition, cross-validation statistics (RMSE, rRMSE and R2) were employed to evaluate the accuracy of the HMLS method. Key results The results indicated strong correlations among all fuel characteristics. However, overstorey fuel characteristics were more favourable (r-values between 0.804 and 0.996, P &lt; 0.01) than understorey (r-values between 0.483 and 0.612, P &lt; 0.01). There was no significant difference between FFM and HMLS datasets in all fuel characteristics (P &gt; 0.05). Conclusions The results indicated that the HMLS was practical, cost-effective, time-efficient and required less labour as compared to traditional FFM in plot-level (i.e. 0.1 ha) inventories.

An artificial intelligence framework for predicting fire spread sustainability in semiarid shrublands

Background Fire behaviour simulation and prediction play a key role in supporting wildfire management and suppression activities. Aims Using machine-learning methods, the aim of this study was to predict the onset of fire propagation (go vs no-go) and type of fire behaviour (surface vs crown fire) in southern Australian semiarid shrublands. Methods Several machine-learning (ML) approaches were tested, including Support Vector Machine, Multinomial Naive Bayes and Multilayered Neural Networks, as was the use of augmented datasets developed with Generative Adversarial Networks (GAN) in classification of fire type. Key results Support Vector Machine was determined as the optimum machine learning classifier based on model overall accuracy against an independent evaluation dataset. This classifier correctly predicted fire spread sustainability and active crown fire propagation in 70 and 79% of the cases, respectively. The application of synthetically generated datasets in the Support Vector Machine model fitting process resulted in an improvement of model accuracy by 20% for the fire sustainability classification and 4% for the crown fire occurrence. Conclusions The selected ML modelling approach was shown to produce better results than logistic regression models when tested on independent datasets. Implications Artificial intelligence frameworks have a role in the development of predictive models of fire behaviour.

Crown fire initiation of a thunderstorm

Understanding bushfire–atmosphere interactions is essential for accurate prediction of fire behaviour, and for the safe and effective strategic management of fires to mitigate risk to people and property. Bushfires with feedbacks to thunderstorms represent the most extreme form of fire–atmosphere interaction, with potential to initiate tornadoes, lightning and hazardous winds causing dangerous fire behaviour and new ignitions many kilometres from the fire front. However, there is very little evidence that links quantitative fire behaviour with observed thunderstorm dynamics. Here we combine stochastic modelling of fire behaviour with satellite and mobile weather radar data of a bushfire thunderstorm in Queensland, Australia. The results show the coupling between fire behaviour and thunderstorm development in a conditionally unstable atmosphere. The process by which the coupling occurs raises questions as to the cause and effect relationship of the bushfire-initiated thunderstorms and associated fire behaviour. Recommendations for future research are made, highlighting the need for understanding links between modelled and observed fire behaviour dynamics and atmospheric thermodynamics.

Improved logistic models of crown fire probability in Canadian conifer forests

Background Crown fires are an ecologically necessary but hazardous process in conifer forests. Prediction of their behaviour in Canada has largely depended on the Canadian Forest Fire Behaviour Prediction System, in which fire weather indices drive primarily fixed fuel type models. The Crown Fire Initiation and Spread (CFIS) system presents a more flexible approach to predicting crown fire occurrence than fixed fuel type models. Aims Using a multi-decadal database of experimental fires carried out in conifer plots (1960–2019, n = 113), our aim was to develop updated models based on the CFIS system approach, fitting crown fire occurrence models to fire environment variables using logistic regression. Methods We tested alternative fuel moisture estimates and compared various model forms using repeated cross-validation. In two-storeyed stands, crown fire occurrence was defined as the involvement of lower canopy stratum fuels. Key results Final models based on wind speed, fuel strata gap, litter moisture and surface fuel consumption predicted crowning events correctly in up to 92% of cases in training data (89% in cross-validation). Conclusions and implications These new models offer improved accuracy and flexibility that will help users assess how competing environmental factors interact under different fuel treatments and wildfire scenarios.