Thomas J. Duff

Operational wildfire suppression modelling: a review evaluating development, state of the art and future directions

Wildfires are an inherent part of the landscape in many parts of the world; however, they often impose substantial economic burdens on human populations where they occur, both in terms of impacts and of management costs. As wildfires burn towards human assets, a universal response has been to deploy fire suppression resources (crews, vehicles and aircraft) to extinguish them, and limit their spread or impacts. The determination of the appropriate levels of investment, resource allocation and suppression tactics is a challenge for managers. As suppression expenses account for a substantial proportion of the cost of fires, and escaped fires account for a large portion of impacts, fire suppression models have been developed to better inform decision-makers. We undertake a review of the literature pertaining to the development of operational models that emulate fire suppression as part of decision support systems. We provide a summary of the development of modelling approaches, discuss strengths and limitations and provide perspectives on the direction of future research.

Predicting continuous variation in forest fuel load using biophysical models: a case study in south-eastern Australia

The increasing potential for wildfires in Mediterranean-type landscapes has resulted in pressure to mitigate fire threats. This is typically achieved by strategic reduction of fuel. To prioritise fuel management, it is necessary to understand vegetation dynamics and the relationships between plants and fuel. As the direct measurement of fuel in the field is labour intensive, mapped vegetation classes are typically used as to estimate fuel load. As vegetation properties vary continuously, the error in such estimates can be high. Remotely sensed and biophysical data are commonly used for vegetation classification, but rarely for estimating fuel load. This study investigated how fuel load varied with vegetation composition in an Australian woodland and assessed the potential for using biophysical models to create continuous estimates. Fuel was found to be influenced by species abundance, with some species having a greater contribution to load than others. Fuel was found to be somewhat predictable, with quantities related to fire history and several other biophysical variables. Models were applied to create continuous maps of fuel load; these provided a more precise representation of fuel variation than using discrete classes. Improved maps have the potential to facilitate improved prediction of fire behaviour and assist targeted fuel management.

Using discrete event simulation cellular automata models to determine multi-mode travel times and routes of terrestrial suppression resources to wildland fires

Forest fires can impose substantial social, environmental and economic burdens on the communities on which they impact. Well managed and timely fire suppression can demonstrably reduce the area burnt and minimise consequent losses. In order to effectively coordinate emergency vehicles for fire suppression, it is important to have an understanding of the time that elapses between vehicle dispatch and arrival at a fire. Forest fires can occur in remote locations that are not necessarily directly accessible by road. Consequently estimations of vehicular travel time may need to consider both on and off road travel. We introduce and demonstrate a novel framework for estimating travel times and determining optimal travel routes for vehicles travelling from bases to forest fires where both on and off road travel may be necessary. A grid based, cost-distance approach was utilised, where a travel time surface was computed indicating travel time from the reported fire location. Times were calculated using a discrete event simulation cellular automata (CA) model, with the CA progressing outwards from the fire location. Optimal fastest travel paths were computed by recognising chains of parent–child relationships. Our results achieved comparable results to traditional network analysis techniques when considering travel along roads; however the method was also demonstrated to be effective in estimating travel times and optimal routes in complex terrain.

Hillslope-scale prediction of terrain and forest canopy effects on temperature and near-surface soil moisture deficit

Soil moisture has important effects on fuel availability, but is often assessed using drought indices at coarse spatial resolution, without accounting for the fine-scale spatial effects of terrain and canopy variation on forest floor moisture. In this study, we examined the spatial variability of air temperature, litter temperature and near-surface soil moisture (θ, 0–100 mm) using data from field experiments at 17 sites in south-east Australia, covering a range of topographic aspects and vegetation types, within climates from semiarid to wet montane. Temperatures and θ in mountainous environments were found to vary at much finer spatial scales than typical drought index grid dimensions (several kilometres). Using terrain elevation, local insolation ratio and plant area index, we developed semi-empirical microclimate models for air and litter temperatures, then used modelled temperatures as input into calculations of the Keetch–Byram Drought Index, a widely used index of soil moisture deficit. Drought index results based on predicted litter temperature were found to explain 91% of the spatial variation in near-surface soil moisture at our experimental sites. These results suggest the potential for routine hillslope-scale predictions of forest floor moisture status, which may be useful in the management of fire, particularly prescribed burning, in complex terrain.

Patterns of plant abundances in natural systems: is there value in modelling both species abundance and distribution?

In plant ecology it is common to use biophysical models to predict species distribution; however, spatial quantitative models of plant species remain rare. In practice, occupancy models are often assumed to indicate habitant quality and are used as surrogate abundance models. This study assessed the potential value of quantitative models of plants for ecosystem management applications by assessing patterns of occupancy and abundance within two closely related understorey plant species, Xanthorrhoea australis and X. caespitosa. Vegetation quadrats were surveyed in Eucalyptus woodland and cover-abundances were assessed using a metric pin intersection technique. A zero inflated generalised additive modelling process was used to assess the relationship of species occupancies and cover-abundances to environmental properties. The models were applied to mapped environmental data to create spatial predictions of occupancy and cover-abundance. Both species shared several predictor variables, but differing responses to these variables resulted in mutually exclusive distributions. No significant correlation was observed between occupancy and cover-abundance for X. australis, but strong correlation was evident for X. caespitosa. The strength of the occupancy and abundance relationship was found to differ greatly between the two species and is therefore likely to be species specific. Occupancy models have been used successfully as proxies for habitat quality models of plant species; however where occupancy and abundance of plants are driven by different influences occupancy will be a poor surrogate for abundance. Outcomes may be improved if occupancy models are validated for abundance or quantitative models are developed and tested for individual species.

Indices for the evaluation of wildfire spread simulations using contemporaneous predictions and observations of burnt area

Methods to objectively evaluate performance are critical for model development. In contrast to recent advances in wildfire simulation, there has been limited attention to evaluating fire model performance. Information to validate fire models is typically limited, commonly to a few perimeter observations at a small number of points in time. We review metrics for comparing two burnt areas at a point in time: observed and predicted. These are compared in an idealised landscape and with a case study evaluating the performance of simulations of an Australian wildfire. We assessed: Shape Deviation Index (SDI), Jaccards coefficient, F1, Sorensens Similarity and Area Difference Index (ADI). For decomposing fit into error components (overprediction and underprediction) we assessed the partial indices of SDI and ADI, Precision and Recall. The various metrics were evaluated for their ability to represent error and their suitability for use in model improvement frameworks. Fire simulation models are of increasing importance to wildfire management.Verification of fire simulation models is necessary before operational use.There are no indices of model performance being used consistently.The properties of various fire model performance indices were evaluated.Performance indices properties should be considered when interpreting results.

Dryness thresholds for fire occurrence vary by forest type along an aridity gradient: evidence from Southern Australia

ContextWildfires are common in localities where there is sufficient productivity to allow the accumulation of biomass combined with seasonality that allows this to dry and transition to a flammable state. An understanding of the conditions under which vegetated landscapes become flammable is valuable for assessing fire risk and determining how fire regimes may alter with climate change.ObjectivesWeather based metrics of dryness are a standard approach for estimating the potential for fires to occur in the near term. However, such approaches do not consider the contribution of vegetation communities. We aim to evaluate differences in weather-based dryness thresholds for fire occurrence between vegetation communities and test whether these are a function of landscape aridity.MethodsWe analysed dryness thresholds (using Drought Factor) for fire occurrence in six vegetation communities using historic fires events that occurred in South-eastern Australia using logistic regression. These thresholds were compared to the landscape aridity for where the communities persist.ResultsWe found that dryness thresholds differed between vegetation communities, and this effect could in part be explained by landscape aridity. Dryness thresholds for fire occurrence were lower in vegetation communities that occur in arid environments. These communities were also exposed to dry conditions for a greater proportion of the year.ConclusionsOur findings suggest that vegetation driven feedbacks may be an important driver of landscape flammability. Increased consideration of vegetation properties in fire danger indices may provide for better estimates of landscape fire risk and allow changes to fire regimes to be anticipated.

Quantifying wildfire growth rates using smoke plume observations derived from weather radar

Fast-moving wildfires can result in substantial losses of infrastructure, property and life. During such events, real-time intelligence is critical for managing firefighting activities and public safety. The ability of fixed-site weather radars to detect the plumes from fires has long been recognised; however, quantitative methods to link properties of radar observed plumes to fire behaviour are lacking. We investigated the potential for weather radars to provide real time estimates of the growth of large fires in south-eastern Australia. Specifically, we examined whether the rate of change in fire area could be approximated using the change in volume represented by radar returns. We evaluated a series of linear mixed-effects models predicting fire-area growth using radar data representing a range of dBZ thresholds and search volumes. Models were compared using an information–theoretic approach. Radar return volume was found to be a robust predictor of fire-area change. The best model had a minimum threshold of 10 dBZ and a search radius of 60 km (R2 = 0.64). Fire area and radar relationships did not vary significantly between radar stations, suggesting broad applicability beyond the dataset. Further development of the use of weather radars for wildfire monitoring could yield substantial benefits because of their high frequency of scan and broad coverage over many populated areas.

Messmate stringybark: bark ignitability and burning sustainability in relation to fragment dimensions, hazard score and time since fire

Messmate stringybark is common in forests across south-eastern Australia. The bark of these trees is persistent and produces firebrands that contribute to house loss and the difficulty of fire suppression during wildfires. The trees typically survive fire with the amount of bark depleted. We compared two common methods to assess messmate bark fuels: (1) field-based hazard assessment, and (2) desk-based assessment using mapped time since fire. Our measurements included space-for-time field surveys and laboratory flammability tests. Although several physical properties of bark could be approximated from both assessment methods, some bark properties important to flammability were not captured. Ignitability was found to be dependent on the amount of char on bark fragments and could be predicted by the site assessment methods, whereas sustainability was dependent on bark fragment dimensions and could not be predicted by current methods. Bark fragment properties were found to be partially a function of tree size. Overall, these findings indicate that current bark assessment methods do not capture all the key bark properties that contribute to messmate bark’s flammability. Further research is warranted to improve bark assessment methods so they better reflect bark’s contribution to fire behaviour.