A Risk-based Approach to Wildfire Management

A Risk-based Approach to Wildfire Management Sam Nicol Honours Dissertation November 2004 Supervisors: Prof. Murugesu Sivapalan & Dr. Ray Steedman 1 Acknowledgements No thesis could possibly be completed without the help of a wide range of people. This page is my opportunity to thank those who have helped me reach this point. Firstly I would like to thank my supervisors for this project, Siva and Ray, for their patience, knowledge, enthusiasm and professionalism. I have the utmost respect for you both and thank you for the opportunity to contribute to such an exciting project. My thanks also to the Department of Conservation and Land Management Fire Services Division, especially Darren Wallace, Chris Muller and Rick Sneeuwajgt for their help and advice. I would finally like to thank my family and friends – thank you for putting up with the stress, the silences as I stewed over the latest problem, and for providing me with an outlet I can always turn to for support. Thanks especially to my friends at the CWR – it would not have been the same without you all. i Abstract Wildfires in the northern jarrah forest of south-west Western Australia are a frequent occurrence and cause considerable disturbance to ecosystems and property. A reliable risk based management strategy to manage wildfire may enhance protection of life, property and the environment. This study uses techniques derived from flood hydrology and the Australian Standard for risk management to calculate a quantitative, probability based measure of the risk of wildfire. The present study represents a first step in an intricate problem. Seventeen years of fire data were obtained from the Department of Conservation and Land Management’s Fire Management Services Division. These data were used to calculate probability distribution functions which were used to randomly simulate, in Monte Carlo fashion, a series of fire intensities for 100 years of record using an existing empirical and deterministic model. Extreme value series fire intensities and their associated return periods were calculated based on the time series of simulated fires. The contribution of Fire Danger Index (FDI), fuel load and surface moisture content to extreme fire intensity were examined. It was found that the fuel load reduction is effective only if fuel loads are maintained at low levels. Once the fuel load exceeds some threshold, the extreme fires are dominated by the weather conditions (FDI). The return period of uncontrollable fires in the northern jarrah forest was found to be close to 1 year as a result of the occurrence of extreme weather conditions. Risk management may improve the existing hazard management approach to wildfire management by shifting the focus of management to the asset to be protected. This allows for the consideration of alternative mitigation strategies not considered using pure hazard management. A simple application of a design standard to a test pole was demonstrated to illustrate how an alternative mitigation strategy can be practically applied. ii Glossary of Terms Annual Exceedance Series: Extreme value series in which events are selected if they have an intensity which exceeds a predefined value. The number of events selected is equal to the number of years of record. Annual Maximum Series: Extreme value series in which the largest intensity event in each year of record is selected. Asset Protection: A mitigation approach that seeks to identify and protect the assets at risk by selecting the most appropriate technique/techniques from a range of strategies. CALM: Abbreviation for the Department of Conservation and Land Management, the management authority responsible for much of the forest in Western Australia. Fire Danger Index: An aggregate index comprised of surface moisture content and wind speed that represents the forward rate of headfire spread under well-defined conditions. Fuel: Combustible plant material, both living and dead, that is capable of burning in a wildland situation (National Wildlife Coordinating Group 2001). Hazard Management: A mitigation strategy which seeks to reduce risk by the removal of the object causing the potential danger. Intensity: The heat energy released per unit of time for each unit length of the leading fire edge. The units are kilowatts per metre of fire front. Monte Carlo Method: The process of using random sampling from distributions to approximate solutions to probabilistic or deterministic problems. iii Prescribed Burning: Controlled application of fire to wildland fuels in either their natural or modified state, under specified environmental conditions that allow the fire to be confined to a predetermined area, and produce the fire behaviour and fire characteristics required to attain planned fire treatment and resource management objectives (National Wildlife Coordinating Group 2001). Probability distribution function: A plot of all the possible values that a parameter can assume, each with an associated cumulative probability of occurrence that is derived from the frequency of occurrence in the data set. Risk: The chance of something happening that will have an impact upon objectives. It is measured in terms of consequences and likelihood (Standards Australia & Standards New Zealand 1995). Return Period: The average recurrence interval between events equalling or exceeding a specified magnitude (Chow et al. 1988). Surface Moisture Content: Percentage of water by mass in the top 10mm of the litter bed. Wildfire: A free burning and unwanted wildland fire. Wildland-Urban Interface: The area where structures and other human development intermingle with undeveloped forest or vegetative fuels. iv Table of Contents Acknowledgements.............................................................................................................. i Abstract ............................................................................................................................... ii Glossary of Terms.............................................................................................................. iii Table of Contents................................................................................................................ v List of Figures ................................................................................................................... vii List of Equations .............................................................................................................. viii 1. Introduction................................................................................................................. 1 1.1. Specific Objectives of the Study......................................................................... 1 2. Literature Review........................................................................................................ 3 2.1. Fire in Western Australia .................................................................................... 3 2.1.1. Objectives of the management authority .................................................... 3 2.1.2. Ecological Response to Fire in South-Western Australia........................... 4 2.1.3. Effects of fire on life and property in Western Australia............................ 6 2.2. Prescribed Burning.............................................................................................. 8 2.2.1. The rationale behind prescribed burning: Fuel load reduction and rotation times 8 2.2.2. Criticisms of prescribed burning................................................................. 9 2.3. Predicting Wildfire............................................................................................ 13 2.3.1. Existing models......................................................................................... 13 2.3.2. Current Western Australian Model: The Wildfire Threat Analysis.......... 14 2.4. Risk ................................................................................................................... 18 2.4.1. Definition of risk....................................................................................... 18 2.4.2. Are Risk-Based models Appropriate? ...................................................... 21 2.4.3. Existing Applications of the Return Period Risk-based approach............ 21 3. Background ............................................................................................................... 24 3.1. The Study Site................................................................................................... 24 3.2. Fire Science....................................................................................................... 25 3.2.1. Causes of wildfire ..................................................................................... 25 3.2.2. Factors Influencing Fire Intensity............................................................. 26 4. Methods..................................................................................................................... 31 4.1.1. The Australian Standard for Risk Management applied to Wildfire Management.............................................................................................................. 31 4.1.2. Method overview ...................................................................................... 35 4.1.3. Obtaining probability distribution functions............................................. 36 4.1.4. Intensity Simulation: Monte Carlo techniques ......................................... 42 4.1.5. Selecting the Extreme Fires ...................................................................... 46 4.1.6. Return Period Analysis: Wildfire risk....................................................... 47 4.1.7. Contribution of the parameters to extreme fire events.............................. 49 4.1.8. Spatial Considerations: Effect of the size of the study area...................... 49 4.1.9. Application: Mitigation............................................................................. 50 5. Results....................................................................................................................... 52 5.1.1. Probability Mass function: Cause of fire .................................................. 52 5.1.2. Probability distribution functions: Annual Number of Fires, FDI, Fuel Load and Surface Moisture Content ......................................................................... 53 v 5.1.3. Extreme value distributions - the Annual maximum and Annual Exceedance Series..................................................................................................... 56 5.1.4. Contribution of the fuel load, the FDI and the surface moisture content . 58 5.1.5. Testing the Effects of Extreme Fuel Loads............................................... 63 5.1.6. Correlation between the variables............................................................. 67 5.1.7. Effect of the size of the study area............................................................ 70 5.1.8. A risk-based standard for mitigating the risk of ignition due to radiant heat 71 6. Discussion ................................................................................................................. 73 6.1. Fire in WA - Implications for management..................................................... 73 6.1.1. Cause of ignition ....................................................................................... 73 6.1.2. Return period of extreme intensity wildfire events................................... 74 6.1.3. Mitigation.................................................................................................. 82 6.2. Fire in WA - appropriateness of the model....................................................... 83 6.2.1. Empirical vs. physical model.................................................................... 83 6.2.2. Sensitivity of the model ............................................................................ 84 6.2.3. Spatial Considerations: The size of the study area ................................... 86 6.2.4. Correlations between the variables ........................................................... 86 6.2.5. The Annual Maximum series vs. the Annual Exceedance Series............. 87 6.3. Limitations of the study .................................................................................... 89 7. Conclusions............................................................................................................... 91 8. Recommendations..................................................................................................... 93 9. References................................................................................................................. 95 10. Appendices.......................................................................................................... 100 10.1. MATLAB Scripts used in the analysis ....................................................... 100 vi List of Figures Figure 1: Wildfire Threat Analysis mapping hierarchy ________________________________________ 16 Figure 2: Map of the Swan and Southwest regions of the Northern Jarrah showing fuel age and spatial extent of the regions ___________________________________________________________________ 25 Figure 3: Steps in the risk management process (Australian Standard 4360:1995) __________________ 32 Figure 4: Probability mass function illustrating the probability that ignition occurred due to a given cause ___________________________________________________________________________________ 52 Figure 5: Probability distribution functions for (a) Number of fires with a burnt area greater than 10ha; (b) Fuel load (t/ha); (c) Fire Danger Index; and (d) Surface Moisture Content (%)._________________ 53 Figure 6: Intensity (kW/m) vs. Return period (yrs) for 100 years of record. (a) The plot shows the annual maximum series, which includes the largest intensity fire from every year of record. (b) The plot shows the annual exceedance series, which includes the 100 largest fires in the record. Note the difference in the intensities of the low return period fires between the two figures.________________________________ 57 Figure 7: Fuel load (t/ha) vs. return period for 100 years of simulation. These fires are the same fires shown in Figure 6. (a) shows the contribution of fuel loads to the extreme intensity fires in the annual maximum series in Figure 6a. (b) shows the contribution of fuel loads to the extreme intensity fires in the annual exceedance series in Figure 6b. ____________________________________________________ 59 Figure 8: Fire Danger Index (FDI) vs. return period for 100 years of simulation. These fires are the same fires shown in Figure 6. (a) shows the contribution of FDI to the extreme intensity fires in the annual maximum series in Figure 6a. (b) shows the contribution of FDI to the extreme intensity fires in the annual exceedance series in Figure 6b. __________________________________________________________ 60 Figure 9: Surface moisture content (%) vs. return period (yrs) for 100 years of simulation. These fires are the same fires shown in Figure 6. The plots show the contribution of surface moisture content to the extreme intensity fires in (a) the annual maximum series in Figure 6a and (b) the annual exceedance series in Figure 6b. _________________________________________________________________________ 62 Figure 10: Sensitivity of the fuel load to canopy cover and fuel age. Fuel load (t/ha) is plotted on the vertical axis and fuel age (yrs) on the horizontal axis. The blue, green and red lines represent canopy covers of 10%, 50% and 100% respectively. ________________________________________________ 63 Figure 11: Exponential curve fit to cumulative probability distribution function for fuel load with canopy cover set to 100%. The curve has been extrapolated to include fuel ages up to 70 years. _____________ 65 Figure 12: Fire Intensity (kW/m) and return period (yrs) for extreme intensity fires (annual exceedance series) showing the effect of varying fuel load (canopy cover and fuel age). (a) shows extreme maximum conditions- 100 % canopy cover and fuel age 0-70 years. (b) depicts minimum conditions- 10% canopy cover and fuel age 0-30 years. ___________________________________________________________ 66 Figure 13: Fuel load (t/ha) and return period (yrs) for extreme intensity fires (annual exceedance series) showing the effect of varying canopy cover. (a) shows extreme maximum conditions: 100% canopy cover and fuel age 0-70 years. (b) depicts minimum conditions: 10% canopy cover and fuel age 0-30 years. __ 67 Figure 14: Plots showing the correlation between some of the variables used to determine fire intensity: (a) shows the relationship between area burnt (ha) and fire danger index (FDI); (b) shows the relationship between area burnt (ha) and fuel age (yrs); (c) shows the relationship between fuel age (yrs) and FDI. Correlation coefficients are given in the lower right hand corner of each figure. ___________________ 69 Figure 15: Effect of study area on intensity versus return period plot. ____________________________ 70 Figure 16: Effects of risk mitigation strategy: Intensity (kW/m) vs. Return Period (yrs). The dark blue line represents the intensity if no action is taken, and the green, red and light blue lines show the effects of clearing the fuel load distances of 10m, 30m and 50m from the pole respectively. The dashed orange line at 40kW/m represents the design standard (ignition intensity of the test pole). _____________________ 72 vii List of Equations Equation 1: The Fire Danger Index (FDI) ....................................................................................................27 Equation 2: The Poisson Probability Distribution Function.........................................................................38 Equation 3: Fuel Load...................................................................................................................................40 Equation 4: Fire Intensity Equation (General Form) ...................................................................................44 Equation 5: Fire Intensity Equation Adjusted to Include the Heat of Combustion .......................................45 Equation 6: Rate of Headfire Spread ............................................................................................................45 Equation 7: Equations for the Fuel Quantity Correction Factor ..................................................................46 Equation 8: Return Period ............................................................................................................................49 Equation 9: Parameter of the Poisson distribution with correction for size of study area ...........................50 Equation 10: Radiant Heat Intensity Decay Equation ..................................................................................51 viii Introduction 1. Introduction 1.1. Specific Objectives of the Study The specific objectives of this study and the broad steps taken to achieve them are outlined below: 1) Investigate quantitative wildfire risk and apply it to environmental design and management, consistent with standard AS/NZS 4630:1995 a. Undertake a literature review to investigate quantitative wildfire risk. b. Understand Australian Standard for risk management AS/ NZS 4630:1995. c. Introduce a method for fire risk assessment that can be applied to design and management. 2) From existing data of annual atmospheric and forest parameters, simulate the probability and return period of fire intensity of extreme wildfire events a. Estimate the statistical distributions of weather and fuel variables b. Synthetically generate, in Monte Carlo fashion, the necessary data set to create a random data series. c. Select and apply a fire intensity model. d. Rank all events by fire intensity and estimate the return period using accepted statistical techniques. 3) Evaluate the parameter combinations that lead to the simulated extreme wildfire events a. Examine the combinations of environmental parameters and fuel loads that contributed to the extreme event. 4) Apply the return period of wildfire intensity to mitigation design and management of wooden structures. a. Propose a risk design method directed at fuel load reduction in a restricted area. 1 Introduction Chapter 2 of this dissertation contains a literature review. The purpose of the review is to provide the reader with the background to the effects of fire in the forests of southwestern Australia and to outline the models and methods that are used to manage wildfire, with an emphasis on Western Australia. The literature review presents a scientific argument for a probabilistic risk-based approach to wildfire management. A definition of risk for the study is proposed and the application, in other disciplines, of the method presented in this dissertation is discussed. Background to the study area and some further information on wildfire science and prediction are presented in Chapter 3. Chapter 4 presents the method used to fulfil objectives 2 – 4 above. The results of the analysis are presented in Chapter 5 and their implications are discussed in Chapter 6. Chapters 7 and 8, respectively, present the study conclusions and recommendations for the future. 2 Literature Review 2. Literature Review The literature review is split into four sections. Section 2.1 describes the objectives of the fire management authority in Western Australia, and provides a brief overview of the issues confronting the manager in achieving these objectives. The current management practice of prescribed burning is then introduced and assessed in light of these objectives (section 2.2). An overview of existing models for predicting wildfire is given, and the model currently used to implement the practices described in section 2.2 is introduced and discussed (section 2.3). The need for a quantitative and probabilistic risk-based approach to wildfire management is discussed in section 2.3.2. The final section (section 2.4) introduces the concept of risk and defines its meaning in this study. 2.1. 2.1.1. Fire in Western Australia Objectives of the management authority The forests of south-western Australia are managed by the Department of Conservation and Land Management (CALM). CALM’s current policy was introduced in 1987 under the Fire Management Policy. The aim of the policy is to balance two main objectives: to ensure an acceptable level of protection to human life and property, and to ensure that biodiversity and essential ecological processes in the forests are not disrupted (Muller 2001). These objectives are not unique- the USDA Forest Service has similar objectives (Andrews et al. 2003). It is well documented that fire causes a significant disturbance to ecological processes (Gill & Moore 1997; Knox et al. 1999; Lang 1997) . If this is indeed the case, then there would seem to be an inherent contradiction in the implementation of CALM’s management objectives. Since CALM manages the danger posed by fire to life and property by fuel load reduction- essentially burning the bush- then there must be disturbance to the ecology in the burned areas. However it is commonly accepted that the forests of south-western Australia have adapted to occasional disturbance by fire, and there is evidence that fire has become an essential part of the ecological cycle of succession in the forest (Underwood & Christensen 1981; Knox et al. 1999). This 3 Literature Review assumption is the key to CALM’s forest fire management strategy, the cornerstone of which is the policy of prescribed burning for the purposes of fuel reduction. In light of these objectives, it seems pertinent to discuss some of the commonly accepted theory for each of the management objectives. 2.1.2. Ecological Response to Fire in South-Western Australia Fire in the forests of south-western Australia is a natural phenomenon. Indeed it is frequently argued in the literature that fire is a necessary occurrence in these ecosystems, as a number of plant species require fire in order to reproduce, and others demonstrate impressive adaptations enabling them to regenerate after fires (Knox et al. 1999). Fire has also traditionally been considered to be an essential way of recycling nutrients in the forest ecosystem (Underwood & Christensen 1981). It is commonly argued that the type of vegetation that exists in an area is a direct result of the fire regime in that area (Knox et al. 1999; Underwood & Christensen 1981). There are two aspects of fire that are crucial to determining the vegetation type in an area: fire intensity and fire frequency. Different plant and animal species in the forest have different strategies for recovery after fire. This makes the intensity of the fire a critical factor that determines which species are able to colonise an area directly after a fire (Underwood & Christensen 1981; Morrison 2002). For example, in certain areas of the jarrah forest, a high intensity burn will result in a large number of legume species (Underwood & Christensen 1981). Because a high intensity burn clears the undergrowth, the heat resistant legume seeds are able to grow rapidly with little competition and hence colonise the area rapidly. However if the fire is of a low intensity, the legumes will often be out-competed by surviving rootstock species, or re-sprouters, such as jarrah and other woody scrub species (Underwood & Christensen 1981). 4 Literature Review Fire frequency is also an important factor in determining the ecosystem type. The species composition in an area is largely dependent on the time since the last fire. Although recolonisation in an area is usually dominated by the species that were prevalent before the fire, a number of species not recorded on the site for many years can often be found. It is common to see a dramatic increase in species number and diversity immediately after a fire (Underwood & Christensen 1981; Morrison 2002). However following this initial proliferation, a pattern of succession usually sets in (Morrison 2002). In the jarrah forest, the original colonising plants are usually replaced after 1-2 years by common wildflower species (1-2 years post fire), leguminous species (4-5 years post fire) and larger, woody plants and trees respectively (Underwood & Christensen 1981). Forests that have not been burnt in a long time are often quite open, with large trees and relatively little undergrowth (Underwood & Christensen 1981). It is worth noting that fire does not burn uniformly across the landscape. It is affected by many factors such as topography, moisture and variations in wind speed and fuel availability and type. This results in ‘patchy’ burns, which leave behind a mosaic of habitat types that undergo ecological succession in time (Knox et al. 1999). The only case where this does not occur is when the fire is too intense, which results in the death of almost all species in the area (Knox et al. 1999). Under ‘natural’ fire regimes, it is generally accepted that the fauna of the forests of southwestern Australia follow a pattern of succession similar to the flora (Underwood & Christensen 1981; Commonwealth Government of Australia 1984). Maintenance of a diversity of vegetation types and ages ensures a range of habitats that favours a high diversity of fauna. Humans have been undertaking widespread prescribed burning of the forests of southwestern Australia since 1953 for the purposes of reducing fuel loads in the forest (Haswell & Brown 2002). This practice has ignited considerable debate in the literature about the role of fire and the benefits of the well-documented patterns of succession. There exist a number of papers documenting the adverse effects of frequent deliberate 5 Literature Review fuel reduction burning on various species of invertebrate and vertebrate species (York 1994; Wilson 1994). There are concerns that frequent fire is responsible for the rapid spread of the jarrah dieback fungus (Underwood & Christensen 1981) that has affected large areas of the south-western forests, resulting in the death of many trees. In other parts of Australia, Neyland and Askey-Doran (1994) claimed that fuel reduction burning has resulted in a simplified forest structure in Eastern Tasmania. Fires are lit at relatively regular intervals (to prevent excess fuel build-up) and at relatively constant intensities (to ensure that the fires are controllable), removing the mosaic effect necessary for a wide range of habitat types. The result is a simplified forest structure lacking the diversity of natural forests (Neyland & Askey-Doran 1994). Current forest management practice acknowledges the need for a mosaic landscape, and attempts are often made to create ‘patchy’ areas by manipulating the fire interval and intensity of fuel reduction burns, however there remains uncertainty as to whether or not the regimes applied are optimal for species in the affected areas (Gill & Moore 1997; Department of Conservation and Land Management 2001). What can be concluded from all of the papers reviewed is that there is a great deal of uncertainty about the effects of fire on species and ecosystems (Wilson 1994; Franklin & Agee 2003). More research is needed into the effects of fire regimes on both flora and fauna if the current regimes are to silence their critics, and conversely more evidence that the fire regimes are causing irreparable harm to the environment are required if regimes are to be significantly altered. 2.1.3. Effects of fire on life and property in Western Australia There is a considerable stigma attached to bushfire in Western Australia, evidenced by the eagerness of the press to seize upon any action undertaken by the management authority responsible for the majority of fire management in the State, the Department of Conservation and Land Management (CALM) (for example, The West Australian (2004)). This fascination with fire is likely to stem from the State’s history of severe and sometimes disastrous fires, such as the Donnybrook fires of 1949-50 (Underwood & 6 Literature Review Christensen 1981) and the infamous Dwellingup fires of 1960-61, which provoked a Royal Commission (Rodger 1961). The strong reaction to fire in Western Australia is illustrated by the State’s prescribed burning policy. In 1953, the then Forests Department became one of the earliest authorities to remove the existing policy of fire exclusion and to adopt broad-scale prescribed burning for the purposes of fuel reduction (Underwood & Christensen 1981). Since this time, buoyed by the positive findings of the 1961 Royal Commission regarding the new policy, the forest management authorities have greatly increased the area of forest undergoing prescribed burning. The area managed by prescribed burning now includes all the jarrah forest and the majority of the karri forest on State-owned lands (Underwood & Christensen 1981), a huge area to monitor and control. New techniques such as aerial ignition have been developed to decrease the cost of such a massive undertaking (Foster 1976). There is no doubt that forest fires present a significant risk to life and property, and that where this risk exceeds acceptable levels it must be mitigated. Prescribed burning presents one option to reduce the hazard (fuel) and hence lower the risk to life and property. However burning such a massive area of the state at regular intervals is a costly and logistically challenging exercise. Recent work by the department of CALM has moved away from ‘blanket burning’ large areas of forest, towards a more strategic approach that targets areas where life and property are likely to come under threat (Muller 2001). The wildland-urban fringe is the area where suburban developments are combined with significant patches of forest. These areas carry the highest risk of damage to life and property as a result of wildfire (Fried et al. 1999; Braun 2001; Boura 1994). Extra emphasis needs to be placed upon these areas to ensure the safety of residents (Boura 1994; Keeley et al. 1999). 7 Literature Review The work of Braun (2002) argues that the wildland-urban interface presents another way to manage fire. He argues that there needs to be a fundamental shift in wildfire risk assessment away from the hazard agent (fuel) and onto the community and its capacity and vulnerability to wildfire (Braun 2001; Braun 2002). He claims that a large number of the adverse effects of wildfires on life and property can be mitigated by ensuring that the community is well educated and well prepared for fire events. Braun argues that managing the wildland-urban fringe presents a possible way to improve the management of the risk of wildfire to life and property. Having introduced the objectives of the wildfire authority and having highlighted the complex nature of the problem, the discussion now shifts to the current management policy. As previously mentioned, the management objectives are carried out using prescribed burning for the purposes of fuel load reduction. 2.2. Prescribed Burning 2.2.1. The rationale behind prescribed burning: Fuel load reduction and rotation times The Department of CALM’s objectives include managing fire to conserve biodiversity and ensuring an acceptable level of protection to human life and property. These objectives are realised using the process of fuel load reduction. The logic behind fuel load reduction is simple. Fire intensity is controlled by a large number of factors: air temperature and humidity, solar radiation, wind speed and direction, rainfall, atmospheric stability, the quantity, type and distribution of fuel, and topography (Foster 1976; Burrows 1984a; Luke & McArthur 1986). Of these factors, the only one that can be practically controlled is fuel quantity (Foster 1976; Department of Conservation and Land Management 2000; Luke & McArthur 1986). Fuel can be removed by collecting the fuel or more easily by burning it in a controlled situation. The aim is not to prevent forest fire, but to keep fires under control when they do occur (Underwood & Christensen 1981). 8 Literature Review Based on research and practical experience, CALM’s prescribed burning policy sets out threshold fuel levels (tonnes/ha) above which a forest fire cannot be effectively controlled under average summer conditions (Muller 1993). This threshold level also represents the upper limit to the fuel quantity beyond which the fire intensity will cause unacceptable damage to young trees (Department of Conservation and Land Management 2000). These limits are used to calculate a prescribed burn time (‘rotation’) within a strategic buffer zone. The idea of the buffer is that there will be areas of low-density fuel where natural fires can be effectively suppressed. In these areas fire protection must take precedence over wilderness or other forest values. The rotation depends on the site characteristics, including soil fertility and forest type and structure. On average the rotations are 5-7 years for jarrah, 7-10 years for jarrah-wandoo and 6-8 years for karri (Department of Conservation and Land Management 2000). 2.2.2. Criticisms of prescribed burning Prescribed burning policy has many critics for a variety of reasons. Arguments against prescribed burning can be classified into two broad classes: Specific and general. Specific arguments focus on a specific aspect of prescribed burning, such as hazardous smoke levels in towns and the metropolitan area (Department of Conservation and Land Management 2000; The West Australian 2004) or the effects of burning on a specific species of plant or animal (York 1994). General arguments accept that fire is a natural occurrence in Western Australian forests, but argue that there is no need to deliberately burn the forest for the purposes of preserving biodiversity, as forests existed for many thousands of years without significant human interference. The claim is that the inherent variability of natural forest fires can naturally maintain a mosaic of habitats that promotes a wide species diversity (York 1994; Neyland & Askey-Doran 1994). Specific arguments Specific arguments against prescribed burning focus upon the ecological effects of fire. There are few complaints about prescribed burning as a strategy to protect life and property. 9 Literature Review There is a broad area of research documenting the effects of fire on different species and ecosystems in the forest (Morrison 2002; Underwood & Christensen 1981; Wilson 1994). There is evidence to show that current fire regimes detrimentally affect the abundance and distribution of some species within the forest (York 1994; Neyland & Askey-Doran 1994; Morrison 2002), and that fire results in the export of nutrients from the forest, altering the soil properties and ecosystem dynamics (Commonwealth Government of Australia 1984). Water pollution can occur due to fires (Water and Rivers Commission 2000; Commonwealth Government of Australia 1984), and this has been raised as an argument against prescribed burning. However the weakness of these arguments is that a decrease in habitat quality for one species is likely to open a niche for another species in the cycle of succession discussed previously (Knox et al. 1999). Hence it is not clear whether prescribed burning can be considered detrimental to the ecosystem as a whole. Air pollution is a common complaint arising as a result of fires. In Western Australia, hazy days in the Perth metropolitan area are often attributed to smoke from CALM’s prescribed burning program (The West Australian 2004). However records of air quality maintained by CALM in 2000 showed that the department has not exceeded the national standard for particulates since 1994 (Department of Conservation and Land Management 2000). Some burns have exceeded the Department of Health guidelines, but CALM claims these exceedances were due to wildfires and not prescribed burns (Department of Conservation and Land Management 2000). Evidence from boreal forests in the northern hemisphere has shown that low fire frequency results in storage of carbon in the humus of forest soils, and that burning causes greenhouse gases to be released (Wardle et al. 2003). This has sparked argument in Western Australia that the prescribed burning policy is releasing significant greenhouse gases. However CALM’s research shows that litter in Western Australian forests is not decomposed into the humus as it is in wetter climates (McCaw et al. 2002). There is also evidence (Ward & Van Didden 1997) that fire frequency has decreased since European settlement, so the greenhouse gas argument remains debateable. 10 Literature Review Evidence also exists that prescribed burning is ineffective, as the number of wildfires under such regimes do not necessarily decrease (CALM unpublished data 2002). However such statistics are bedevilled by the fact that the monitored area has increased, and hence it is not certain if the frequency of fire has truly increased. Some papers show that the rate of accumulation of dead litter following a low-intensity burn is very rapid as a result of dead leaves and unburnt trash litter, and that prescribed burning is not having the desired fuel reduction effect (Franklin & Agee 2003). There is some substance to this argument, as evidence supports a rapid increase in fuel immediately following a fire (McCaw et al. 2002), however the amount of fuel may not reach equilibrium for up to 70 years (McCaw et al. 2002). It is argued that not burning the fuel presents a significant risk of a very intense uncontrollable burn (Foster 1976). The many arguments and criticisms directed at the prescribed burning policy illustrate the complex nature of the problem. There are many arguments against the burning, but almost all the arguments can be countered with contrasting evidence from CALM or other sources. The problem lies in the complexity of the fire regime, which can have a variety of impacts depending on a myriad conditions and variables (Franklin & Agee 2003), and it seems impossible that a management authority can satisfy all stakeholders operating under such complexity. General arguments The crux of most arguments over whether or not prescribed burning should occur is the question: what is a natural fire regime? The argument centres on whether or not humans have altered the environment so greatly that they need to interfere to maintain nature’s balance, or whether nature can survive without the prescribed burning policy. Central to the arguments of pro-burn advocates is the claim that fire was frequent throughout the forests prior to European settlement. This claim is based upon anecdotal evidence from the early Western Australian European settlers and explorers that the Nyoongar people of the south-west were burning the forests prior to European settlement 11 Literature Review in 1829 (Ward 2000; Ward & Van Didden 1997; Foster 1976; Underwood & Christensen 1981). It is claimed in some reports that Aboriginals maintained a mosaic of fuel and habitat types similar to today’s fire managers (Burrows 2003; Underwood & Christensen 1981). However while most reports accept that the Aboriginal people utilised fire, there is disagreement about the scale of the areas burnt. Horton (2002) questions why Aboriginal burning is so widely accepted with so little evidence, claiming that the selected historical quotes describing isolated incidents do not justify the current broad-scale fire regime (Horton 2002). More quantitative justification for the extent of Aboriginal burning is provided by Ward and Van Didden (1997), who used the bark of grass trees to estimate the fire history of the jarrah forest at discrete points. It was found that the current fire regime utilises considerably longer rotations than that applied prior to European settlement, which Ward and Van Didden (1997) attributed to broad-scale Aboriginal burning. Another general question concerns the type of forest that is desirable. There is evidence that prescribed burning is simplifying the forest structure (Neyland & Askey-Doran 1994). If the findings of this study are true, it is necessary to question how we value diversity. Franklin and Agee (2003) questioned whether it was desirable to return to preEuropean settlement fire regimes given the widespread land use changes that have occurred. The moralistic dilemma is whether management agencies should be allowed to manipulate nature to suit human purposes to the detriment of certain natural ecosystem types (Franklin & Agee 2003). Both sides of the moralistic debate call upon the precautionary principle, that action should not be deterred based upon a lack of full scientific evidence, to support their case (Ward & Van Didden 1997). Opponents of the policy claim that burning should be prevented until it is certain that a fire regime is not detrimental to the ecology of the affected area. Supporters of the policy claim that the converse is equally true - that based upon the recent history of the forests, it is logical to continue to burn until it is certain that burning will not alter the soil chemistry and nutrient cycling in the forest (Ward & Van Didden 1997). 12 Literature Review Based on the above discussion, it is clear that there does not appear to be a definite right or wrong way to manage fires, there simply isn’t enough knowledge of the complexity of the forest to have an easy answer. Certainly blanket burning of the forest will not do the complexity and diversity of the forest justice, but there is no guarantee that a total ban would benefit the forests. Indeed, such a plan would present substantial risks to humans living near and within the forest. CALM has responded to the moralistic dilemma by attempting to maintain a mosaic of landscapes and ecosystems. Given that the objectives of the management authority do not allow them to do nothing (Department of Conservation and Land Management 2000), this seems a logical approach. This dissertation aims to present the framework for a quantifiable and probabilistic means to prioritise this mosaic to ensure that the forest is strategically burned only when the management objectives are jeopardised. 2.3. Predicting Wildfire Maintaining a mosaic of habitats across the vast area of CALM managed lands is a complex task, and requires the use of a model to prioritise the prescribed burning. The following section gives a brief overview of the major types of quantitative wildfire models currently available and details the model used by CALM to manage wildfire hazard in Western Australia. 2.3.1. Existing models There is an abundance of models that have been designed to model fire. For simplicity, this dissertation groups the models reviewed into three groups; empirically-based models, physical models and stochastic models (Keane et al. 2003). As the name suggests, empirical models are those models that make basic assumptions about the behaviour of fires and relate the variables via empirical coefficients, the values of which are based upon numerous observations of fire behaviour under a range of conditions. The relationships derived are deterministic. Models that employ this approach 13 Literature Review include Rothermal (1972) and Catchpole et al. (1999). Empirical approaches can become highly complex, employing techniques such as neural networks to predict the regime (McCormick et al. 1999). It is noteworthy for this dissertation that the department of CALM employs an empirical model, the Wildfire Threat Analysis (WTA) (Muller 1993), which is based upon a series of tables derived from experimental fire observation (Sneeuwajgt & Peet 1985). Physical models are based upon an understanding of fire mechanics and natural processes to predict deterministic fire regimes. Conventional two-dimensional spatial models assign properties to grid cells representing the topography of the study area, and calculate the rate of spread of the fire based upon the characteristics of the cells (Hargrove et al. 2000; Green 1983; Dupuy & Larini 1999). Stochastic models use probability and stochastic functions to determine wildfire spread and intensity. They differ from physical and empirical models in that they do not return a definite answer, but give a probability of a fire of a given intensity occurring. The models thus incorporate a quantitative risk that is useful for management and design. Examples of this type of model can be found in Keane et al. (2003) and Preisler et al. (2004). It should be noted here that a fire model does not need to fit exclusively into one of the categories, but that it is possible to create models that incorporate elements of all three model classes (Keane et al. 2003). 2.3.2. Current Western Australian Model: The Wildfire Threat Analysis The department of CALM has limited resources to carry out vast areas of burning, and its prescribed burning policy receives considerable public scrutiny (Muller 1993). In light of these constraints, the department has recognised that there is a clear need to formalise a systematic and uniform approach to strategically prioritise which areas should be burned. To fulfil this need, the Wildfire Threat Analysis (WTA) system was introduced in the early 1990s (Muller 1993). The WTA is a decision-making tool that provides a 14 Literature Review framework to incorporate all the factors leading to a wildfire threat, and allows fire managers to prioritise their controlled burns in an accountable way (Sneeuwajgt 1998). The WTA is used to determine the rotation time for each forest based on the likelihood of a severe fire occurring. CALM’s Wildfire Threat Analysis (WTA) package determines the risk of a wildfire becoming uncontrollable based on the values of the forest at risk (human and environmental), the potential for fires to start, the capacity of CALM to respond to a fire, and the expected fire behaviour. CALM also takes into account the juvenile period of the slowest maturing plants in the forest, and the post-fire responses of flora and fauna in the forest when calculating the rotation time. Different burn times are used to encourage different species. In areas where endangered fauna are present, the fire regime is based around encouraging the plant species that these animals require (Department of Conservation and Land Management 2000). The WTA attempts to assess the risk of wildfire by incorporating the following four factors (Muller 1993): • Values: Identification of the valuable areas in a region and prioritisation of these values based upon the severity of the consequences of loss of these values. There are seven classes of values including such values as human life (bush townships, settlements etc), high property values, fire vulnerable threatened species and • plantations (Sneeuwajgt 1998). Values are not based on economic (dollar) value. Risk of Ignition: Determination of the likelihood that a fire will start. There are many reasons a fire can start, including natural causes (lightning, friction between trees), deliberate lighting/arson, and accidental causes (escapes from burn-offs, campfires, sparks from trains, power line failure etc) (Luke & McArthur 1986). The WTA includes three ignition risk classes: high, medium and low. These are based on factors such as fire history in the area, visitor access to the area and the • proximity to areas undergoing land clearing (Sneeuwajgt 1998). Suppression Response: The likelihood that, given that a fire exists, it will be able to be extinguished. The response is determined by a combination of the detection time and travel time to the area, and the production time (time to construct a 15 Literature Review 1000m fireline) (Sneeuwajgt 1998). The longer the response time, the greater the • risk of wildfire in an area. Fire Behaviour: The expected behaviour of the headfire. The headfire is the section of the fire moving in the same direction as the wind or slope and is the most intense part of the fire (Burrows 1984a). The headfire behaviour determines the severity of the damage that may be caused by the fire and is measured as a rate of spread or intensity (Muller 1993). Headfire behaviour in the WTA is influenced by fuel, topography and weather conditions. Five classes of behaviour are employed in the WTA based upon the likelihood of a suppression attack on the fire of the given intensity succeeding (Sneeuwajgt 1998). Figure 1 shows these four factors and how they contribute to the wildfire threat (Muller 1993). Figure 1: Wildfire Threat Analysis mapping hierarchy These four factors are not combined together to create a single index. The Department of CALM recognised the importance of knowing the relative importance of each of the above factors, and the WTA is published as a series of maps showing the above factors 16 Literature Review (Muller 1993). The maps can be used as overlays to get some sense of the aggregate of the factors (Muller 1993). The WTA represents a move by the Department of CALM to a formalised, risk-based approach to fire management (Department of Conservation and Land Management 2000). Analysis using this tool has enabled the department to reduce duplication of suppression response zones, and hence to become more cost efficient (Sneeuwajgt 1998). The fire detection system was also optimised to reduce the cost of spotter flights. Spotter planes were able to be reduced by 30 per cent and wildfire detection time has decreased under the new system (Sneeuwajgt 1998). The WTA has been used to create master burning plans, which strategically allocate low fuel buffers on a large scale (Sneeuwajgt 1998). Dissertation Context The WTA considers risk in a broader sense than the traditional approach of assessing the contribution of the fuel load to the probable intensity of fire. It does this by incorporating the values at risk, the likelihood of ignition, the suppression response capabilities, and the expected fire behaviour (Muller 1993). This is a first step towards a new way of thinking about the risk of fire. However the mitigation strategies proposed to manage the risk of fire remain centred on managing the hazard (the fuel load). This dissertation seeks to determine the relative importance of the physical parameters that contribute to an extreme wildfire- the fuel load, the weather conditions and the moisture content of the fuel. The dissertation will argue that the focus of risk management should be shifted from hazard management to asset protection based upon the dominant source of risk to an asset, which may not necessarily be the fuel load. Whilst the WTA is a step towards a quantitative risk-based management of wildfire, the system is not quantitative in the sense that it does not lead to a standard that can be used in design. Such a design standard requires the assessment of risk in terms of probabilities. If the probability of a destructive fire occurring is known, then an acceptable probability of destruction can be defined as a standard. Structures can then be designed to withstand the standard fire intensity so that the structure has an acceptable probability of damage 17 Literature Review from fire. This study uses the framework from the Australian Standard 4360:1995 for risk management to illustrate how a very simplified design standard can be applied to wildfire risk and suggests some mitigation strategies that will target the dominant source of risk, not just the hazard. 2.4. Risk 2.4.1. Definition of risk The notion of risk has many connotations and meanings in different contexts (Standards Australia & Standards New Zealand 1995). This review defines the meaning of risk for the purposes of this dissertation and considers the meaning of risk as it has been applied to fire management. A pertinent point of departure for a risk definition is the Australian and New Zealand Standard for risk management AS/NZS 4360:1995, which defines risk as “the chance of something happening that will have an impact upon objectives. It is measured in terms of consequences and likelihood” (Standards Australia & Standards New Zealand 1995). This definition implies a need to have defined objectives, and that risk is an event that may impact the implementation of these objectives. The definition requires that risk be measured in terms of consequences and likelihood, implying that risk needs to be evaluated in terms of the probability of the management objectives being impacted and the severity of the consequences if the objectives are not completed. The definition of risk presented in the Australian Standard is employed in this dissertation. The standard contains a framework of five steps that allow for the assessment of risk using this definition and these steps are followed in the methodology of this dissertation. The definition is well suited to the problem of wildfire management because the objectives of the wildfire management authority are already well-defined and the impact of fire upon these objectives is understood. The concepts of likelihood and consequence in the definition are easily quantified and translated into probability theory. 18 Literature Review A slightly different definition is employed by the International Standard ISO/NWI 14140, which describes risk as “the combination of the chance that a specified undesirable event will occur and severity of the consequences of the event” (Goff & Steedman 1997). This definition of risk reduces the scope of events that need to be considered to include only those which are likely to have an undesirable impact. It does not explicitly state that objectives need to be defined. Goff and Steedman (1997) employed the ISO definition of risk to propose a framework for environmental risk management, which they applied to management of oil spills. Firstly the risk is classified into four independent categories: (a) Primary risk, which is the probability of an undesirable event occurring; (b) Secondary risk, which is the probability of the consequences of the event reaching an area of value given that the event has occurred; (c) tertiary risk, which is the probability of the event causing damage (severity) given that the consequences of the event reach an area of value; and (d) quaternary risk, which is the probability that an area damaged by the undesirable event can recover from the impact given that it has been damaged (Goff & Steedman 1997). The power of this framework is that it separates the components of environmental risk into independent probabilities, and hence it is possible to determine where the majority of the risk lies for a given area and prioritise mitigation strategies to target areas of high risk. Separating the elements of risk in this way and prioritising mitigation strategies ensures that a holistic view of risk is taken in which the protection of the asset at risk is the focus, rather than a view which focuses on one hazard. Because the events are independent, an overall risk can be obtained by the simple combination of the four probabilities representing the four types of probabilities (Manno 1999). For the purposes of wildfire risk management, it is possible to apply Goff and Steedman’s (1997) approach. The primary risk can be defined as the probability of a fire igniting in a given area. Secondary risk can be defined as the probability of a fire spreading through the landscape to a particular area of value (such as a town or high conservation area) given that it has ignited. Tertiary risk may be defined as the probability that a fire causes significant damage to an area of value given that it reaches 19 Literature Review the area. Quaternary risk may be defined as the probability that an area of value can recover from damages caused by fire given that damage occurs (for example, the forest can recover from low-intensity burns rapidly due to the adaptations of fire-resistant plants (Knox et al. 1999)). Although Goff and Steedman (1997) used the ISO definition for their framework for risk management, the approach they advocate is generic and would not be affected by using a similar definition of risk such as the Australian Standard definition. There is the potential to employ the framework of Goff and Steedman (1997) whilst continuing to comply with the Australian Standard. Goff and Steedman’s approach is not unique. A spatial model for estimating wildfire risk in the state of Oregon (USA) successfully applied a similar approach to that of Goff and Steedman (1997). In the model the probability of ignition based upon historical fire records was combined with the probability that a fire spreads to become a large fire to obtain the unconditional probability of a large fire occurring (Preisler et al. 2004). Spatial probability maps were produced showing areas of high predicted wildfire risk that fitted well with the historical record given the patchy data sets used (Preisler et al. 2004). Fried et al. (1999) assessed the unconditional probability of a fire occurring at the wildlandurban interface combined with the probability of the fire causing damage to a structure as part of a study to determine the willingness of residents to pay for a reduced risk of wildfire (Fried et al. 1999). Dissertation Context The approach of Goff and Steedman (1997) provides an overall goal for the assessment of wildfire risk. As a first step towards such a tool, this dissertation uses the Australian Standard for risk management definition of risk to demonstrate how the likelihood and consequences of wildfire can be assessed using the framework proposed in the Australian Standard for risk management (Standards Australia & Standards New Zealand 1995). The historical record of wildfire events is used to assess the return period (likelihood) and intensity (potential severity of consequences) of extreme wildfire events. This study is 20 Literature Review thus focussed upon the primary and tertiary risk (what is the probability that a fire will occur, and what damage will it cause?). The study does not attempt to assess the other steps in Goff and Steedman’s (1997) framework, but rather presents this approach as a vision for future research towards a more holistic definition of risk. 2.4.2. Are Risk-Based models Appropriate? A paper by Dovers (1995) addresses the issue of whether or not it is appropriate to apply risk-based models to an environmental problem. Dovers (1995) argues that human interactions with the environment are confounded by uncertainty. In light of this, it is argued that improved techniques of quantifying uncertainty or risk in the environment are essential to avoid purely political decision making. It is likely there will always be uncertainty in environmental decisions, and attempting to avoid making decisions due to the uncertainty does not solve the problem (Dovers 1995). Decisions will still be made, but with even less information than that which can be gleaned from an aggregate model (Dovers 1995). While improved techniques are critical to ensuring that decisions are well-informed, there is a need to ensure that the uncertainty inherent in the analysis of the techniques is presented (Dovers 1995). A probabilistic approach to uncertainty management provides a means of objectively designing in the face of uncertainty, based upon the best available information. Naturally all decisions will still be influenced by politics, however risk assessment allows for informed rather than purely emotional decisions, and provides a logical and justifiable basis for decisions (Muller 1993). 2.4.3. Existing Applications of the Return Period Risk-based approach This dissertation applies a risk-based model that is determined from the return period of wildfire. This approach is not unique, and forms the cornerstone of a wide variety of engineering design techniques (Kennedy & Neville 1986). 21 Literature Review Return period analysis is commonly used to create risk estimates for a wide range of natural disasters. This dissertation borrows heavily from the field of flood hydrology, where return periods are frequently used to design around waterways. An outline of some of the approaches taken by flood hydrologists can be found in (Chow et al. 1988; Kennedy & Neville 1986; Plate 1982; Pattison et al. 1977). Other areas where return period analysis has been used to quantify risk include cyclones (Steedman 1987; Petrauskas & Aagaard 1971) and storm surges (United States Army Corps of Engineers 2003). Risk of wildfire in central Idaho, USA, was evaluated using a combination of simulation and return period analysis (Keane et al. 2003). Dissertation Context The literature review has outlined the current wildfire management approach of prescribed burning on a specified rotation. It was argued that the current management approach lacks a design standard and remains focussed upon hazard reduction as a mitigation strategy (Braun 2001). It was proposed that a probabilistic risk approach could be utilised to address these issues in terms of likelihood of fire occurrence and the severity of the consequences. In the preceding section, risk was defined for the study and the application of return period analysis introduced and justified by its common use in other disciplines. The dissertation proposes that the dual objectives of the management authority may be balanced using a risk-based system to prioritise mitigation. It is argued that the areas of high value to human life and property are likely to be localised around the wildland-urban interface (Bradstock et al. 1998). Hence it is possible that mitigating the risk to life and property in these areas will suffice to fulfil the first objective of the management authority. The remainder of the forest may then be managed by using a prescribed burning rotation designed primarily for the benefit of biodiversity, fulfilling the second objective of the management authority. The remainder of this dissertation presents a methodology to probabilistically assess the risk of wildfire in the northern jarrah forest. It argues that the risk-based method 22 Literature Review presented has the potential to allow the management authority to satisfy both of their objectives. In addition, the dissertation seeks to show how quantitative risk can be used to determine the driving forces behind extreme fires, and argues that this allows for a shift in thinking from hazard reduction to asset protection as the focus of mitigation strategies (Braun 2001). 23 Background 3. Background 3.1. The Study Site The study uses historical fire data collected by the Department of Conservation and Land Management (CALM) from the Swan and Southwest regions of the northern jarrah forest. The area includes the CALM-managed forested area to the east of the Darling Scarp and stretches from near Lancelin in the north to close to Augusta on the south coast. The climate in the region is warm Mediterranean, with annual rainfall varying from 1300mm near the Darling Scarp to 700mm in the east of the region (Burrows et al. 1999). About 80% of this precipitation falls during the winter months (Burrows et al. 1999). A summer ‘drought’ is experienced every year, with 140-160 days a year across the region occurring where the fuel is considered dry enough to burn (Burrows et al. 1999). The area is dominated by jarrah (Eucalyptus Marginata) and marri (Corymbia calophylla) forest, with some wandoo-marri in the eastern part (Williams & Mitchell 2001; Burrows et al. 1999). Banksia woodlands and bullich and blackbutt communities are also dispersed amongst the forest (Williams & Mitchell 2001). The soil type is predominantly laterite gravel with an increasing dominance of clays in the east of the region (Williams & Mitchell 2001). The forest has a high species diversity, particularly around the granite outcrops where there are rapid changes in soil conditions over a small area (Williams & Mitchell 2001). The dominant land uses in the area are forestry (both native forest and plantations), conservation, grazing (improved pastures), cultivation (dry land agriculture) and mining (Williams & Mitchell 2001). There are growing areas of rural residential and urban land use (Williams & Mitchell 2001). Prescribed burning has been undertaken in the northern jarrah since 1953 (Underwood & Christensen 1981). Fire rotations for the jarrah forest are approximately every 5-7 years with the aim of keeping fuel loads below 8t/ha (Department of Conservation and Land 24 Background Management 2000; Burrows et al. 1999). Figure 2 shows the spatial extent of the regions, as well as the fuel age (time since last fire) of the forest (image courtesy of the Department of Conservation and Land Management, 2004). Figure 2: Map of the Swan and Southwest regions of the Northern Jarrah showing spatial extent of the regions and fuel age. The northern jarrah is a large region, and hence there is significant spatial variation. Forest structure, diversity, density and composition vary with rainfall and human land use impacts in the area (Williams & Mitchell 2001). By undertaking the study at such a large scale, considerable smoothing of these local variables will occur. Despite this, the northern jarrah is the best studied and most well understood of the forested regions in Western Australia, making it the most suitable region for study. 3.2. Fire Science 3.2.1. Causes of wildfire Wildfires occur naturally in the forests of Western Australia (Knox et al. 1999; Luke & McArthur 1986). Every year from about mid-October to mid-April the south west of Australia suffers a predictable drought, which dries the soil and vegetation. Fanned by strong dry summer winds, a small ignition can rapidly erupt into an intense wildfire as the fuel ignites (Department of Conservation and Land Management 2000). 25 Background The most frequent natural cause of wildfire ignition is lightning strikes, which occur frequently in the summer when the fuel bed is dry (CALM unpublished data 2002; Luke & McArthur 1986). There are approximately 15-30 lighting-caused wildfires in the northern jarrah every year (Burrows et al. 1999). The presence of man in Western Australia has had a significant effect upon the fire regime. The effects of logging, clearing and prescribed burning have all greatly affected the fire regime (Underwood & Christensen 1981). Of the six largest causes of fire in Western Australia, five of them are induced by humans (CALM unpublished data 2002). By far the largest cause of fire in Western Australia is deliberate ignition (Foster 1976; CALM unpublished data 2002). Arsonists tend to operate around settlements, (Kocsis & Irwin 1997; Rogerson & Sun 2001), and hence wildfires are often clustered around areas of human settlement. Two other ignition causes are significant. Accidental fires are caused by the timber and other industries, and also by recreational users of the forest. Escapes from prescribed burns, carried out by the department of CALM or by farmers, foresters or other industries also occur each year (CALM unpublished data 2002). 3.2.2. Factors Influencing Fire Intensity Once a fire has ignited, its rate of spread and intensity will be controlled by a combination of the local weather conditions, the fuel quantity and type and the topography (Beck 1995; Burrows 1984b; Burrows 1984a; Luke & McArthur 1986; Sneeuwajgt & Peet 1985; Clayton et al. 1987). All of these factors can become exceedingly complex depending on the depth and scale at which they are measured. This study focuses on the parameters used by the Department of Conservation and Land Management (CALM) to measure fire intensity (Beck 1995). The CALM model greatly simplifies the processes involved in a wildfire. However the equations were derived from experimental fires and yielded a good fit to the empirical data, suggesting that the simple 26 Background model is an effective means to obtain a first approximation of the intensity of a fire (Beck 1995). Fire weather factors: The Fire Danger Index There are many important weather variables affecting the intensity of a fire. These include temperature and season, relative humidity, precipitation, surface moisture content, wind velocity and wind direction (Clayton et al. 1987; Foster 1976; Luke & McArthur 1986). While there appear at first glance to be many weather variables affecting the intensity of the fire, it should be noted that many of these variables are not independent of one another (e.g. temperature and season, or temperature and surface moisture content), and it will often suffice to consider a reduced number of these variables in a model (Andrews et al. 2003). Fire managers often tend to aggregate a number of important weather variables into a single index. In Western Australia, the fire danger index (FDI) is constructed from wind speed and the surface moisture content (a function of time since last rain and the time of day) (Beck 1995). The empirical equations are rather complex, however for a standard northern jarrah forest, the FDI can be determined using Equation 1 (Beck 1995): Equation 1: The Fire Danger Index (FDI) FDI = Y + A exp(WIND * N ) where: FDI= fire danger index Y = 21.37 − 3.42SMC + 0.085SMC 2 A = 48.09 SMC exp(−0.60 SMC ) + 11.90 N = −0.0096 SMC 1.05 + 0.44 and WIND = wind speed (km/hr) at a standard height of 1-2m above the forest floor (application bounds 0-11.2km/hr) SMC = surface moisture content (application bounds 3-26%) The application bounds represent the conditions that were experienced during the experimental fires. If the equations are applied outside of these bounds, there is no 27 Background certainty that the fire will behave according to Beck’s equations (Beck 1995). This poses a problem for the prediction of extreme fire events, as these high intensity events may exceed the application bounds. However, developing empirical equations for extreme intensity fires is difficult and may be dangerous as these fires are difficult to suppress, and hence the behaviour of extreme fires under these conditions has not been adequately studied (Fernandes & Botelho 2003). Consequently this dissertation assumes that the equations of Beck (1995) are valid for the extreme fires, however it should be noted that the empirical equations have not been validated outside their bounds. There is a need to develop equations that are valid at extreme fire intensities, and this is a current priority for research (Fernandes & Botelho 2003). The FDI is derived for Western Australian forests, but is very similar to weather indices in other parts of the world. In New South Wales and the Australian Capital Territory, the very similar McArthur Forest Fire Danger Meter (Mark 5) is used (Noble et al. 1980; Vernon et al. 2004). The National Fire Danger Rating System is a similar, albeit more complex series of fire weather indices that describes the potential fire danger based upon the weather in the USA (Andrews et al. 2003). Topography The effect of topography on fire intensity can be highly complex. However topographical effects, like weather effects, are often not independent. For example, the topography affects the local wind speed (Beer 1991), vegetation structure (e.g. riparian vegetation in a river valley compared with vegetation on a hill slope), temperature (shading) and soil moisture content (greater in a river valley than on a hill slope) (Luke & McArthur 1986). Since this study does not attempt to capture the local spatial and temporal variability during a bushfire, the complex interactions between topography and the other variables are neglected. The CALM model assumes that topographical factors can be captured by one variable – the slope of the terrain (Beck 1995). The slope of the terrain affects the rate of spread of the fire, which in turn affects the intensity. Essentially fires become more intense as they 28 Background move upslope. This occurs because the flames lean over the fuel bed when the fire moves uphill, increasing the radiant heat intensity transferred to the fuel and increasing the amount of fuel that ignites (Burrows 1984b; Luke & McArthur 1986). Moving down slope, the fire is less intense as the flames lean away from the unburnt fuel bed, transferring less radiant heat and making ignition more difficult (Burrows 1984b; Luke & McArthur 1986). Given the broad spatial area over which this study is carried out, as a first simplifying step, this study assumes a flat topography and thus neglects the effects of topography in the intensity calculations. Fuel load Fuel load is directly related to fire intensity - the greater the available fuel load, the more intense will be the resulting fire (Luke & McArthur 1986; Clayton et al. 1987; Foster 1976). Whilst this is a trivial relation, it is not so simple to obtain the amount of fuel available to the fire. The amount of fuel available is determined by the weight and moisture content of the fuel bed (Beck 1995). Fuels can be broken into three classeslitter, trash fuel and scrub. Litter is the dead material including leaves, twigs and branches that regularly fall from the vegetation in the forest. Litter (t/ha) can be estimated from the canopy cover and the number of annual leaf falls since the last fire (Beck 1995; McCaw et al. 2002). Litter falls rapidly immediately after a fire, as dead material is shed by the vegetation. The rate of accumulation slows after 3-5 years, but litter can continue to accumulate for up to 70 years due to the low rates of decomposition in the jarrah forest, resulting in fuel loads in excess of 20t/ha (Foster 1976; Underwood & Christensen 1981). In the jarrah forest, fifty years of prescribed burning has ensured that the litter depth is less than 20mm, and hence all the litter is available to the fire (Beck 1995). If the litter depth is greater than this, the litter often has a significant moisture gradient which reduces the amount of fuel available to the fire (Luke & McArthur 1986). 29 Background Trash fuel refers to dead tree branches and scrub debris that contribute to the available fuel load. This component of the fuel is more common in karri fuel types than in jarrah (Beck 1995). Jarrah fuels aged less than 10 years do not carry a significant trash component (Beck 1995) and thus trash is not considered in this study. Scrub represents the flammable undergrowth in the forest that will contribute to the fuel load. This is a function of undergrowth height and density, and a correction factor representing the expected intensity of the fire (Beck 1995). This study assumes that the undergrowth is sparse and does not have a dominant impact upon the total fuel load as no information was obtained from CALM about scrub type at the historical fire events in the data set. Total available fuel quantity is obtained by simply summing the litter, trash and scrub components. Since trash and scrub fuel components are not included in the study, the total available fuel quantity is represented by the litter component of the fuel only. 30 Methods 4. Methods The study aims to demonstrate how quantitative risk assessment can be applied to wildfire management. The methodology was designed to be coherent with the Australian Standard for risk management (AS/NZS 4360:1995). This chapter is split into two sections to reflect this. The first section outlines the risk standard and shows how the study interpreted and applied the standard in a broad sense. The second section elaborates on the first, giving the specifics of the methods used. 4.1.1. The Australian Standard for Risk Management applied to Wildfire Management The Australian Standard for risk management (AS/NZS 4360:1995) provides a generic framework for the identification, analysis, assessment, treatment and monitoring of risk (Standards Australia & Standards New Zealand 1995). The main steps in the risk management process are shown in Figure 3 below. 31 Methods Figure 3: Steps in the risk management process (Australian Standard 4360:1995) There are five steps in this framework: Establish the context, identify the context, analyse the risks, assess the risk, and treat the risk. These five steps and how they are applied to wildfire management in this dissertation are outlined in the remainder of this section. 32 Methods Establish the context The first step involves establishing the strategic context in which the management organisation must operate. The aim of this step is to determine the elements of the environment that will impair or support the ability of the organisation to manage risk (Standards Australia & Standards New Zealand 1995). In the context of wildfire, this is a highly complex step. The Department of CALM have the difficult task of balancing the ecological fire regime requirements of the forest and managing fuel loads for the protection of life and property. There is a broad and diverse range of stakeholders including property owners and farmers at the urban fringe, forestry workers, tourists, conservation groups, and community groups. In addition, the Department of CALM is funded by the government of Western Australia, and hence must operate within financial and political boundaries. This dissertation does not intend to emphasise this first step of the risk management standard, as the responsibility of setting goals, strategies and objectives for the organisation is seen to lie with CALM. The literature review serves to provide an overview of the complexity of the environment and show some of the concerns that CALM currently consider. Identify risks Identification of the risks involves answering two questions – what can happen, and how (and why) can it happen? In the case of wildfire management, these questions are well studied. This dissertation simplifies “what can happen?” to the notion that the consequence of extreme wildfire is the loss of property. The effects of fire on biota and the risk posed to human life are not covered in this dissertation. The study considers the question of how a fire may ignite by analysing the probability of ignition due to a given cause. Historical fire records of the cause of ignition were used to create a probability mass function depicting the probability that a given fire started due to a given cause. 33 Methods Once a fire has ignited, the consequences of the fire are dependent upon the conditions in the ambient environment. This dissertation makes the assumption that the equations derived by Beck (1995) capture the variables that cause intense fires, and further assumes that an extreme fire will have detrimental consequences on an area of value. No account is taken of how transport processes or suppression efforts may affect the probability of a wildfire causing damage. Analyse risks Risk analysis seeks to determine the likelihood and consequences of damage as a result of fires. The aim of this step is to separate major and minor risks and provide a base for comparison with design standards (Standards Australia & Standards New Zealand 1995). Probability distribution functions of the controlling variables (Beck 1995) were created and the size of the original data set was increased using Monte Carlo simulation methods. Following these steps, risk was quantified by determining the return period of the extreme fire events. These steps are further described in sections 4.1.3-4.1.6. This process gives an indication of the likelihood of an extreme event occurring. The intensity of the extreme events can be compared to design standards to determine the consequences of the extreme events if they occur. In this way likelihood and consequence are combined to give a level of risk. Return period analysis is widely used in other fields for prediction and design (e.g. (Pattison et al. 1977; Plate 1982; Chow et al. 1988; Petrauskas & Aagaard 1971). Assess risks Risk assessment requires the existence of predefined risk criteria. The level of risk determined from the ‘analyse risk’ stage in the Australian Standard framework is compared against the risk criteria to determine whether the risk can be tolerated, or if mitigation is necessary (Standards Australia & Standards New Zealand 1995). Risks can also be prioritised at this stage in the process (Standards Australia & Standards New Zealand 1995). 34 Methods No predefined risk criteria exist for the acceptable wildfire intensity. However there has been considerable testing done on the temperature at which wood ignites, and limited study of the intensity at which wood ignites (Babrauskaus 2001). Based upon the ignition intensity of wood, a simple standard of 40kW/m2 was proposed (Babrauskaus 2001). This was used to compare against the return period of intense fires. If a wooden structure is to be designed for a given lifetime, it should be designed to ensure that the intensity of the fire with a return period corresponding to the project lifetime has an intensity that is less than the standard. The intensity versus return period plot from the ‘analyse risk’ stage of the Australian Standard was used to compare the intensity of the extreme fires to the design standard. Conclusions can then be made about whether the risk posed by extreme wildfires can be tolerated, or if mitigation is necessary. Treat risks If the intensity of the fire with a return period equal to the design lifetime is greater than the standard fire intensity, then mitigation is necessary. This dissertation considers only one mitigation option, which involves clearing the fuel load away from the asset to reduce the risk of ignition by radiant heat intensity. However it should be noted that there are many other mitigation options available, such as using fire retardant building materials or having a water supply close to the house (Australian Capital Territory Emergency Services Authority 2004; Braun 2002; Braun 2001). 4.1.2. Method overview The above section serves to demonstrate how the dissertation is coherent with the Australian Standard 4360:1995 for risk management. This section provides a very brief overview of the methods to orient the reader. The remaining sections in the chapter expand upon these steps and provide details of the methods used to generate results. MATLAB (version 6.1, release 12.1) was used to undertake the all of data analysis in the study. Scripts containing the code used to carry out the data analysis are included in Appendix 1 (section 10.1). 35 Methods The first step was to use the historical wildfire data to construct probability distribution functions for the key variables that lead to intense wildfires (Fire Danger Index, Fuel load and Surface Moisture Content). Probability distribution functions were also constructed for the annual number of fires and the cause of fire. These distributions were used to construct 100-year simulated records for each variable based upon the 17 years of historical fires using Monte Carlo techniques. The variables were next combined using the equations of Beck (1995) to create a synthetic 100-year record of fire intensity. The fire intensity values were ranked and extreme value series (annual exceedance and annual maximum series) were generated. The return period of the extreme fires was determined, and the parameters that led to the extreme events were examined to determine the critical parameters for management. A simple mitigation example was carried out to demonstrate the benefits that quantitative risk analysis can provide. In the example a wooden test pole is placed into the fuel load. The pole ignites at 40kW/m2, which is set as the design standard. A mitigation strategy of totally removing the fuel load around the pole to reduce the radiant heat intensity is applied to the scenario. How each of these steps was carried out is outlined in detail in the following sections. 4.1.3. Obtaining probability distribution functions Data was obtained from the Department of Conservation and Land Management (CALM) for 17 years of recorded wildfires in the Swan and Southwest regions of the northern jarrah forest of south-western Western Australia. Continuous data were available from 1980-1993 inclusive, and from 2000-2004 inclusive. No data were available for the period 1994-1999 (Wallace (personal communication) 2004). It is assumed that the fact that the data are not continuous in time does not affect the probability distribution. This is justified as the year from which the data are obtained should not affect the probability distribution functions from which the random sampling was undertaken. Making this assumption neglects variability in the distribution on a decadal scale. Given the importance of weather in determining the fire intensity, this may require further research 36 Methods – there is some evidence from eastern Australia that the El Nino phenomenon increases the likelihood of weather conditions conducive to high intensity fires (Vernon et al. 2004). These data were used to create probability distribution functions for the key variables contributing the fire intensity. A probability distribution function is a representation of the possible values that a variable can take, and a measure of the likelihood (probability) of these values occurring. When these probability distribution functions are plotted in a cumulative sense, there will be a unique probability associated with each value in the distribution. Cumulative probability distribution functions allow for the creation of simulated fires because they contain all possible values that the distribution can take with a unique probability of occurrence – selecting a random number between zero and one represents a unique value of a parameter on the distribution. Probability distribution functions for the annual number of fires, the fire danger index (FDI), fuel load, surface moisture content and the cause of ignition were derived from the data. The methods used to create the probability distribution functions are described in the remainder of this section. Annual Number of Fires The number of fires per year was used to obtain the probability of the occurrence of a given number of fires in a year. A fire was considered to have occurred only if the recorded burnt area was greater than 10ha. Thus the distribution is a measure of the number of fires that successfully ignite, given that an ignition has occurred. The annual occurrence of fire is assumed to follow a Poisson distribution, which makes the following assumptions (Kennedy & Neville 1986): • The probability of occurrence of a fire at a given point in time and space is very small, and this probability is constant. The fire will occur at random in time and within the domain. 37 Methods • • The probability of more than one fire occurring in each time subinterval is negligible. The occurrence of fires is independent- the occurrence of a fire does not influence the probability of other fires occurring. The assumption that the occurrence of fires is independent may be challenged in future work. For example, if a fire occurs, it will consume the majority of the fuel load in the area. Hence if a second ignition occurs in the same area soon afterwards, the probability of it exceeding 10ha is very small, as it will have little fuel to consume. Hence the occurrence of the first fire will influence the chance of success of the second fire, and the events are not independent. The system has a ‘memory,’ which is a phenomenon not observed in other extreme events (e.g. cyclone prediction) where this assumption has been made (Petrauskas & Aagaard 1971). However, incorporating this ‘memory’ requires an understanding of the spatial and temporal occurrence of fire, which is not yet quantified. The use of the Poisson distribution provides a useful first approximation to the number of fires per year until such an understanding is reached. The number of fires per year that exceeded 10ha was determined from the raw data. The mean number of fires per year (µ) was obtained and the probability of obtaining n fires greater than 10ha was assumed to follow the Poisson distribution (Equation 2): Equation 2: The Poisson Probability Distribution Function Pr( n _ fires ) = e − µ (µ ) n n! where n is the number of fires. The cumulative probability distribution function for n was obtained by summing all probabilities from the probability density function that had an n less than or equal to a given number of fires. 38 Methods The Fire Danger Index Fire Danger Index (FDI) data was converted to a probability distribution function. A number of frequency classes were created and the number of times that FDI values fitted into these classes was counted. The size of the frequency classes was set to 20 (i.e. the first interval was 0-20, followed by 21-40 etc). This frequency histogram was then used to create a cumulative histogram by summing the frequencies of all FDI classes less than or equal to (not exceeding) the FDI class in question. This cumulative histogram was normalised to a scale from zero to one (a probability scale) by dividing all frequencies by the total number of fires. Fuel Load Fuel load was converted to a probability distribution function in the same way as FDI. A cumulative frequency histogram was created and normalised to a scale of zero to one by dividing by the number of recorded fuel loads. CALM records the fuel age of a forest based upon the number of years that have elapsed since the last fire (prescribed burn or wildfire). Although fuel age is related to the total fuel load, it is not entirely the same variable. This was discussed in section 3.2.2. It should be noted that the fuel load represented in the probability distribution function is a measure of litter fuel, and does not include trash and scrub fuels, which would increase the total fuel load if included. Trash and scrub fuels were not included as no data on these parameters was supplied by CALM, and estimation of these fuels would be purely conjecture. 39 Methods Fuel load was derived from fuel age according to the equations included in Beck (1995) (Equation 3): Equation 3: Fuel Load L = (0.18CC + 11.06)(1.0 − exp(−0.086 * LF ) where L = weight of available leaf litter (t/ha) CC = percentage canopy cover (assumed to be constant 50%) LF = number of annual litter falls (years) It is noteworthy that the fuel age was not recorded for all fires recorded by CALM. Of the 4122 fires in the 17 years of data, fuel age was recorded for 1016 fires. Surface Moisture Content The Department of CALM do not keep a record of surface moisture content (Wallace (personal communication) 2004), as this variable is captured in the aggregate Fire Danger Index. However the equations used by CALM to predict fire intensity require the calculation of a fuel quantity correction factor where the fuel load differs from the standard 8t/ha in the jarrah forest (Beck 1995). This fuel quantity is exceeded in the probability distribution function, and hence calculation of the intensity requires the fuel quantity correction factor to be calculated. Surface moisture content varies from 3-26% in the northern jarrah forest, dependent upon the time since last precipitation and the time of day (Beck 1995). These bounds were used to generate a uniform probability distribution from which surface moisture can be randomly sampled. Because surface moisture content is included in the calculation of the Fire Danger Index (FDI), the FDI and the surface moisture content will not be independent in reality. However the Monte Carlo simulation technique assumes that all parameters are independent (Koller 2000). To show how this is a problem, consider an example where 40 Methods surface moisture content is randomly selected and found to be low. This means that we expect a high FDI and the resultant intensity of the fire to be high due to the dry fuel load. However the FDI is selected randomly from the distribution, and it is possible to select a low FDI which will result in a low intensity fire despite a low surface moisture content. The random selection process does not capture the dependence between the two variables. This concern highlights the need to record the raw variables contributing to the FDI. If these variables were known, it would be possible to create a more reliable probability distribution function for the surface moisture content, and more importantly, to eliminate the problem of dependence between the fuel quantity correction factor and the FDI. In this dissertation, the surface moisture content and FDI are assumed independent based upon the assumption that FDI is an aggregate of surface moisture content and wind speed. It is assumed that it is possible to obtain a high FDI due to a high wind speed, regardless of the value of the surface moisture content. However it is recommended that this assumption should be revised if the raw variables contributing to the FDI become available in the future. 41 Methods Cause of Fire The department of CALM records cause of fire in one of nine categories. These categories are given in Table 1 below. Table 1: Cause of Fire categories used by the Department of Conservation and Land Management Fire Cause Description 1 Deliberate 2 Escape from CALM prescribed burn 3 Escape from other burning off 4 Accidental by timber industry 5 Accidental by other industry 6 Accidental by recreational forest users 7 Lightning 8 Unknown 9 Other To create a probability mass function for the cause of ignition, the number of occurrences of wildfire due to each cause was obtained from the recorded wildfire data supplied by CALM. The number of fires due to each cause was then normalised to a scale between zero and one by dividing the number of fires due to each cause by the total number of recorded fires. This process yielded nine numbers, one representing each cause of wildfire. Each of the nine numbers represents the probability that, given that a fire has started, the fire was started by the cause that the number represents. These probabilities were plotted to create a probability mass function representing the cause of ignition. 4.1.4. Intensity Simulation: Monte Carlo techniques Monte Carlo simulation, also known as distribution sampling, is the process of using random sampling from distributions to approximate solutions to probabilistic or deterministic problems (Manno 1999; Hammersley & Handscomb 1964; Sobol 1974; 42 Methods Koller 2000). The process is useful as it allows simple deterministic relations to be used to simulate complex physical processes and generate probable results (Koller 2000; Manno 1999). The most common application of Monte Carlo methods involves determining the probability that a certain event will occur (Koller 2000). This is achieved by randomly sampling values from frequency distributions which represent the variables in the analysis (Koller 2000). Monte Carlo Simulation of Fire Intensity Monte Carlo techniques were applied in this study to generate a distribution of fire intensities. This section outlines how random sampling from the probability distribution functions was utilised to create a synthetic time series of wildfire intensities. The equations of Beck (1995) describe the relationships between fire intensity and the variables recorded by CALM. These equations are used in this study. Fire intensity is a function of the fuel load, the Fire Danger Index (FDI), and the fuel quantity correction factor (a function of fuel load and surface moisture content). The slope of the terrain also affects the fire intensity - this study assumes a flat domain and thus removes this variability. Beck’s equations are based upon empirical observations of a number of experimental fires (Sneeuwajgt & Peet 1985), and are deterministic – by themselves, they cannot give an indication of the likelihood of a fire occurring. Despite the relations between variables being deterministic, it is possible to create a quantitative probability of a fire of given intensity occurring by using probabilistic inputs. To do this, the probability distribution functions described in section 4.1.3 are used. This can be extended to a time series by repeating the process numerous times based upon the distribution of the annual number of fires. MATLAB’s random number generator was used to select a random number between zero and one. This number represents the probability that the number of fires in a year of simulation will exceed a value given by the cumulative probability distribution function for the number of fires. This process is repeated for the desired number of years of 43 Methods simulation to yield the number of fires occurring each year in a simulated period of record. For the purposes of this dissertation, 100 years of fire data was simulated. Having simulated the number of fires occurring per year, the next task is to determine the characteristics of each fire. There are three variables that need to be determined – fuel load, FDI, and surface moisture content. Cumulative probability distribution functions for these variables were determined from the data supplied by CALM (see section 4.1.3). For each of the three variables, random numbers representing the probability of occurrence of the variable were generated and the corresponding value of the variable was determined from the cumulative probability distribution function. This process was repeated for each fire in the simulated record. The result is a 100 year time series of fires, with each fire event specified by a fuel load, FDI and surface moisture content. The simulated data are finally combined using the equations of Beck (1995), yielding 100 years of simulated fire intensity data. Calculating Fire Intensity from the CALM parameters Fire intensity is a function of fuel load, FDI and surface moisture content. The relation between these variables and fire intensity is given by Beck (1995). Fire intensity is given by Equation 4 (Muller 1993; Luke & McArthur 1986): Equation 4: Fire Intensity Equation (General Form) I = HwR where I = intensity of the fire in kW/m (measured along the head fire front) H = heat of combustion (kJ/kg) w = the weight of fuel in kg/m2 R = rate of spread of the head fire front (m/sec) This relation can be simplified by assuming a value of 16 920 kJ/kg (Muller 1993; Commonwealth Government of Australia 1984) for the heat of combustion. This yields Equation 5 (Muller 1993): 44 Methods Equation 5: Fire Intensity Equation Adjusted to Include the Heat of Combustion I = 0.47 wR where I = intensity of the fire in kW/m (measured along the head fire front) w = the weight of fuel in tonnes/ha R = rate of spread of the head fire front (m/hr) The rate of spread of the head fire front is derived from Beck’s (1995) equations. Neglecting the effects of slope, the rate of spread is given by Equation 6 (Beck 1995): Equation 6: Rate of Head Fire Spread R = FDI × FQCF where R FDI = Rate of spread of the head fire front (m/hr) = Fire Danger Index FQCF= Fuel quantity correction factor 45 Methods Data for the Fire Danger Index was supplied by the department of CALM. However no fuel quantity correction factors were supplied and these had to be calculated using Beck’s piecewise-defined equations (Equation 7) (Beck 1995): Equation 7: Equations for the Fuel Quantity Correction Factor FQCF = 1.02 + 0.10 1 + 7266.83 exp(−1.36 AFQ) FQCF = 6.03 + 5.81AFQ 53.44 FQCF = FQCF = 11.19 + 2.92 AFQ 35.02 0.055 + 0.0023 AFQ 0.074 2.5<AFQ<8.0; 3.0<SMC<26.0 8.1<AFQ<25.0; 3.0<SMC<9.0 8.1<AFQ<25.0; 9.1<SMC<18.0 8.1<AFQ<25.0; 18.1<SMC<26.0 where FQCF = Fuel Quantity Correction Factor AFQ = Available fuel quantity (tonnes/ha) SMC = Surface moisture content (%) Available fuel quantity can be calculated from the data supplied by CALM (see section 4.1.3). Surface moisture content was not recorded by CALM and hence was sampled from a uniform distribution (see section 4.1.3) with a range of 3-26% based upon the above equations. The 100 years of simulated data is combined according to the above equations to yield the simulated 100-year fire intensity record. 4.1.5. Selecting the Extreme Fires The large intensity fires are the most likely to cause damage, and hence these are the fires that most need to be managed. Hence it is useful to determine the extremes of the simulated fire intensity distribution. 46 Methods Extreme fire intensities can be selected from the complete record in two major ways. Selecting all fires with an intensity greater than a predefined value will yield the partial duration series (Chow et al. 1988). If the number of fires selected is fixed to equal the number of years of record, the series is called the annual exceedance series (Chow et al. 1988; Pattison et al. 1977). As an alternative to the partial duration series, the largest fires in a predefined subset of the series can be selected - this is the extreme value series (Chow et al. 1988). If the largest fire in each year of record is selected, the series is called the annual maximum series (Chow et al. 1988; Pattison et al. 1977). Annual maximum and annual exceedance series were determined from the 100 years of simulated record by selecting the largest fire in each year of record, and by ranking the data and selecting the 100 largest intensity fires respectively. 4.1.6. Return Period Analysis: Wildfire risk Risk analysis requires an understanding of the likelihood and consequences of wildfire occurrence. The severity of the consequences of a wildfire is measured by its intensity (Luke & McArthur 1986). Hence from a risk perspective, it is important to know how frequently extreme intensity fires will occur. This can be quantified by determining the return period of the fires in the extreme annual maximum and annual exceedance distributions. The return period of an event of a given magnitude is defined as the average recurrence interval between events equalling or exceeding a specified magnitude (Chow et al. 1988). Return period is inversely proportional to the probability of exceedance (the probability that a fire will have an intensity greater than or equal to the intensity specified) (Chow et al. 1988; Pattison et al. 1977). 47 Methods Return period analysis is based on the demonstrated fact that the magnitude of an extreme event is inversely proportional to its frequency of occurrence. The data in the annual maximum and annual exceedance distributions were ranked in order of decreasing intensity. The return period of each intensity is then given by Equation 8 (Chow et al. 1988; Kennedy & Neville 1986; Pattison et al. 1977): 48 Methods Equation 8: Return Period Tm = where T N +1 m = Return period of fire (yrs) N = Number of years in record (yrs) m = Ranking of the fire intensity in the distribution The return period for each fire in the annual maximum and annual exceedance distributions was determined and plots of return period against fire intensity generated. This provides a visual representation of the risk of wildfire - it shows the statistical average interval between events of the given intensity. 4.1.7. Contribution of the parameters to extreme fire events The simulated parameters (fuel load, FDI and surface moisture content) that led to the extreme wildfire intensities in the annual maximum series were plotted against return period. This gives an indication of the dominant factors leading to high fire intensities. Knowledge of the driving parameters will ensure that mitigation is directed effectively. 4.1.8. Spatial Considerations: Effect of the size of the study area The methods presented in this dissertation up to this point have been derived from data collected across the entire northern jarrah forest. In order to apply the results in practice, it is necessary to understand how the results will alter if scaled down to a specific area within the jarrah forest. Before the results could be applied to the proposed mitigation example, a method to apply the results to a specific point had to be derived. The Poisson distribution applied to simulate the number of fires occurring in each year assumes that the probability of a fire occurring in time and space is constant and occurs at random (Kennedy & Neville 1986). Hence the parameter of the distribution represents the expected mean spacing between fires in the study area over the period of record. It is thus expected that if the size of the study area is reduced, the number of fires occurring in 49 Methods the reduced area will decrease. Since the probability of occurrence is constant, it is assumed that the parameter will decrease in proportion to the relative study area size. In order to bring the size of the study area into the simulation, the mean value used in the Poisson distribution was altered according to Equation 9: Equation 9: Parameter of the Poisson distribution with correction for size of study area µ= where µ N Ai × T A = parameter of the Poisson distribution (#fires/yr) N = number of fires in record T = number of years in record (yrs) Ai = area of sub-area of interest (ha) A = total area of northern jarrah forest (2 255 904 ha) (Williams & Mitchell 2001) The effects of this change were demonstrated on a sub-area of half the size of the northern jarrah forest (Ai =1 127 952ha), and on a single point in the study area. To assess the effect of the change on a point, it was assumed that the smallest independent area within the total area was equivalent to the area of the largest fire in the record. That is, it is assumed that a point in the jarrah forest can be affected by any fire that occurs within an area less than or equal to the area of the largest fire that can occur. The largest recorded fire area in the CALM data is 18 042ha. 4.1.9. Application: Mitigation The return period of extreme wildfires is particularly useful when applied to mitigation design. For example, if a structure is to be designed to last 100 years, the intensity of fire that it should be built to withstand can be determined from the 100-year return period fire intensity. Standards can be set that ensure that fire damage does not occur. This part of the analysis required the results from the whole northern jarrah forest to be scaled down to a point using the methods outlined in section 4.1.8. 50 Methods A simple example of mitigation design was undertaken in which a wooden pole is placed into the fuel load. Wood ignites at a radiant heat intensity of 18-60kW/m2 dependent upon the type of wood, exposure time and other conditions of the ignition test (Babrauskaus 2001). Based upon these values, a tentative standard of 40kW/ m2 was proposed as a design target. A mitigation strategy of completely removing the fuel load around the pole was tested to determine the necessary cleared distance to reduce the fire intensity of the 100-year fire to a level below the design standard and thus prevent ignition by radiant heat intensity. It was assumed that fire intensity decreases according to Equation 10: Equation 10: Radiant Heat Intensity Decay Equation I (r ) = where I r I ( 0) r2 = intensity of the fire (kW/m) = cleared distance from the object of value (pole) (m) It may be noted that the units for head fire intensity are measured in kW/m, while the units for the design standard are the more conventional kW/m2. Head fire intensity is measured in units of kW/m2 along one metre of the fire front, hence kW/m (Muller 1993). For the purposes of the mitigation example, it is assumed that the problem can be thought of in 1-D. This means that the fire propagates to the pole as a point source, and hence it has no dimensions along the fire front. The intensity of the fire is thus measured in units of kW/m2, and can be directly compared with the standard. Another pertinent point is that this mitigation example greatly simplifies the risk of ignition by assuming that the asset will ignite due to radiant heat intensity alone. While radiant heat intensity is a concern, the majority of buildings ignite as a result of flying embers settling on the roof (New South Wales Rural Fire Service 2003) This should be kept in mind when interpreting the results. 51 Results 5. Results 5.1.1. Probability Mass function: Cause of fire A probability mass function was created using the 17 years of historical fire data provided by the Department of CALM. Figure 4 shows the probability that, given that a fire has started, it was started by the cause given on the x-axis. Figure 4: Probability mass function illustrating the probability that ignition occurred due to a given cause The figure shows that if a fire starts, the most likely cause is ‘deliberately lit’ (45%). This cause is considerably larger than any other, being followed by lightning (11%) and escapes from burn offs (10%). It is noteworthy that the ignition classes break down the cause of fire into different types of accident and separates escapes from burn offs into different groups, which makes the deliberately lit category appear more dominant than it actually is. However even if all accidental fires and all escapes from burn offs are classed together, they are still considerably less than the deliberately lit probability. In this case the probability a given fire starts due to an accidental cause is 12%, and the probability of a fire starting due to an escape from a burn off is 14%. 52 Results Unknown causes of fire make up 14% of the total record. 5.1.2. Probability distribution functions: Annual Number of Fires, FDI, Fuel Load and Surface Moisture Content Probability distribution functions for the annual number of fires with a burnt area exceeding 10ha, the Fire Danger Index (FDI), fuel load and surface moisture content were (a) (c) constructed. These are shown in Figure 5 below. (b) (d) Figure 5: Probability distribution functions for (a) Number of fires with a burnt area greater than 10ha; (b) Fuel load (t/ha); (c) Fire Danger Index; and (d) Surface Moisture Content (%). 53 Results The probability distribution function for the number of fires with a burnt area exceeding 10 ha (Figure 5a) is Poisson distributed. The mean value is 40.33 fires per year, meaning that on average 40.33 fires will occur that result in a burnt area of greater than 10ha. The maximum number of fires in the 17-year record with a burnt area greater than 10ha is 106 fires. Hence the largest number of fires in any year of the simulation is 106 fires. The probability distribution function for fuel load (Figure 5b) appears to be roughly exponentially distributed. The mean of the distribution is 7.79 t/ha. This value compares well with the CALM management goal which aims to maintain fuel loads between 68t/ha (Beck 1995; Underwood 1988). The standard deviation of the distribution is 7.89t/ha. The probability of ignition occurring with zero fuel is quite large (0.25). If ignition occurs with zero fuel, the resulting fire will have zero intensity. Figure 5b also shows that there is zero probability of a fire occurring in fuels of 1t/ha or 4t/ha. This is unlikely to be physically the case, and it is likely that these figures have been rounded down to 0t/ha and up to 5t/ha respectively by the data collectors. A significant peak (probability 0.115) occurs in the fuel load (Figure 5b) at 12t/ha. This point does not follow the general decreasing trend with increasing fuel load. Another noteworthy feature of the Figure 5b is that the range of the fuel loads is quite limited. Although large fuel loads exist, extreme values are not present and the distribution does not have a long tail. Figure 5b contains some significant assumptions. In the original data set, there were six data points with recorded fuel ages (years since last burn) of greater than 30 years. Examination of these extreme fuel loads showed that these fires corresponded to very small burnt areas (e.g. in a 70 year fuel load, the recorded burnt area was 7ha). Consultation with the Department of CALM revealed that these fuel ages were highly 54 Results unlikely given the 7 year prescribed burning rotation cycle for the northern jarrah (Sneeuwajgt (personal communication) 2004). It is likely that these extreme values were entered incorrectly into the data set (Wallace (personal communication) 2004) and did not physically occur. Consequently the six extreme fuel age values were dropped from the data set. This action has an effect on subsequent results as extreme fuel loads are likely to correspond to extreme intensity fires. The maximum fuel age in Figure 5b is 30 years, corresponding to a fuel load of 19t/ha. The other major assumption in Figure 5b is that the canopy cover at all areas in the northern jarrah is 50%. The canopy cover is used to determine the fuel load from the fuel age (Equation 3) (Beck 1995). This equation has been given application bounds from 20% to 80% (Beck 1995). Canopy cover varies throughout the forest and no information on this parameter was available from CALM. Hence a mean canopy cover of 50% was taken from Beck’s application bounds and this was used to calculate the fuel load. The sensitivity of the fuel load and intensity to extreme values of this parameter was also tested. Figure 5c depicts the probability distribution function for the Fire Danger Index (FDI), a measure of the weather conditions during wildfires. The distribution displays an exponential decay in probability of occurrence with increasing FDI. The majority of wildfires occur at low FDI values. The mean value of the distribution corresponds to an FDI of 188. The standard deviation of the distribution of the distribution is 317. Sixtyeight percent of wildfires occur on days with a FDI of less than 100, and 90% of wildfires occur on days with an FDI less than 500. Despite most of the wildfires occurring on days with a relatively low FDI, Figure 5c shows that there is a significant probability of very extreme weather conditions coinciding with wildfire events. The probability of a wildfire occurring on a day with a FDI greater than 1000 is 0.03, and the probability of exceeding a FDI of 2000 is 0.005. Although these probabilities are small, the selection of such an extreme FDI is likely to lead to a very extreme intensity. To put this into context, consider that there is an average 55 Results of 40.33 wildfires per year (Figure 5a), and that the record is to be extended from 17 years to 100 years. Using these probabilities, the expected number of fires occurring on a day when the FDI exceeds 1000 is 121 (statistically, one occurrence every 10 months), and the number of fires with a FDI exceeding 2000 is 20 (statistically, one occurrence every 5 years). Hence over large time periods such as 100 years, there is a significant chance that extreme weather conditions will coincide with wildfire events. Figure 5d shows the probability distribution function for the surface moisture content. As described in section 4.1.3, surface moisture content was sampled from a uniform distribution. Surface moisture content ranges from 3-26% (Beck 1995), resulting in a uniform probability of 1/23 (~0.043). The expected value of the distribution is 14.5% and the standard deviation is 6.6%. 5.1.3. Extreme value distributions - the Annual maximum and Annual Exceedance Series Figure 6 shows the return periods of extreme intensity fires for 100 years of simulated record. Figure 6a shows the annual maximum series, and Figure 6b shows the annual exceedance series (see section 4.1.5). Fifty simulations were undertaken. Note that both figures represent the same simulations. Each simulation is represented as a dot on the figure, and the mean of all fifty simulations is depicted by the solid green line. The vertical spread of the dots gives an indication of the variation in the mean intensity for a given return period. 56 Results (a) Annual Maximum Series (b) Annual Exceedance Series Figure 6: Intensity (kW/m) vs. Return period (yrs) for 100 years of record. (a) The plot shows the annual maximum series, which includes the largest intensity fire from every year of record. (b) The plot shows the annual exceedance series, which includes the 100 largest fires in the record. Note the difference in the intensities of the low return period fires between the two figures. In both the annual maximum and annual exceedance series, the variation in the simulated values increases dramatically with increasing return period. This suggests that the intensities of the large return period extreme fires are highly variable and prediction of these events is more difficult than prediction of the low return period fires. Figure 6a and Figure 6b have the same mean values for return periods of greater than 12.5 years, and follow a similar trend for return periods greater than 10 years. The 100 year fire event has a mean intensity of 34 108 kW/m. There was significant variation in this value, which fluctuated between 54 861 kW/m and 24 045 kW/m over the 50 simulations. The annual maximum series (Figure 6a) differs significantly from the annual exceedance series (Figure 6b) for return periods of less than 10 years. The annual maximum series has intensities that are considerably lower than the annual exceedance series for these return periods. The mean intensity of the 1-year fire is 1086 kW/m in the annual maximum series and 5667 kW/m in the annual exceedance series. This difference in 57 Results intensity is due to the way in which the intensities leading to the annual maximum series and annual exceedance series are selected and is discussed further in section 6.2.5. CALM have a rule of thumb which states that fires of intensity greater than 3000kW/m are too intense to attempt to control with direct attack methods (Muller 1993). The 100 year fire event has a mean magnitude of just over ten times this value, suggesting that the 100 year fire event will be very severe and uncontrollable. The 3000 kW/m threshold can also be compared to the 1 year fire intensities. Intensities exceed 3000kW/m with a return period of 1.12 years in the annual maximum series, and with a return period of less than 1 year in the annual exceedance series. This suggests that, statistically, a fire will occur that cannot be controlled with direct attack somewhere in the northern jarrah forest approximately every year. 5.1.4. Contribution of the fuel load, the FDI and the surface moisture content The fire intensity values in Figure 6 were created by sampling from the probability distribution functions for fuel load, Fire Danger Index (FDI) and surface moisture content. The values of these parameters that contributed to the extreme fires plotted in Figure 6 were plotted against the return period of the extreme intensity fires. This allows the trends in each parameter to be examined, and allows conclusions to be made about the influence that each parameter has on the fire intensity of extreme events. Figure 7, Figure 8 and Figure 9 show the fuel loads, FDI, and surface moisture contents respectively that contributed to the extreme wildfires. As in Figure 6, the green line represents the mean of a 50 simulations, and the parameter selected for each simulation is represented as a dot on the plot. 58 Results (a) Annual Maximum Series (b) Annual Exceedance Series Figure 7: Fuel load (t/ha) vs. return period for 100 years of simulation. These fires are the same fires shown in Figure 6. (a) shows the contribution of fuel loads to the extreme intensity fires in the annual maximum series in Figure 6a. (b) shows the contribution of fuel loads to the extreme intensity fires in the annual exceedance series in Figure 6b. Figure 7 shows that the mean fuel load increases with increasing fire intensity, regardless of whether the annual maximum or annual exceedance series is used. In both series, the 100-year fire intensity is driven by a mean fuel load of 16.64t/ha, and the 1-year fire intensity is driven by a mean fuel load of 12.64t/ha (from the annual exceedance series) or 11.58t/ha (from the annual maximum series). Although both of these fuel loads are above the 6-8t/ha fuel load that CALM attempts to maintain in the northern jarrah (Underwood 1988; Beck 1995), most of the extreme fuel loads lie within one standard deviation of the mean fuel load (7.8t/ha). One standard deviation above the mean corresponds to a fuel load of 15.68t/ha. This fuel load has a return period of close to 50 years (annual exceedance series). Hence although all fuels contributed to extreme fires, they are not at the extreme end of the fuel load distribution as they lie within one standard deviation of the mean. The exceptions to this statement are the 50 (15.84t/ha) and 100 year fire events (16.64t/ha), which lie just above one standard deviation of the mean. 59 Results Figure 8 below shows the contribution of the FDI to the extreme fire events in Figure 6. Both the annual exceedance (Figure 8a) and annual exceedance series (Figure 8b) are shown. (a) Annual maximum Series (b) Annual Exceedance Series Figure 8: Fire Danger Index (FDI) vs. return period for 100 years of simulation. These fires are the same fires shown in Figure 6. (a) shows the contribution of FDI to the extreme intensity fires in the annual maximum series in Figure 6a. (b) shows the contribution of FDI to the extreme intensity fires in the annual exceedance series in Figure 6b. Figure 8 shows that as intensity increases, FDI also increases dramatically. This demonstrates a strong correlation between the weather conditions and the occurrence of extreme wildfire events. The FDI of the 100 year fire event is 2310, and the FDI of the 1 year event is 868 (annual exceedance series) or 240 (annual maximum series). From the probability distribution function (Figure 5c) the mean FDI when a bushfire occurs is 188. This value is considerably less than the extreme values in Figure 8. The standard deviation of the FDI distribution is 317, so the mean annual exceedance series 1-year fire event is over 2 standard deviations above the mean FDI conditions, and the 100-year fire event is 6.7 standard deviations above the mean. The annual maximum series FDI exceeds 1 standard deviation above the mean with a return period of 5 years. Thus almost all of the FDI values in Figure 8 are in the extreme tail of the FDI probability distribution function and 60 Results can be classified as extreme weather conditions. This shows that extreme fire intensities correspond to extreme FDI (weather) conditions. It is worth emphasising that the extreme FDI values are considerably more extreme (up to 6.7 standard deviations) than the corresponding fuel loads, which are rarely outside 1 standard deviation. This suggests that extreme intensity wildfires are driven not by extreme fuel loads, but rather by the weather conditions (FDI) on the day. It should be noted that this analysis is based upon the extremes of the intensity distribution. Hence although it appears from Figure 7 and Figure 8 that the fire weather and not the fuel load are dominating the fire intensity, this needs to be considered in context. At low fuel loads, the fire intensity will be limited by the amount of fuel (Fernandes & Botelho 2003). Hence even if the worst weather conditions existed, if there was no fuel, there could not be a fire (Equation 4). What this analysis is showing is that fuel load ceases to be a dominant factor in the extreme fire intensity events – it does not show the effects of fuel in low and moderate intensity fires. Figure 9 shows the effect of surface moisture content on the extreme intensity fires in Figure 6. Both the annual maximum (Figure 9a) and the annual exceedance series (Figure 9b) are shown. 61 Results (a) Annual maximum Series (b) Annual Exceedance Series Figure 9: Surface moisture content (%) vs. return period (yrs) for 100 years of simulation. These fires are the same fires shown in Figure 6. The plots show the contribution of surface moisture content to the extreme intensity fires in (a) the annual maximum series in Figure 6a and (b) the annual exceedance series in Figure 6b. Figure 9 shows that the surface moisture content of the extreme intensity fires is restrained to a narrow band of surface moisture contents. The 100-year surface moisture content is 10.6% and the 1-year surface moisture content is 14.0% (annual exceedance series) or 15.1% (annual maximum series). There is a weak decreasing trend as the intensity of the fire increases. The decreasing trend suggests that as the fuel load becomes drier, the intensity of the fire increases. The expected value of the uniform distribution from which the surface moisture content was sampled is 14.5%. Since the distribution has a standard deviation of 6.6%, an extreme surface moisture content can be described as a moisture content of less than 7.9% or greater than 21.1%. None of the mean surface moisture contents in Figure 9 are outside one standard deviation of the mean. The annual maximum and annual exceedance series show similar trends for high return periods for all three parameters. At low return periods the annual maximum series has an 62 Results appreciably lower fuel load (Figure 7) and fire danger index (Figure 8) than the annual exceedance series. There is no appreciable difference in the surface moisture content (Figure 9) between the two series. 5.1.5. Testing the Effects of Extreme Fuel Loads The results in Figure 7 showed that the largest intensity fires rarely coincide with extreme fuel events (i.e. when fuel load recorded at the fire event is greater than one standard deviation from the mean). This was an interesting result and raised the question - what is the effect of extreme fuel events? To answer this question, it was necessary to vary the probability distribution function for fuel load to include extreme fuel loads. The fuel load increases with increasing canopy cover and increasing fuel age according to Equation 3. A sensitivity analysis of this equation was carried out to determine the relative contribution of these parameters to the total fuel load. The results of this analysis are shown in Figure 10 below. Figure 10: Sensitivity of the fuel load to canopy cover and fuel age. Fuel load (t/ha) is plotted on the vertical axis and fuel age (yrs) on the horizontal axis. The blue, green and red lines represent canopy covers of 10%, 50% and 100% respectively. Figure 10 shows that increasing the canopy cover increases the fuel load for all fuel ages. This was expected, since a greater canopy cover implies that a greater weight of leaf litter will fall per unit time, and hence the fuel load will increase with increasing canopy cover 63 Results (McCaw et al. 2002). The results presented in Figure 6-Figure 9 all assume an average canopy cover of 50%, however it is possible to obtain canopy covers of 10-80% (Beck 1995). Consequently there will be significant variation from the mean results presented in Figure 6-Figure 9 based upon the local canopy cover. One of the interesting features of Figure 10 is that fuel load ceases to increase after 70 years (McCaw et al. 2002; Underwood & Christensen 1981), and that the rate of accumulation of fuel decreases dramatically with increasing time since fire, The fuels shown in Figure 10 reach half of their initial accumulation rate at 10, 15 and 18 years for 10%, 50% and 100% canopy covers respectively. For the 50% canopy cover case, the increase in fuel load from a 30 year old fuel to a 100 year old fuel is just 1.6t/ha. Even in the 100% canopy cover fuel, the change in fuel load from a 30 year old fuel to a 100 year old fuel is only 2.4 t/ha. The results of this sensitivity analysis were used to generate maximum and minimum fuel loads and determine the effects of these extreme fuels on the extreme fire intensities. Jarrah forests in the south-west of Western Australia can accumulate fuel for up to 70 years (McCaw et al. 2002; Underwood & Christensen 1981). Hence a maximum fuel age of 70 years was substituted into Equation 3 to derive a maximum fuel load. Canopy cover was assumed to be 50% in the previous simulations, but in theory the parameter can vary from 0-100%. A maximum fuel load will occur when this variable is set to 100%. Beck’s (1995) equations were applied to a minimum canopy cover of 10%, so this minimum value was used when testing the fuel load sensitivity. To derive a new probability distribution function for fuel load, the canopy cover was set to a constant value of 100%, and the existing fuel age data (0-30 years) was substituted into Equation 3 to derive a new probability distribution function for fuel load. An exponential function was then fitted to this probability distribution function and the curve was extrapolated to a fuel load that corresponds to a 70 - year fuel age. The results of the 64 Results curve fitting and extrapolation of the cumulative probability distribution function are shown in Figure 11 below. Figure 11: Exponential curve fit to cumulative probability distribution function for fuel load with canopy cover set to 100%. The curve has been extrapolated to include fuel ages up to 70 years. Figure 11 shows that increasing the fuel age does not result in a considerable increase in the fuel load. Increasing the maximum fuel age from 30 to 70 years only increases the fuel load from 26t/ha to 30t/ha. The reason that the increase is so slight is that fuel load follows a logarithmic curve (Equation 3), with a rapid initial increase in fuel load that levels off as fuel age increases. Hence, after the first 30 years, the increase in fuel load is considerably slower than during the first 30 years. Figure 11 shows considerably higher fuel loads than those in the fuel load probability distribution function (Figure 5b). This is driven by the increase in canopy cover from the assumed 50% cover to the theoretical maximum canopy cover (100%). The fuel load is highly sensitive to the canopy cover. Thus it is worth determining the range of fuel loads and intensities of the extreme fire events due to different canopy covers. Figure 12a shows the theoretical maximum fire intensity versus return period based upon the probability distribution function in Figure 11. Figure 12b gives an idea of the minimum expected fire intensity based upon the probability distribution function of fuel load for fuel ages from 0-30 years (Figure 5b) and a canopy cover of 10% (minimum 65 Results value of canopy cover observed in Beck (1995)). Both figures are derived from the annual exceedance series. (a) 100% Canopy cover Fuel age 0-70 years (b) 10% Canopy cover Fuel age 0-30 years Figure 12: Fire Intensity (kW/m) and return period (yrs) for extreme intensity fires (annual exceedance series) showing the effect of varying fuel load (canopy cover and fuel age). (a) shows extreme maximum conditions- 100 % canopy cover and fuel age 0-70 years. (b) depicts minimum conditions- 10% canopy cover and fuel age 0-30 years. The percentage of canopy cover seems to be a very significant parameter determining the intensity of extreme wildfire events based upon the equations of Beck (1995). The mean intensity of the 100-year fire event varies from a maximum of 80 455kW/m in Figure 12a, (with a fuel load of 27t/ha, FDI of 2272, and a surface moisture content (SMC) of 9.6%) down to a minimum of 15 493kW/m in Figure 12b (fuel 10.5tha, FDI 2534, SMC 12.7%). The mean intensity of the 1-year fire event varies from a maximum of 10 707kW/m, (with a fuel load of 19.9t/ha, FDI of 755, and SMC of 14.0%) down to 2247kW/m, (with a fuel load of 8.54t/ha, FDI 666, SMC 12.8%). There is thus considerable variation in all intensity as a result of fuel loads with variation in canopy cover. The effect of varying the canopy cover and fuel age had a significant effect on the intensity of wildfires. To verify that the intensity was being driven by increased fuel load, 66 Results the fuel loads of the extreme intensity fires in Figure 12 were plotted. Figure 13 shows the fuel age of the most intense fires of return period plotted against return period. Figure 13a shows the fuel age for the theoretical maximum fire (100% canopy cover and fuel age varying from 0-70yrs). Figure 13b shows the fuel loads of the extreme intensity fires resulting from the minimum canopy cover (10%) and the fuel ages in the CALM data set (0-30yrs). (a) 100% Canopy cover Fuel age 0-70 years (b) 10% Canopy cover Fuel age 0-30 years Figure 13: Fuel load (t/ha) and return period (yrs) for extreme intensity fires (annual exceedance series) showing the effect of varying canopy cover. (a) shows extreme maximum conditions: 100% canopy cover and fuel age 0-70 years. (b) depicts minimum conditions: 10% canopy cover and fuel age 0-30 years. There is a marked increase in fuel load with increasing canopy cover and fuel age (Figure 13a). Similarly there is a marked decrease in fuel load with minimum observed fuel conditions (Figure 13b). Figure 13 verifies that fuel load is highly sensitive to the input parameters canopy cover and fuel age. However based upon the cumulative probability distribution function in Figure 11, canopy cover appears to be the driving parameter rather than fuel age. 5.1.6. Correlation between the variables The correlations between some of the key parameters in the CALM data set were examined for trends. Plots were created to illustrate the relationships between the FDI, 67 Results area burnt and fuel age. Surface moisture content was not correlated to the other variables as it was randomly sampled from a uniform distribution, and hence will not be related to the other variables. The results from the correlations are shown in Figure 14. The correlation coefficient between the two data sets was also calculated to give a quantitative indication of the relationship between the two variables. The magnitude of the correlation coefficient ranges from zero to one – the larger the magnitude of the correlation coefficient, the stronger the relationship between the variables. A negative correlation coefficient suggests an inverse relationship between the variables. The correlation coefficient relating the two variables in each plot are given in the lower left hand corner of each plot in Figure 14. 68 Results (b) (a) (c) r = 0.0626 r = 0.0748 r = -0.1690 Figure 14: Plots showing the correlation between some of the variables used to determine fire intensity: (a) shows the relationship between area burnt (ha) and fire danger index (FDI); (b) shows the relationship between area burnt (ha) and fuel age (yrs); (c) shows the relationship between fuel age (yrs) and FDI. Correlation coefficients are given in the lower right hand corner of each figure. Figure 14 shows that there is little relationship between any of the variables plotted. This is verified by the correlation coefficients, all of which have insignificant magnitudes. The fact that none of the variables are strongly related suggests that FDI, fuel load and burnt area are independent, a necessary condition for random Monte Carlo simulation. 69 Results It is an interesting observation that the burnt area is not correlated to FDI and fuel load. Since FDI and fuel load are directly related to fire intensity, it was expected that there would be a positive correlation between these variables and the area burnt. 5.1.7. Effect of the size of the study area To account for the change in study area size when applying the results from the whole northern jarrah region to a smaller area, it was necessary to vary the mean parameter in the Poisson distribution representing the number of fires. The effects of this change were tested on a sub-area of half the size of the northern jarrah forest, and on a single point in the study area. The results of these trials are plotted in Figure 15 below. The blue, green and red lines show the intensity vs. return period determined using the whole forest area, half of the forest area, and a point in the forest respectively. The plots are mean intensity plots of the annual exceedance series. Figure 15: Effect of study area on intensity versus return period plot. For a given intensity, smaller study areas correspond to greater return periods. Hence decreasing the size of the study area decreases the probability of extreme fires occurring. The 100-year fire events were 31 309kW/m, 26 902kW/m and 22 485kW/m for the whole study area, half study area and point fire respectively. 70 Results The return period of the 3000kW/m uncontrollable fire is 3.5 years at a point in the forest. This is larger than the corresponding return period (< 1 year) if the entire forest is taken as the study area, reflecting the fact that it becomes more likely that an extreme fire will occur within the domain as the size of the study area increases. 5.1.8. A risk-based standard for mitigating the risk of ignition due to radiant heat Australian Standard 4360:1995 for risk management requires that risk be assessed, and if necessary, treated. To illustrate how these steps might be implemented, a simple risk assessment was carried out on a wooden test pole placed in the fuel load. Wood ignites at an intensity of 18-60kW/m2 (Babrauskaus 2001), and so a mean intensity of 40kW/ m2 was set as a design standard. A mitigation strategy of totally clearing the fuel load around the fuel load was employed in the theoretical example. It was assumed that radiant heat intensity would decline according to an inverse square law with increasing distance from the test pole. Since the mitigation is being applied to a single point in the study area, the Poisson distribution parameter was altered to apply to an area equivalent to the area burnt by the largest fire on record (see section 4.1.8). The effects of clearing distances of 10m, 30m and 50m from the test pole are plotted as intensity versus return period in Figure 16 below. Note that both axes in the plot have logarithmic scales. The dark blue line in the plot represents the do-nothing scenario, in which no clearing is undertaken and the fuel load is in contact with the test pole. This line is equivalent to the red line in Figure 15 (but appears different due to the logarithmic scales on the axes). The green, red and light blue lines represent the intensities of the extreme fires at cleared distances of 10m, 30m and 50m respectively. The dashed orange line at 40kW/m represents the design standard at which the test pole will ignite. 71 Results Figure 16: Effects of risk mitigation strategy: Intensity (kW/m) vs. Return Period (yrs). The dark blue line represents the intensity if no action is taken, and the green, red and light blue lines show the effects of clearing the fuel load distances of 10m, 30m and 50m from the pole respectively. The dashed orange line at 40kW/m represents the design standard (ignition intensity of the test pole). To assess the risk it is necessary to compare the existing situation (dark blue line) to the design standard (dashed orange line) for the desired lifetime (return period) of the test pole. Assume that we wish to design the pole to withstand the 100-year fire intensity. In this case the existing scenario shows that the fire intensity of the 100 year fire is 22 485 kW/m compared to the ignition standard of 40kW/m. Clearly the test pole will not withstand the 100 year fire event and mitigation is thus necessary. Clearing the fuel load decreases the intensity according to an inverse square law. Clearing the fuel back a distance of 10m from the test pole decreases the intensity of the 100 year fire to 217 kW/m. clearing the fuel to distances of 30m and 50m decreases the intensity of the 100 year fire to 24 kW/m and 8 kW/m respectively. Hence clearing the fuel load around the test pole just 10m decreases the radiant heat intensity of the fire at the pole by two orders of magnitude, and clearing a distance of approximately 30m will ensure that the pole will not ignite under the mean 100 year fire intensity conditions. Additional clearing further reduces the fire intensity at the test pole. These clearing distances compare well with the cleared distances recommended by the New South Wales Rural Fire Service, who recommend clearing a distance of 20-50m from the object of value, depending on the slope of the terrain (New South Wales Rural Fire Service 2003). 72 Discussion 6. Discussion The discussion of results is devoted to answering two questions i) What are the implications of the study for the management of wildfire in the northern jarrah forests of Western Australia? ii) Were the methods used in the study appropriate, and what were their limitations? 6.1. Fire in WA - Implications for management 6.1.1. Cause of ignition Figure 4 demonstrated that a very large proportion of wildfires (45% of ignitions) are deliberately lit. Lightning was the only non-human induced cause of fire (11% of ignitions). Thus approximately 89% of fire ignitions occur due to anthropogenic causes. This result has been documented in the literature and recognised by CALM (Underwood & Christensen 1981; Underwood 1988; Luke & McArthur 1986). Given that such a large number of fires are started by people, the existence of people in an area could be argued to provide the one of the largest risks of ignition. CALM consider the accessibility of areas of forest in their Wildfire Threat Analysis (WTA), arguing that inaccessible sites will not be likely to be ignited by arsonists (Kocsis & Irwin 1997) or recreational accidents (Muller 1993). Visual observation of the WTA maps appears to show increased numbers of wildfires concentrated around population centres (Muller 1993). This result suggests that the spatial distribution of ignition is an important parameter in the risk of wildfire occurrence. Given that towns and population centres are often considered to be the highest priority and most valuable built assets in an area (Underwood 1988), the fact that most wildfires are concentrated around these areas reinforces existing research (Fried et al. 1999; Braun 2001) that suggests that the wildland-urban fringe is a crucial area for wildfire management. 73 Discussion Given that relatively few fires are started in the absence of humans, the likelihood of wildfire occurring away from populated areas could be argued to be minimal. Similarly since the majority of assets lie near population centres, the severity of the consequences of wildfires occurring away from population centres could also be argued to be minimal (excluding ecological consequences). This suggests that, from a risk management point of view, emphasis for wildfire management could be focussed largely upon the wildlandurban fringe, rather than distributed throughout the forested area. There is the potential to quantitatively prioritise the wildland-urban fringe spatially based upon the risk of ignition. It needs to be remembered that this study has only looked at the risk of ignition at the wildland-urban interface without considering other factors. Since this area is the most populous in the study area, there are many social, organisational and political concerns that also need to be considered before this proposal is accepted or rejected. If a strategy of intensive risk-based burning along the wildland-urban interface was found to be desirable after considering the many other facets of wildfire management (e.g. legislative, organisational, political, public expectations etc.), prescribed burning of areas outside the interface could focus wholly on providing conditions conducive to the local flora and fauna. This strategy has the potential to address many of the specific criticisms of prescribed burning (e.g. (York 1994; Neyland & Askey-Doran 1994)) by ensuring that minimal prescribed burning is undertaken, and that it is undertaken for conservation purposes only. More general concerns (e.g. (Neyland & Askey-Doran 1994; Horton 2002)) may also be addressed by designing fire regimes that are optimised for the single purpose of protecting nature outside of the wildland-urban interface. 6.1.2. Return period of extreme intensity wildfire events The plots of extreme fire intensity against return period for 100 years of simulated record (Figure 6) showed that uncontrollable fires (>3000kW/m) are not uncommon in the jarrah forest. The return period of an uncontrollable fire (>3000kW/m) was found to be 1.12 years in the annual maximum series, and less than a year in the annual exceedance series. 74 Discussion The intensity of the 100-year fire event is over ten times the threshold above which the fire becomes uncontrollable. Before interpreting these results it is necessary to consider some of the assumptions behind them. Recall the size of the study area (Figure 2), which covers a large proportion of south-western Australia. Thus a return period of approximately 1 year suggests that (statistically) an uncontrollable fire will occur once every year somewhere within the study area. The study area is very large and hence the statistic is not quite as severe as it initially sounds. The effects of scaling the size of the study area down to a specific point are examined in section 5.1.7. It is also worth noting that the analysis assumed a percentage canopy cover of 50% when calculating the fuel loads for these simulations. Subsequent testing of the sensitivity of the fuel load to this parameter (section 5.1.5) showed that the choice of canopy cover has a significant effect on the intensity (Figure 12). Hence the fire intensities shown in Figure 6 should be interpreted with care as the actual fire intensity predicted by the model may be significantly greater or less than the value depicted. The uncertainty in the intensity of wildfire events increases as return period increases. The value of the 100 year fire event showed significant variability, fluctuating between intensities of 54 861 kW/m and 24 045 kW/m over the 50 simulations (Figure 6). It needs to be remembered that the values given are averages, and that significant variation exists in these values, particularly as the return period increases. Despite these assumptions, there is still significant cause for concern. Even if the highly sensitive canopy cover variable is reduced to a minimum value of 10%, the 1-year fire event is 2247kW/m, and this increases above the uncontrollable fire threshold when it reaches a return period of close to 5 years (Figure 12). These represent minimum average conditions, so it is likely that the intensity is greater than these values in reality. According to the results presented in this study, the current strategies will not eliminate extreme wildfires. 75 Discussion In light of this result, the study looked at the trends in the variables controlling the intensity of the extreme wildfire events. Three variables were examined: fuel load, the Fire Danger Index (FDI), and the surface moisture content. The Fuel Load As expected, the mean fuel load increased with increasing mean fire intensity (Figure 7). In the annual exceedance series, the mean fuel load ranged from 12.64 t/ha (1-year fire event) up to 16.64 t/ha (100-year fire event). Despite the fact that all the fuel loads in the extreme value series exceeded the 6-8t/ha threshold desired by CALM (Underwood 1988; Beck 1995), only the 50 and 100 year fuel loads were greater than one standard deviation from the mean fuel load of 7.8t/ha. Hence the fuel loads contributing to the extreme fire events tend to be large but not extreme. This result initially suggests that CALM’s hazard reduction strategy is effectively controlling extreme fuel loads- the mean fuel load lies within the desired range of 6-8t/ha and the extreme fires do not coincide with extreme fires unless they have return periods of approximately 50 years or greater. Reducing the fuel load has a dramatic effect on the fire intensity (Figure 12 and Figure 13). Figure 12 shows that the mean intensity of the 100 year fire can be decreased from 80 455 kW/m to 15 493 kW/m by decreasing the fuel conditions from maximum fuel age and 100% canopy cover down to minimum canopy cover and existing fuel age. The fuel load in these fires is 27t/ha and 10.5t/ha for maximum and minimum fuel conditions respectively. Figure 12 shows that there is clearly a significant reduction in intensity if the fuel load is reduced. However as noted above, the intensities of the extreme fires are still frequently above the 3000kW/m threshold even under minimum fuel conditions. By taking the derivative of Equation 3, it can be shown that the rate of accumulation of fuel load decreases exponentially with fuel age. Hence once fuel age increases above about 30 years (the oldest fuel used in the CALM data set), there is very little increase in 76 Discussion fuel load with increasing age. This is significant for extreme fires as the extreme fire intensities occur in large fuel loads (which should correspond to old fuel ages). The fact that fuel loads decreased so dramatically in Figure 12 is not due to the fuel age but to due to the variation of the canopy cover parameter. This can be clearly seen in Figure 11, where the fuel age is extrapolated from 30 years to 70 years old – the increase in fuel load is just 4t/ha, but the increase in the 30 year fuel load as a result of increasing from 50% canopy cover to 100% canopy cover is 8.3t/ha. Once the fuel becomes old, the fuel age does not seem to have a major influence on the extreme values in any of the figures plotted. The effect of the canopy cover is simply to increase the total amount of fuel available. Hence changing the canopy cover will increase the intensity of all fires in the area as more fuel is available to the fire. This variable will change with locality and is subject to significant variation. The difference in the fuel load contributing to the 100-year and the 1-year fire in Figure 7a is just 3.76t/ha, despite an increase in intensity of over 30 000kW/m. The 1-year fire intensity corresponds to a fuel age of 11.9 years, and the 100-year fire event corresponds to a fuel age of 20.6 years. The reason that the difference between the two fuel loads is so small is that fuel age ceases to cause a significant increase as it ages. Fuel age makes a significant difference to fire intensity for low fuel ages, but for large fuel loads/old fuels, the age of fuel ceases to make a significant difference. Similar observations have been made by Bradstock et al. (1998) and are discussed in Fernandes & Botelho (2003). The significance of this result is that it is not the fuel load that controls the intensity of an extreme fire. Although the amount of fuel is proportional to the intensity of the fire, after about 10-18 years (dependent on canopy cover- see Figure 10), the rate of fuel accumulation will decrease to less than half its original accumulation rate, and the fire intensity will begin to be controlled by the other variables - the FDI or the surface moisture content. Essentially, if the fuel load is very low, then fires can be controlled 77 Discussion even in extreme weather conditions as there is simply not enough fuel to contribute to a large fire (Fernandes & Botelho 2003). However, if the fuel load rises above a certain threshold age, then it will cease to dominate the intensity equation (Equation 5) as it will cease to increase dramatically while the other variables are free to fluctuate. The ramifications of this finding for hazard management are that fuel loads need to be kept below a threshold level where the fuel load still dominates the intensity equation (Equation 5). CALM’s rotation times of 5-7 years in the jarrah (Department of Conservation and Land Management 2000) are likely to fall within this threshold level as the rate of accumulation of fuel is still significant at these ages (rate of change is 0.78 and 1.77 for 10% and 100% canopy cover respectively) . The mean fuel age in this study is 5.7 years. However, whilst the older fuel loads that contribute to the extreme fire events do increase with increasing return period, the increase is not dramatic since the fuel cannot accumulate to very extreme levels – it is not the fuel load that dominates these extreme fires. In order to eliminate the possibility of these extreme fires using hazard management, the management authority must increase the frequency of their prescribed burning to ensure that fuel ages are consistently kept below the threshold level at which the other variables begin to dominate the intensity equation. This is likely to be an expensive and unpopular result, as the frequency of prescribed burning already receives considerable criticism (Horton 2002; Neyland & Askey-Doran 1994). However risk should not be considered simply as hazard management, and there may be other more effective ways to manage environmental risk (Braun 2001; Braun 2002; Steedman 2002). The Fire Danger Index Figure 8 shows that the Fire Danger Index (FDI) increases dramatically as intensity increases. This demonstrates a strong correlation between the weather conditions (the FDI is an aggregation of the effects of wind speed, moisture content, dry temperature and relative humidity (Beck 1995)) and the occurrence of extreme wildfire events. Almost all of the fire events in the extreme distributions in Figure 8 were found to have FDI values greater than at least one standard deviation above the mean FDI conditions, suggesting that all the FDI values were from the extreme tail of the FDI probability distribution 78 Discussion function. Thus it can be concluded that extreme intensity fires coincide with extreme weather conditions. The previous section of this discussion demonstrated that extreme intensity fires are not controlled by the accumulation of large fuel loads because the fuel load ceases to accumulate rapidly enough to make a difference to fire intensity once it reaches some threshold fuel load. Comparison of the contribution of the three parameters influencing fire intensity (Figure 7, Figure 8 and Figure 9) shows that the FDI values are increasing much more dramatically than the fuel load or surface moisture content values. The FDI values are also almost all from the extreme tail of the probability distribution function, whereas the fuel load and surface moisture content values are rarely outside one standard deviation of the mean. From these results it is argued that it is the FDI that dominates the intensity of extreme fire events. Hence the major factor determining the intensity of an extreme fire event is the weather conditions and not the moisture or weight of the fuel load. Although there are many factors that affect the intensity of a fire, the only one that can be practically controlled is the fuel load (Foster 1976; Department of Conservation and Land Management 2000; Luke & McArthur 1986). The aim of the fuel reduction is not to prevent forest fire, but to keep fires under control when they do occur (Underwood & Christensen 1981). However the analysis of the extreme FDI values undertaken in this study shows that although the fuel load is a necessary condition for an extreme fire, it is the weather conditions that determine the severity of the fire. Thus controlling the fuel load alone is not sufficient to manage the risk of wildfire - the management authority also needs to be prepared to manage against severe weather conditions. As mentioned previously, this either requires frequent burning to dramatically reduce the fuel load (a hazard management approach), or a different, more holistic management strategy based upon the protection of assets. 79 Discussion Surface Moisture Content Figure 9 shows the contribution of the surface moisture content to the extreme fire intensities. A slightly decreasing trend in surface moisture content was observed with increasing fire intensity, suggesting that drier fuels contributed to more intense fires. None of the mean simulated surface moisture content values were found to be outside one standard deviation of the mean of the distribution. This is likely to be due to the fact that the simulation was taken from a uniform distribution. This means that there is an equal probability of selecting any moisture content from 3-26% (the range of the distribution function). Hence the number of high and low surface moisture contents will be roughly equal, and taking the mean of all the simulated moisture contents will cause the effects of extreme values to cancel one another out, resulting in surface moisture contents that are distributed closely around the mean as in Figure 9. This hypothesis is backed up by the range of simulated values (range of the coloured dots) for each return period in Figure 9. For every return period, the whole distribution has been sampled during the simulation (because there is an equal probability of selection in the uniform distribution). In comparison, the fuel load (Figure 7) and FDI (Figure 8) show that the extreme events were sampled from the upper end of the probability distribution function. Because there are no extreme surface moisture content values that contribute to the extreme fire intensities in Figure 6, and very little trend in the surface moisture content values, it is unlikely that surface moisture content as an independent variable plays a major role in determining extreme fire intensity. However this is slightly misleading, as surface moisture content is included within the variables that contribute to the FDI, and so it is likely that surface moisture content will have some effect on fire intensity. A Broader Approach to Bushfire Management The previous discussion showed that uncontrollable wildfires occur (statistically) with a return period of approximately 1 year. It was proposed that extreme wildfire events are driven largely by the weather conditions and not by excessive fuel loads or low moisture 80 Discussion content fuels. The following sections seek to explain the consequences of these findings and propose some solutions to the problems that they pose. The findings of this study are not unique. CALM have found that fuels accumulate to levels where average weather conditions will make the fires uncontrollable in 5-7 years in the jarrah forest (Department of Conservation and Land Management 2000), and accept that prescribed burning is only effective if undertaken in this period. Braun (2002) noted that if this information was adjusted to account for very high and extreme weather conditions, then the time taken for fuels to accumulate to levels where the fire becomes uncontrollable is likely to be even less. In an examination of the wildland-urban interface, another study found that a large proportion of fire events where life and property are lost correspond to extreme weather conditions, and that prescribed burning would have to be undertaken annually across very large areas (~40%) of the wildland-urban interface to obtain a satisfactory level of risk (Bradstock et al. 1998). It appears that the prescribed burning policy currently practiced by the management authority is not sufficient to prevent extreme intensity wildfires from occurring (Figure 6). Given that these extreme intensity events are driven by extreme weather conditions over which the management authority has no control, how can the risk of wildfire be mitigated? In light of the fact that fuel load is the only variable that can be controlled, the logic of prescribed burning has always centred on fuel reduction. However, given that this is not necessarily the most important factor determining the occurrence of extreme intensity wildfires, it becomes necessary to broaden the context in which management is carried out. It is proposed that a better understanding of the true risk of fire can be obtained by considering a much broader range of factors under the notion of risk than simply the fuel load. What is needed is an acceptance that extreme fires cannot necessarily be prevented, and the development of a measure of the capacity of the system to withstand and recover from an event of the intensity predicted by this study. 81 Discussion CALM have already begun taking steps towards a more holistic view of risk with their wildfire threat analysis (WTA) which includes the likelihood of ignition, the probable behaviour of the head fire, the values at risk and the suppression capacity of the management authority (Muller 1993). Other research has also argued that broader mitigation strategies are necessary. Braun (2001) argues that risk needs to include more than just the hazard management provided by CALM in the past. He argues that a well-prepared and educated community can make a significant difference to the severity of the impacts of an intense bushfire. The New South Wales Rural Fire Service provide a number of methods to protect the house against severe fire that can significantly reduce the intensity of extreme wildfires on these assets (New South Wales Rural Fire Service 2003). These strategies are discussed in the following section. The framework provided for the management of environmental risk by Goff and Steedman (1997) is perhaps a goal to which fire risk managers should aspire. In their paper, risk is broken down into four independent categories - primary (cause), secondary (transport), tertiary (damage/severity) and quaternary (recovery) risk. In the context of fires, the primary risk may be assessed as the risk of ignition based upon the cause of fire, the secondary risk as the likelihood of the fire reaching an asset, the tertiary risk as the likelihood of severe damage being caused, and the quaternary risk as the likelihood of the asset recovering from the fire. Such an approach would capture not only the hazard, but also incorporate the ability of the system to withstand and recover from an intense fire. 6.1.3. Mitigation An attempt to demonstrate how an alternative strategy for mitigation might be applied was carried out to complete the ‘assess risk’ and ‘treat risk’ steps of Australian Standard 4360:1995. In the example, a wooden test pole was placed into the fuel load and assumed to ignite at 40kW/m2. To mitigate against ignition, all fuel around the pole was cleared to some radial distance from the pole. Radiant heat intensity was assumed to decay according to an inverse square law and significant reductions in intensity resulted from 82 Discussion relatively small cleared distances from the pole. The results of this analysis for cleared distances of 10m, 30m and 50m are shown in Figure 16. Figure 16 shows that relatively small changes made around the test pole can result in a dramatic reduction in radiant heat intensity at the asset. Extrapolating this result to houses or buildings, it is possible to greatly reduce the risk of fire by clearing or regularly burning around the house (New South Wales Rural Fire Service 2003). Clearing around the asset is only one way in which the broader risk of bushfire can be reduced. Other strategies include using fire-retardant building materials (Braun 2002), ensuring that the community is educated and prepared for fire and has an adequate water supply (Braun 2001), implementing barriers or windbreaks, and selecting fire resistant plant species when landscaping (New South Wales Rural Fire Service 2003). 6.2. Fire in WA - appropriateness of the model 6.2.1. Empirical vs. physical model The study employed an empirical model based on experimental fires to predict fire intensity (Beck 1995). Although such models are useful, they have certain drawbacks. Firstly, the sheer number and complexity of these equations renders them cumbersome - a different equation is required for each vegetation type, fuel class etc. As an illustration of this, consider Beck’s (1995) paper, which contains 72 equations. However, the major drawback to an empirical model is the lack of process understanding that it disguises. An empirical equation is a very powerful tool to obtain an initial predictive tool, but empirical equations can only be applied to the conditions under which they have been tested (Beck 1995). Beck (1995) gives application bounds to all equations, based upon acceptable root-mean-square errors from the original data. It was also noted by Beck (1995) that the parameters leading to the extreme events became less certain as parameters increased. 83 Discussion Extreme fires are often uncontrollable, and hence it is likely that few experimental data exist for extreme conditions. Hence the application of the empirical model to these extreme conditions may not be appropriate as it is likely that it has not been widely tested under these conditions. The conditions contributing to the extreme fires in this study often exceeded the application bounds of Beck’s (1995) equations. This implies that there is significant uncertainty in the results. Combined with the uncertainty resulting from the simulation (Figure 6), the results should be interpreted with caution- although the observed trends are likely to be correct, the magnitude of the values may fluctuate significantly from their stated values. To overcome the problems associated with the empirical model, it would be beneficial to attempt to apply a physical model based upon an understanding of fire transport processes. However these models are often spatially based and require considerable data about the progression of a fire front through spatial cells in time (e.g. Hargrove et al. (2000), Green (1983), Dupuy & Larini (1999)). Since the data on fires is rarely this detailed, these models are often based entirely upon theoretical grounds with little testing outside the laboratory (e.g. Beer (1991), Green (1983)). Given the CALM data set, it was most practical to apply the empirical CALM model when undertaking the analysis. Keane et al. (2003) support the methodology used in this study by arguing that simulation models are the most effective means of capturing temporal elements (such as return period). However it would be beneficial to compare the results from this study to a similar study in the same study area that employs a physical model. 6.2.2. Sensitivity of the model The sensitivity of the model to the assumptions can have significant ramifications for the reliability of the results. A number of assumptions were made which could affect the results and these are discussed in this section. 84 Discussion The CALM data set contained a number of extreme FDI and fuel age values. Consultation with CALM (Wallace (personal communication) 2004; Sneeuwajgt (personal communication) 2004) led to the removal of some of these extreme values from the data set. FDI values were restricted to be 3000 or less (6 values were dropped), and fuel load was restricted to be 30 years old or less (2 values were dropped). This action was based largely upon the area burnt variable, which showed that the extreme values did not correlate with large burnt areas. The data were considered by CALM to be too large to be realistic given their knowledge of the area. The removal of the extreme values is likely to have an effect on the extreme intensity distribution, as the extreme values of FDI and fuel age are selected when extreme intensity fires occur. However a sensitivity analysis of the effects of fuel age showed that fuel ages greater than 30 years have little effect on the fuel load (section 5.1.5, Figure 10) as the rate of accumulation by fuels of this age is very slow. However the effect of the extreme FDI values is likely to have an effect since it is directly related to intensity by the rate of spread of the fire (Equation 6 and Equation 4). This study assumes that CALM’s knowledge of the data set is sufficient reason to remove the extreme FDI values, however it should be noted by the reader that their removal changes the FDI probability distribution function and decreases the extreme intensity fires. Although the fuel age does not significantly affect the fuel load at large fuel ages, the percentage canopy cover makes a dramatic difference to the total fuel load (Figure 10) and hence the intensity (Figure 12). This variable was assumed to be 50% for the analysis in sections 5.1.3 and 5.1.4, however this is likely to vary with location. This phenomenon has already been discussed in some detail in section 5.1.5 and should be kept in mind when interpreting the results. The surface moisture content was sampled from a uniform distribution and as a result varied little with increasing intensity (Figure 9). In reality the surface moisture content is unlikely to follow a uniform probability distribution function, however surface moisture content data was not available from CALM and the uniform distribution was used. Future 85 Discussion research should investigate the distribution of surface moisture content to improve this assumption. Moisture content may have a more significant effect on extreme fire intensities than that observed in this study. 6.2.3. Spatial Considerations: The size of the study area With the exception of the mitigation example, this study assumed that the risk of wildfire can be assessed at a macro level. However it was shown in section 5.1.7 that smaller areas are associated with a smaller probability of extreme fire occurrence. This result accounts for the fact that extreme events are rare in space as well as time – the smaller the area of interest, the less likely that a fire will intersect the area. This result is significant for design – the results in Figure 15 show that there is a difference of 8824kW/m between the 100-year fire events predicted using the full study area and the point study area. Given that 3000kW/m represents an uncontrollable fire (Muller 1993), this is a significant difference. From the point of view of design, more extreme intensity fires require more mitigation to reduce risk, and hence designs for extreme intensity fires are more costly to implement. There is a need to determine the appropriate spatial scale at which to assess the risk of fire before it can be reliably used in design. Another spatial consideration is heterogeneity. The investigation of the effect of study area size accounts for the size of the area of interest, but does not account for the differences in probability of occurrence (e.g. fires are more likely to occur in areas where people are present to cause ignition). The study assumes that the study area is homogenous. Future research should be directed towards improving this assumption. 6.2.4. Correlations between the variables Figure 14 shows the correlation between FDI, fuel load and the area burnt by wildfires. It was found that there was no correlation between any of these variables. This is a necessary condition for Monte Carlo simulation (Koller 2000), and was an expected result for FDI and fuel load, which contribute to the simulation of the intensity. This 86 Discussion result means that there is no relationship between the weather conditions and the fuel load – i.e. the weather conditions do not influence the fuel load and vice versa. However since FDI and fuel load are directly related to fire intensity, it was expected that there would be a positive correlation between these variables and the area burnt (i.e. it was expected that the greater the fuel load and the more severe the weather conditions under which ignition takes place, the greater the area of the fire burnt). In fact it turned out that there was no significant correlation between these variables. One possible explanation for the lack of correlation between the variables could be CALM’s suppression efforts. CALM may have suppressed a large number of fires occurring on days with large FDI and high fuel loads, reducing the correlation between the variables (Fernandes & Botelho 2003). The lack of correlation between burnt area and the other variables is important only in the parts of the simulations where the burnt area variable is used. This variable is used primarily to determine the number of fires per year, in which the number of fires exceeding a burnt area of 10ha is assumed to be Poisson distributed (Figure 5a). The lack of correlation between the extreme fuel ages and FDI values was also used a justification to drop the extreme fuel ages and FDI values from the data set. Since extreme FDI values have the potential to affect the intensity of extreme fires, the lack of correlation between FDI and burnt area should be clarified in future. 6.2.5. The Annual Maximum series vs. the Annual Exceedance Series Both the annual maximum and annual exceedance series of the extreme intensity series were determined (Figure 6). The use of these series was based upon their use in flood hydrology (Chow et al. 1988). In flood hydrology, the annual maximum series is used as a standard (Chow et al. 1988; Pattison et al. 1977). However, there is no standard for which series to use for wildfires. Some discussion about the merits of each series is contained in this section. 87 Discussion In flood hydrology, the argument for using the annual maximum series is that it is difficult to determine whether events in the annual exceedance series are independente.g. a large flood can leave the soil saturated and induce another large flood that is really just a result of the initial flood (Chow et al. 1988). However, this logic does not hold for fires. A large fire will reduce the fuel load in the burnt area, and hence the occurrence of a fire will not be likely to induce another extreme intensity fire (neglecting the possibility that the same fire re-ignites). The other major difference between the application of this method to this study and its application in flood hydrology is that the fires are measured at a wide range of stations in the study area, and not at one discrete gauging station at the catchment outlet. Hence the likelihood of two recorded events influencing one another is greatly diminished. For large return periods, the intensity of the fires predicted by the annual maximum and annual exceedance series are very similar. It is only at low return periods that the two approaches differ (Chow et al. 1988). The annual maximum series predicts lower intensity fires than the annual exceedance series because only the largest fire from each year can be selected. In some cases the second most intense fire in a year can exceed the intensity of the largest fire in another year – this fire will be excluded from the annual maximum series but included in the annual exceedance series, leading to the difference in the two series (Chow et al. 1988). For future research into fires, it is recommended that the annual exceedance series be used as a standard rather than the annual maximum series. The fires are not measured at a single point (as they are in flood hydrology) and there is an equal probability of a fire occurring anywhere in the domain. Extreme weather conditions and high fuel loads can induce two extreme intensity fires at the same time in two different locations within the domain that can burn entirely independently of one another. Both should be included in the extreme value series as they occurred due to different sets of conditions. In fact the 88 Discussion Poisson distribution used to predict the number of fires occurring in a year requires the assumption that fire events are independent (Section 4.1.3). 6.3. Limitations of the study Limitations of the study fall into three categories: • • • Uncertainty about the composition of the recorded FDI values, The inability of the empirical model to capture environmental transport processes, Limited ability to predict the spatial effects of study area and inability to capture spatial heterogeneity within the domain Each of these categories is elaborated below. Composition of the FDI One of the biggest constraints on the analysis was the lack of knowledge of the raw parameters contributing to the fire danger index (FDI). The FDI is based upon the raw variables surface moisture content and wind speed, but the distributions of these parameters was not known. Since it was found that the FDI was the major factor driving extreme intensity fires, it would have been informative to investigate the relative contributions of the component variables of the FDI. The extreme FDI values that were dropped from the set (>3000) were dropped on the basis that they were too large to be realistic. However the values of the raw variables that contributed to these extreme FDI values were not known, so it was difficult to determine whether these extremes were physically realistic. Given the impact that these extreme FDI values can have on the intensity, an understanding of the raw variables would be a powerful tool towards a physical process understanding of the problem. These limitations were also noticed by Beck (1995), who noted that the CALM equations work very effectively in operation. However, it was also noted that the incompleteness of published data behind the prediction system detracted from its scientific credibility, and recommended that further publications be undertaken to shed light on the prediction system (Beck 1995). Given the complex multivariate nature of wildfire occurrence, environmental ignition and transport processes, and geographical spatial distribution of 89 Discussion the problem and the constraints preventing excessive data collection, it seems that an empirical model is the most practical management tool at this time. However, as noted by Beck (1995), it is desirable that efforts be made to ensure that the assumptions behind the prediction system are made transparent. The fact that the raw parameters behind the FDI were not provided also meant that the surface moisture content had to be sampled from a uniform distribution to determine the fuel quantity correction factor (Equation 7). The limitations imposed by this uncertainty have already been outlined in (section 6.2.2). Environmental transport mechanism Another limitation of the study is that it can only give a macro-level picture of wildfire risk. Events at specific locations cannot be forecast. This is due to the fact that the precise location and transport mechanisms of the individual fires within the data set are not known. The equations of Beck (1995) cannot capture transport processes as they do not account for directional variables (e.g. wind direction and the acceleration of the fire front) – a different model would need to be used to capture this variability. Other spatial considerations The large spatial domain (Figure 2) over which fires were analysed was used to ensure that the data set was large enough to give reliable probability distribution functions. However, the size of the study area means that the risk of wildfire applies to a very large area, and this reduces the applicability of the results. For example, a 1-year return period of an extreme intensity fire means that a fire is statistically expected somewhere in the domain every year, but it is not certain where the fire will occur. Although an attempt was made to account for the size of the study area, the analysis assumed that the domain was homogeneous (i.e. that there was a uniform probability of a fire occurring anywhere in the domain). However over such a large domain, spatial heterogeneity is known to exist (Williams & Mitchell 2001). If this analysis is to be applied in practice, it is crucial that the spatial heterogeneity is well understood. 90 Conclusions 7. Conclusions The major findings of the study are: • CALM run a practical fuel load hazard reduction management program achieved by prescribed burning The probability distribution function of fuel loads showed that the mean fuel load in the northern jarrah forest was below the controllable threshold (8t/ha). Although the fuel loads selected to create the extreme value series often exceeded the controllable threshold, these high fuel loads were almost all within one standard deviation of the mean and were not considered to be extreme fuel loads. The absence of extreme fuel loads was attributed to CALM’s policy of prescribed burning. • FDI is weather dependent and independent of available fuel load and moisture content for a given wildfire; consequently fuel load hazard reduction will not necessarily result in reduced wild fire intensity It was found that the wildfire intensities exceeded the controllable threshold (>3000kW/m) with a return period of close to 1 year. This result implies that (statistically), there is likely to be an uncontrollable fire somewhere in the northern jarrah every year. An analysis of the factors contributing to the extreme intensity fires determined that although a moderate fuel load contributed to the extreme intensity fires, it is the weather conditions (FDI) that drive the extreme fires. Further, the fuel age was found to have the most influence at low fuel ages - this was only an important variable for young fuels. This led to the conclusion that prescribed burning will be effective only if it can maintain very low fuel ages (and hence fuel loads) throughout the forest. If fuels are allowed to accumulate above a moderately high level, then the occurrence of extreme weather conditions is likely to result in uncontrollable fires. • Hazard reduction method has served CALM well and may be improved by considering the widely accepted risk management approach described by Australian Standard 4360:1995 91 Conclusions CALM are already moving towards a risk management approach with the Wildfire Threat Analysis program. The framework for risk management outlined in Australian Standard 4360:1995 was applied to this study. This approach allows for a broader perspective of risk by changing the focus of management from the hazard to the asset to be protected. Alternative mitigation strategies such as prioritised areas for prescribed burning, localised clearing, alternative building materials and community education can be considered using the Australian Standard approach. A simple application of a design standard to a test pole was demonstrated to illustrate how an alternative mitigation strategy can be practically applied. • Where possible, wildfire risk analysis should consider weather and spatial geographic parameters in addition to fuel load hazard management. An analysis of the parameters driving extreme fire events showed that fuel load, weather conditions and spatial parameters were important factors driving extreme fire events. Analysis of the risk of wildfire should include all of these parameters. Further research should be directed towards determining the effects of spatial parameters (transport processes, heterogeneity and study area size) affecting wildfire risk. 92 Recommendations 8. Recommendations Given that the fire danger index (FDI) seems to be the most important factor driving the extreme intensity fires, it is important that the factors contributing to the FDI are understood. Unfortunately, data for the parameters contributing to the FDI (wind speed and surface moisture content) were not available in this study. An understanding of the most important weather factors driving wildfires is essential if a physical rather than empirical model for wildfire is to be developed. In addition, the surface moisture content had to be sampled from a uniform distribution in this study, so the contribution of the surface moisture content to extreme fire events is not known - this could be rectified if the probability distribution function for surface moisture content was known. Further research into quantifying the probability of fire weather factors in Western Australia is thus recommended. The move from an empirical to a physical model for wildfire intensity is recommended. An empirical model allows for the prediction of intense wildfires, but does not allow for an understanding of the processes driving the fires. A process understanding is desirable as it makes it possible to quantify the dominant processes and controls of the system. Mitigation strategies to target these dominant processes can then be designed to most effectively reduce fire intensity. A number of physical models exist in the literature (Hargrove et al. 2000; Green 1983; Dupuy & Larini 1999). If sufficient data to run these models becomes available, it is recommended that a physical model is applied to Western Australian forests and the results compared to the results from Beck’s (1995) empirical model. The study showed that the size of the study area has an effect on the extreme fire intensity (Figure 15). The variation in intensity with size of the study area was found to be significant and it is recommended that further research is directed towards determining the appropriate spatial scale at which to assess risk of wildfire. This is a common problem in spatial statistics and applied science (e.g. Cressie (1993), Ripley (1981)). The 93 Recommendations techniques developed in other fields (e.g. Cressie (1993), Ripley (1981)) should be applied to this study. Although it still requires much research, it is recommended that eventually a risk management strategy similar to Goff and Steedman’s (1997) model be adopted. The processes of ignition, transport, damage and recovery should be quantified and risk should be managed on the basis of a combination of these processes. CALM’s Wildfire Threat Analysis has already taken steps in this direction (Muller 1993; Sneeuwajgt 1998). Such a strategy would give a broad picture not just of the risk of fire occurring, but also of the consequences of the fire on assets, the preparedness of the community to deal with the fire, and the ability of the assets to recover from the fire. This is clearly a more powerful management tool than simply the chance of an extreme fire occurring. 94 References 9. References Andrews, P., Loftsgaarden, D. & Bradshaw, L. 2003, 'Evaluation of fire danger rating indexes using logistic regression and percentile analysis', International Journal of Wildland Fire, vol. 12, pp. 213-226. Australian Capital Territory Emergency Services Authority 2004, Strategic Bushfire Management Plan for the Act. Version 1- Draft, Australian Capital Territory Emergency Services Authority,, Canberra. Babrauskaus, V. 2001, 'Ignition of wood- a Review of the State of the Art', in Interflam 2001, Interscience communications Ltd, pp. 71-88. Beck, J. 1995, 'Equations for the forest fire behaviour tables for Western Australia', CALMScience, vol. 1, no. 3, pp. 325-348. Beer, T. 1991, 'The interaction of wind and fire', Boundary-Layer Meterology, vol. 54, no. 3, pp. 287-308. Boura, J. 1994, 'Reconciliating Fire Protection and Conservation issues at the UrbanForest Interface', in Fire and Biodiversity- The effects and effectiveness of fire management, Department of the Environment, Sport and Territories, Biodiversity Unit, Footscray, Melbourne, pp. 247-252. Bradstock, R., Gill, A., Kenny, B. & Scott, J. 1998, 'Bushfire risk at the urban interface estimated from historical weather records: consequences for the use of prescribed fire in the Sydney region of south-eastern Australia.' Journal of Environmental Management, vol. 52, pp. 259-271. Braun, K. 2001, 'Wildfire risk mapping- including the community', in Proceedings of the Australasian Fire Authorities Annual Conference., D, Darwin, NT. Braun, K. 2002, 'Bushfire threat to homeowners', in Fire in Ecosystems of South-Western Australia: Impacts and Management, Department of Conservation and Land Management, Perth WA. Burrows, N. 1984a, Describing Forest Fires in Western Australia: A Guide for Fire Managers, Technical Paper 9, Forests Department of Western Australia, Perth. Burrows, N. 1984b, Predicting Blow-Up Fires in the Jarrah Forest, Technical Paper 12, Forests Department of Western Australia, Perth. Burrows, N. 2003, 'Fire for life', Landscope, Winter 2003, vol. 18, no. 4, pp. 21-26. Burrows, N., Ward, B. & Robinson, A. 1999, 'The Role of Indicators in Developing Appropriate Fire Regimes', in Bushfire 99- Australian Bushfire Conference, Albury. CALM unpublished data 2002, Unpublished data- number of fires and area burnt by selected cause 1937-1999, Department of Conservation and Land Management. Catchpole, W., Bradstock, R., Choate, J., Fogerty, L., Gellie, N., McCarthy, G., McCaw, W., Marsden-Smedley, J. & Pearce, G. 1999, 'Cooperative Development of Prediction Equations for Fire Behaviour in Heathlands and Shrublands', in Australian Bushfire Conference, eds I. Lunt, D. Green & B. Lord, Albury, NSW. Chow, V. T., Maidment, D. R. & Mays, L. W. 1988, Applied Hydrology, McGraw-Hill Book Company. Clayton, B., Day, D. & McFadden, J. 1987, Wildland Firefighting, State of California Resources Agency, Department of Forestry and Fire Protection. 95 References Commonwealth Government of Australia 1984, Bushfires and the Australian Environment: Report by the Standing Committee on Environment and Conservation, Canberra. Cressie, N. A. C. 1993, Statistics for spatial data, Wiley. Department of Conservation and Land Management 2000, Fire management on CALM managed lands in the South West of Western Australia, Department of Conservation and Land Management (CALMfire branch), Perth. Department of Conservation and Land Management 2001, Department of Conservation and Land Management response to the recommendations of the review of fire operations in forest regions managed by the Department (Muller review 2001), Department of Conservation and Land Management, Perth. Dovers, S. 1995, 'Risk, uncertainty and ignorance: policy process and institutional issues', in Fenner Conference on the environment: Risk and Uncertainty in Environmental Management, Canberra. Dupuy, J. & Larini, M. 1999, ' Fire spread through a porous forest fuel bed: a radiative and convective model including fire-induced flow effects', International Journal of Wildland Fire, vol. 9, no. 3, pp. 155-172. Fernandes, P. M. & Botelho, H. S. 2003, 'A review of prescribed burning effectiveness in fire hazard reduction', International Journal of Wildland Fire, vol. 12, pp. 117128. Foster, T. 1976, Bushfire: History, Prevention, Control, AH & AW Reed Pty Ltd, Sydney. Franklin, J. & Agee, J. 2003, 'Forging a Science-Based National Forest Fire Policy', Issues in Science and Technology, vol. 20, no. 1, pp. 59-66. Fried, J., Winter, G. & Gilless, J. 1999, 'Assessing the benefits of reducing the risk of wildfire at the wildland-urban interface: A contingent valuation approach', International Journal of Wildland Fire, vol. 9, no. 1, pp. 9-20. Gill, A. & Moore, P. 1997, Contemporary Fire Regimes in the Forests of South Western Australia, CSIRO Centre for Plant Biodiversity Research. Goff, H. & Steedman, R. 1997, 'Environmental Risk Management: A proposed scheme', APPEA Journal 1997, pp. 714-720. Green, D. 1983, 'Shapes of Simulated fires in Discrete Fuels', Ecological Modelling, vol. 20, pp. 21-32. Hammersley, J. & Handscomb, D. 1964, Monte Carlo Methods, Spottiswoode, Ballantyne & Co. Ltd, London. Hargrove, W., Gardner, R., Turner, M., Romme, W. & Despain, D. 2000, 'Simulating fire patterns in heterogeneous landscapes', Ecological Modelling, vol. 135, pp. 243263. Haswell, D. & Brown, N. 2002, 'Approaches to community safety and bushfire in SouthWestern Australia', in Fire in Ecosystems of South-Western Australia: Impacts and Management, Department of Conservation and Land Management. Horton, D. 2002, 'Fire and Australian Society', in Fire in Ecosystems of South-West Australia: Impacts and Management. Proceedings of the April 2002 Symposium organised by the Department of Conservation and Land Management. 96 References Keane, R., Cary, G. & Parsons, R. 2003, 'Using simulation to map fire regimes: an evaluation of approaches, strategies and limitations', International Journal of Wildland Fire, vol. 12, pp. 309-322. Keeley, J., Fotheringham, C. & Morais, M. 1999, 'Reexamining fire suppression impacts on brushland fire regimes', Science, vol. 284, no. 5421, pp. 1829-1832. Kennedy, J. & Neville, A. 1986, Basic statistical methods for engineers and scientists, Third Edition edn, Harper & Row Inc, New York. Knox, B., Ladiges, P. & Evans, B. 1999, Biology, McGraw-Hill, Sydney. Kocsis, R. & Irwin, H. 1997, 'An analysis of spatial patterns in serial rape, arson and burglary: The utility of the circle threory of environmental range for psychological profiling', Psychiatry, psychology and law, vol. 4, no. 2, pp. 195206. Koller, G. 2000, Risk Modeling for determining value and decision making, Chapman and Hall/CRC, Boca Raton, Florida. Lang, S. 1997, Burning the Bush: A spatio-temporal analysis of Jarrah forest fire regimes, Australian National University. Luke, R. & McArthur, A. 1986, Bushfires in Australia, 2 edn, CSIRO Division of Forest Research- Department of Primary Industry, Canberra. Manno, I. 1999, Introduction to the Monte-Carlo Method, Akademiai Kiado, Budapest. McCaw, W., Neal, J. & Smith, R. 2002, 'Stand Characteristics and fuel accumulation in a sequence of even-aged Karri (Eucalyptus diversicolor) stands in south-west Western Australia', Forest Ecology and Management, vol. 158, pp. 263-271. McCormick, R. J., Brandner, T. A. & Allen, T. F. 1999, 'Toward a theory of meso-scale wildfire modeling- A complex systems approach using artificial neural networks', in The Joint Fire Science Conference and Workshop, eds L. F. Neuenschwander, K. C. Ryan & G. E. Gollberg, Boise, Idaho. Morrison, D. 2002, 'Effects of fire intensity on plant species composition of sandstone communities in the Sydney region', Australian Journal of Ecology, vol. 27, pp. 433-441. Muller, C. 1993, Wildfire Threat Analysis- A decision support system for improved fire management, Department of Conservation and Land Management, Bunbury. Muller, C. 2001, Review of fire operations in forest regions managed by the Department of Conservation and Land Management: Report to the executive director of the Department of Conservation and Land Management., Department of Conservation and Land Management, Perth. National Wildlife Coordinating Group, Fire Effects Guide [Online], Available: http://fire.r9.fws.gov/ifcc/monitor/EFGuide/index.htm. New South Wales Rural Fire Service 2003, Guidelines for Asset Protection Zones, New South Wales Rural Fire Service, Granville, NSW. Neyland, M. & Askey-Doran, M. 1994, 'Effects of repeated fires on dry sclerophyll (E. Sieberi) forests in Eastern Tasmania', in Fire and Biodiversity- The effects and effectiveness of fire management, ed. J. Barnett, Department of the Environment, Sport and Territories, Biodiversity Unit, Footscray, Melbourne, pp. 81-85. Noble, I., Bary, G. & Gill, A. 1980, 'McArthur's fire-danger meters expressed as equations', Australian Journal of Ecology, vol. 5, pp. 201-203. 97 References Pattison, A. E., Ward, J. E., McMahon, T. E. & Watson, B. E. 1977, Australian Rainfall and Runoff- Flood Analysis and Design, Institution of Engineers Australia, Canberra. Petrauskas, C. & Aagaard, P. 1971, 'Extrapolation of Historical Storm Data for Estimating Design-Wave Heights', Society of Petroleum Engineers Journal, vol. 251, pp. 23-37. Plate, E. 1982, Engineering Meteorology, Elsevier Scientific Publishing Company, Amsterdam. Preisler, H., Brillinger, D., Burgan, R. & Benoit, J. 2004, 'Probability-based models for estimation of wildfire risk', International Journal of Wildland Fire, vol. 13, pp. 133-142. Ripley, B. D. 1981, Spatial statistics, Wiley, New York. Rodger, G. 1961, Report of the Royal Commission appointed to enquire into and report upon the bush fires of December, 1960 and January, February and March, 1961 in Western Australia, the Measures necessary or desirable to prevent and control such Fires and to protect life and property in the future and the Basic Requirements for an effective State Fire Emergency Organisation. Rogerson, P. & Sun, Y. 2001, 'Spatial analysis of geographic patterns: An application to crime analysis', Computers, Environment and Urban Systems, vol. 25, pp. 539556. Rothermal, R. 1972, A mathematical model for predicting fire spread in wildland fuels: General Technical report INT-115, USDA Forest Service, Intermountain Forest and Range Experiment Station. Sneeuwajgt & Peet 1985, Forest Fire Behaviour Tables for WA- Revised Edition., 3rd Edition edn, Forests Department of Western Australia. Sneeuwajgt (personal communication), R. 2004, Fire Operations Manager, Department of Conservation and Land Management, Perth, WA. Sneeuwajgt, R. 1998, 'Application of Wildfire Threat Analysis in South-Western Forests of WA', in 3rd International Conference on Forest Fire Research & 14th Conference on Fire and Forest Meteorology, ed. D. Viegas, pp. 2155-2176. Sobol, I. 1974, The Monte Carlo Method, The University of Chicago Press, Chicago. Standards Australia & Standards New Zealand 1995, Australian/New Zealand Standard: Risk Management, AS/NZS 4360:1995, Standards Australia. Steedman, R. 1987, 'A method for estimating the joint probability of occurrence of extreme storm water levels, with application to Exmouth Gulf', in 8th Australasian Conference on Coastal and Ocean Engineering, Launceston, Tasmania, pp. 28-32. Steedman, R. 2002, 'Wildfire and Risk Management', in Fire in Ecosystems of SouthWestern Australian Ecosystems: Impacts and Management, Department of Conservation and Land Management, Perth. The West Australian 2004, 'CALM must resist greens on burn-offs', The West Australian, Monday May 31, 2004, p. 14. Underwood, R. 1988, Fire Management and Prescribed Burning on CALM Lands- A Paper Prepared for the NPNCA, Department of Conservation and Land Management. 98 References Underwood, R. & Christensen, P. 1981, Forest Fire Management in Western Australia, Forests Department of Western Australia, Perth. United States Army Corps of Engineers 2003, Coastal Protection Manual EM1110-21100 Part V, United States Army Corps of Engineers. Vernon, D., Kiem, A. & Franks, S. 2004, 'Multi-decadal variability of forest fire riskeastern Australia', International Journal of Wildland Fire, vol. 13, pp. 165-171. Wallace (personal communication), D. 2004, Darren Wallace, Systems Analyst, Department of Conservation and Land Management (CALM), Perth, WA. Ward, D. 2000, Trouble in the Tuart: A brief fire history, Department of Conservation and Land Management, Perth. Ward, D. & Van Didden, G. 1997, Reconstucting the fire history of the Jarrah forest of south-western Australia, Department of Conservation and Land Management, Perth. Wardle, D., Hornberg, G., Zackrisson, O., Kalela-Brundin, M. & Coomes, D. 2003, 'Long-term effects of wildfire on ecosystem properties across an island area gradient', Science, vol. 300, no. 5621, pp. 972-975. Water and Rivers Commission 2000, Water Notes: Wetlands and Fire, Water and Rivers Commission, Government of Western Australia. Williams, K. & Mitchell, D. 2001, Jarrah Forest 1 (JF1- Northern Jarrah Forest region), Department of Conservation and Land Management. Wilson, B. 1994, 'Fire Effects on Vertebrate Fauna and Implications for fire management and conservation', in Fire and Biodiversity- The effects and effectiveness of Fire Management, Department of the Environment, Sport and Territories, Biodiversity Unit, Footscray, Melbourne. York, A. 1994, 'Long-term effects of fuel reduction burning on invertebrates in a dry sclerophyll forest', in Fire and Biodiversity- the effects and effectiveness of fire management, ed. J. Barnett, Department of the Environment, Sport and Territories Biodiversity Unit, Footscray, Melbourne. 99 Appendices 10. Appendices 10.1. MATLAB Scripts used in the analysis CALManalysis.m CALManalysis.m reads the raw wildfire data, creates probability distribution functions for cause of fire, number of fires with a burnt area greater than 10ha, FDI, fuel load, and surface moisture content (via pdf.m). The Monte Carlo simulation is carried out (via monte_carlo.m), and graphs of return period vs. intensity and the contributing variables (FDI, fuel load, surface moisture content) are produced. Note that the code requires two data files to run (‘date-cause-FDI-area.txt’ and ‘fuel_loads.txt’). Fuel_loads.txt is required to run the function m-file pdf.m. The code is presented in this appendix. %CALManalysis is the master file that creates the figures in the dissertation. clear all; no_simulations= input('enter number of simulations \n'); yrs= input('enter desired length (in years) of simulated record \n'); CC=50; %The line below allows the user to enter the canopy cover if it is activated: % CC= input('enter canopy cover (0-100%) to determine fuel load (default value- enter 50) '); %READ IN DATA-----------------------------------------------------------------------------[fire_date, cause, FDI, sum_area] = textread('date-cause-FDI-area.txt', '%s %d %d %f', 'delimiter', '\t'); %CREATE PDFs WITH GRAPHS-------------------------------------------------------------------%note that the function 'pdf.m' requires the use of the 'dateconverter.m' m-file. [p_greaterthan10, cum_FDIprob, FDI_bins, cum_fuel, litter_class, SMC_class]= pdf(fire_date, cause, FDI, sum_area, CC); %MONTE CARLO SIMULATION--------------------------------------------------------------------% fire_array= cell(1,no_simulations); for k=1:no_simulations %input desired number of years of simulation %use the function 'monte_carlo.m' to do the simulation [sim_num, sim_FDI, sim_fuel, sim_SMC, FQCF, ROS, sim_Inten]= monte_carlo(yrs,p_greaterthan10, cum_FDIprob, FDI_bins, cum_fuel, litter_class, SMC_class); %ANNUAL EXCEEDANCE SERIES--------------------------------------------------------------------------% select largest intensity fire from each year (column) in the matrix annual_ex_series(k,:)= max(sim_Inten); %do the same for the other variables: aes_FDI(k,:)= max(sim_FDI); aes_fuel(k,:)= max(sim_fuel); aes_SMC(k,:)= max(sim_SMC); 100 Appendices %sort the events in decreasing order sorted_aes(k,:)= -1.*sort(-1.*(annual_ex_series(k,:))); sorted_aes_FDI(k,:)= -1.*sort(-1.*(aes_FDI(k,:))); sorted_aes_fuel(k,:)= -1.*sort(-1.*(aes_fuel(k,:))); sorted_aes_SMC(k,:)= -1.*sort(-1.*(aes_SMC(k,:))); %assign a rank m to each element (largest -> m=1) %obtain return period using T= (N+1)/m where N is length of record for i=1:length(sorted_aes(k,:)) ret_period(i)= (length(sorted_aes)+1)/i; gumbel_y(i)= -log(log(ret_period(i)/(ret_period(i)-1))); end %Ranking fires by FDI, Intensity, and Fuel load %take the top x fires from each distribution: %need to convert from an array to a column matrix s=1; for i=1:yrs for j = 1:max(sim_num) col_sim_inten(k,s) = sim_Inten(j,i); col_sim_fuel(k,s)= sim_fuel(j,i); col_sim_SMC(k,s)= sim_SMC(j,i); col_sim_FDI(k,s)= sim_FDI(j,i); col_FQCF(k,s)= FQCF(j,i); s=s+1; end end aesfire_matrix = [col_sim_FDI(k,:);col_sim_SMC(k,:);col_sim_fuel(k,:); col_FQCF(k,:); col_sim_inten(k,:)]'; %sort the fires by intensity aes_sorted_sim_inten= -sortrows(-(aesfire_matrix), 5); %take the most intense fires (number of fires= ret_period) for each simulation k for i=1:length(ret_period) aes_sorted_by_inten(i,:)=aes_sorted_sim_inten(i,:); end %place the simulated data into an array aes_fire_array{k}= aes_sorted_by_inten; %ANNUAL MAXIMUM SERIES-------------------------------------------------------------------------------------% select largest intensity fire from each year (column) in the matrix annual_max_series(k,:)= max(sim_Inten); %do the same for the other variables: ams_FDI(k,:)= max(sim_FDI); ams_fuel(k,:)= max(sim_fuel); ams_SMC(k,:)= max(sim_SMC); %sort the events in decreasing order sorted_ams(k,:)= -1.*sort(-1.*(annual_max_series(k,:))); sorted_ams_FDI(k,:)= -1.*sort(-1.*(ams_FDI(k,:))); sorted_ams_fuel(k,:)= -1.*sort(-1.*(ams_fuel(k,:))); sorted_ams_SMC(k,:)= -1.*sort(-1.*(ams_SMC(k,:))); %Ranking fires by FDI, Intensity, and Fuel load 101 Appendices %take the top x fires from each distribution: %need to convert from an array to a column matrix for i=1:yrs for j=1:max(sim_num) column_sim_mat(i,1)=max(sim_Inten(:,i)); a=find(sim_Inten(:,i)==max(sim_Inten(:,i))); column_sim_mat(i,2)= sim_FDI(a(1),i); column_sim_mat(i,3)=sim_SMC(a(1),i); column_sim_mat(i,4)=sim_fuel(a(1),i); end end %sort the fires by intensity ams_sorted_sim_inten= -sortrows(-(column_sim_mat), 1); %place the simulated data into an array ams_fire_array{k}= ams_sorted_sim_inten; end %end 'k' loop %delete extraneous/"dummy" variables: clear ams_sorted_sim_inten aes_sorted_by_inten clear column_sim_mat clear col_sim_FDI col_sim_SMC col_sim_fuel col_sim_inten col_FQCF a clear sorted_sim_inten sorted_FQCF sorted_sim_FDI sorted_sim_SMC sorted_sim_fuel clear ams_sorted_sim_inten aes_sorted_sim_inten clear aesfire_matrix clear ams_FDI ams_SMC ams_fuel %OUTPUT PLOTS OF RESULTS------------------------------------------------------------------------%annual exceedance simulation %obtain relevant infomation from the array for i=1:no_simulations aes_FDI_matrix(:,i)= aes_fire_array{i}(:,1); aes_SMC_matrix(:,i)=aes_fire_array{i}(:,2); aes_fuel_matrix(:,i)=aes_fire_array{i}(:,3); aes_inten_matrix(:,i)=aes_fire_array{i}(:,5); end end %obtain mean from each simulation aes_mean_FDI_matrix= mean(aes_FDI_matrix'); aes_mean_fuel_matrix=mean(aes_fuel_matrix'); aes_mean_SMC_matrix= mean(aes_SMC_matrix'); aes_mean_inten_matrix= mean(aes_inten_matrix'); clear aes_fire_array; %plot the results of each aes simulation: figure semilogx(ret_period, aes_FDI_matrix, '.', ret_period, aes_mean_FDI_matrix, '-*', 'LineWidth', 2) ylabel('FDI') xlabel('Return period (yrs)') title('AES: Return period vs. Fire Danger Index (FDI)') figure semilogx(ret_period, aes_fuel_matrix, '.', ret_period, aes_mean_fuel_matrix, '-*','LineWidth', 2) ylabel('Fuel (t/ha)') 102 Appendices xlabel('Return period (yrs)') title('AES: Return period vs. Fuel load') figure semilogx(ret_period, aes_SMC_matrix, '.', ret_period, aes_mean_SMC_matrix, '-*','LineWidth', 2) axis([0, 10*max(ret_period), 0, max(max(aes_SMC_matrix))+1]); ylabel('SMC (%)') xlabel('Return period (yrs)') title('AES: Return period vs. Surface Moisture Content') figure semilogx(ret_period, aes_inten_matrix, '.', ret_period, aes_mean_inten_matrix, '-*', 'LineWidth', 2) ylabel('Intensity (kW/m)') xlabel('Return period (yrs)') title('AES: Return period vs. Intensity') %ANNUAL MAXIMUM SERIES--------------------------------------------------------------------------%obtain relevant infomation from the array for i=1:no_simulations ams_FDI_matrix(:,i)= ams_fire_array{i}(:,2); ams_SMC_matrix(:,i)=ams_fire_array{i}(:,3); ams_fuel_matrix(:,i)=ams_fire_array{i}(:,4); ams_inten_matrix(:,i)=ams_fire_array{i}(:,1); end %obtain mean from each simulation ams_mean_FDI_matrix= mean(ams_FDI_matrix'); ams_mean_fuel_matrix=mean(ams_fuel_matrix'); ams_mean_SMC_matrix= mean(ams_SMC_matrix'); ams_mean_inten_matrix=mean(ams_inten_matrix'); clear ams_fire_array; %plot the results of each ams simulation: figure semilogx(ret_period, ams_FDI_matrix, '.', ret_period, ams_mean_FDI_matrix, '-*', 'LineWidth', 2) ylabel('FDI') xlabel('Return period (yrs)') title('AMS: Return period vs. Fire Danger Index (FDI)') figure semilogx(ret_period, ams_fuel_matrix, '.', ret_period, ams_mean_fuel_matrix, '-*','LineWidth', 2) ylabel('Fuel (t/ha)') xlabel('Return period (yrs)') title('AMS: Return period vs. Fuel load') figure semilogx(ret_period, ams_SMC_matrix, '.', ret_period, ams_mean_SMC_matrix, '-*','LineWidth', 2) axis([0, 10*max(ret_period), 0, max(max(ams_SMC_matrix))+1]); ylabel('SMC (%)') xlabel('Return period (yrs)') title('AMS: Return period vs. Surface Moisture Content') figure semilogx(ret_period, ams_inten_matrix, '.', ret_period, ams_mean_inten_matrix, '-*', 'LineWidth', 2) ylabel('Intensity (kW/m)') xlabel('Return period (yrs)') title('AMS: Return period vs. Intensity') 103 Appendices %MITIGATION EXAMPLE: %ESTIMATING THE CLEARING DISTANCE-----------------------------------------------------------------%determine the intensity at cleared distances (verify against the APZ from NSWRFS- recommends around 30m) cleared_dist= [10, 30, 50]; for i=1:length(cleared_dist) %assume a 1/R^2 drop in intensity cleared_Int(i,:)= ams_mean_inten_matrix./cleared_dist(i).^2; end %set standard ignition intensity 40kW/m^2 (Babrauskas, 2001) for i=1:length(ret_period) standard(i) = 40; end %plot intensity vs. return period for various cleared distances (loglog scale) figure semilogy(ret_period, ams_mean_inten_matrix, ret_period, cleared_Int, ret_period, standard, '-.'); legend('no cleared distance', '10m cleared', '30m cleared', '50m cleared', 'ignition standard'); title('return period under different scenarios') xlabel('return period (yrs)') ylabel('intensity at value') %-------------------------------------------------------------------------------------------%how many fires can CALM actually fight? (based on CALM standards in the WTA) no_dir_attack_std= 3000; no_machine_attack_std= 2000; no_hand_attack_std= 800; no_dir_attack=0; no_machine_attack= 0; no_hand_attack=0; total_sim_fires= 0; for i=1:length(sim_Inten(1,:)) for j=1:length(sim_Inten(:,1)) if sim_Inten(j,i) ~= 0 total_sim_fires= total_sim_fires+1; if sim_Inten(j,i) > no_dir_attack_std no_dir_attack = no_dir_attack+1; end if sim_Inten(j,i) > no_machine_attack_std no_machine_attack= no_machine_attack+1; end if sim_Inten(j,i) > no_hand_attack_std no_hand_attack= no_hand_attack+1; end end end end %calculate the % of fires in the record that exceed CALM's thresholds (Muller '93) percent_unfightable= 100*no_dir_attack/total_sim_fires; percent_no_machine_att= 100*no_machine_attack/total_sim_fires; percent_no_hand_attack= 100*no_hand_attack/total_sim_fires; clear no_dir_attack_std no_machine_attack_std no_hand_attack_std no_dir_attack clear no_machine_attack no_hand_attack 104 Appendices Pdf.m Pdf.m creates probability distribution functions for the FDI, the fuel load, the surface moisture content and the number of fires. It also generates a probability mass function for the cause of fire. function [p_greaterthan10, cum_FDIprob, FDI_bins, cum_fuel, litter_class, SMC_class]= pdf(fire_date, cause, FDI, sum_area, CC) %-CAUSE OF FIRE------------------------------------------------------------%create histogram of fire causes %initiate fire cause variables by cause: deliberate= 0; calm_escape= 0; escape_burn_off =0; rec_accident= 0; ti_accident =0; lightning =0; other_accident=0; other=0; unknown=0; %count the number of fires that occur due to each cause: for i=1:length(cause) if cause(i) == 1 deliberate = deliberate+1; elseif cause(i) == 2 calm_escape= calm_escape+1; elseif cause(i)== 3 escape_burn_off= escape_burn_off+1; elseif cause(i)== 6 rec_accident= rec_accident+1; elseif cause(i)== 4 ti_accident= ti_accident+1; elseif cause(i)== 5 other_accident= other_accident+1; elseif cause(i)== 7 lightning = lightning+1; elseif cause(i)== 9 other= other+1; elseif cause(i)== 8 | cause(i)== 0 unknown = unknown+1; end end ycause= [deliberate, calm_escape, escape_burn_off, rec_accident, ti_accident, other_accident, lightning, other, unknown]; for i=1:length(ycause) prob_cause(i)= ycause(i)/sum(ycause); end figure bar(prob_cause,1) 105 Appendices colormap hsv % legend(xcause); xlabel('Cause') ylabel('Probability of ignition') title('Probability mass function of fire ignition by cause') %-------FIRE DANGER INDEX (weather)-------------------------------------------------------------%if FDI>3000, then assume an error and force FDI=3000 count_extreme_FDI=0; for i=1:length(FDI) if FDI(i)>3000 FDI(i)=0; count_extreme_FDI= count_extreme_FDI+1; end end %create a histogram of FDI (aggregate weather factors) %set bin width h h=20; %create bins FDI_bins = 0:h:max(FDI)+h; %create a histogram with bins centred on elements of FDI_bins=FDIclass, and frequency FDIfreq [FDIfreq, FDIclass]=hist(FDI,FDI_bins); FDIprob= FDIfreq./sum(FDIfreq); %plot the histogram figure bar(FDI_bins, FDIprob,1); title('Probability Distribution Function for Fire Danger Index (FDI)') xlabel('Fire Danger Index') ylabel('Probability') axis([-h/2,max(FDI_bins),0,max(FDIprob)+0.1]); % %create cumulative histogram ready to fit a distribution FDI_counter=0; for i=1:length(FDIprob); cum_FDIprob(i)=FDIprob(i)+FDI_counter; FDI_counter= cum_FDIprob(i); end %plot the cumulative histogram figure plot(FDI_bins, cum_FDIprob); title('Cumulative Probability Distribution Function for Fire Danger Index (FDI)') xlabel('Fire Danger Index') ylabel('Probability') axis([-h/2,max(FDI_bins),0,1]); clear h; clear FDI; clear FDI_counter; %-FUEL-----------------------------------------------------------------------------------------%read in the fuel load data years_since_burn = textread('fuel_loads.txt', '%d', 'delimiter', '\t'); %equations only valid for years_since_burn <30 -> reduce all years_since_burn that are greater 106 Appendices %than 30 to equal 30, or resultant annual max intensities will be too large to be realistic for i= 1:length(years_since_burn) if years_since_burn(i) > 30 years_since_burn(i)= 30; end end %convert from years since burn to fuel load: Use eqn for available litter from Beck p334, Karri forest prediction eqn %for jarrah, available fuel factor AFF =1, assume canopy cover =50% AFF= 1; NJ_L_wt= (0.18.*CC+11.06).*(1-exp(-0.086.*years_since_burn)); Avail_litter = NJ_L_wt.*AFF; %available litter is in tonnes per ha % define frequency intervals for histogram litter_class= 0:1:max(Avail_litter)+1; % create a histogram with bins centred on elements of litter_weight, and frequency fuelfreq [fuelfreq, litter_weight] = hist(Avail_litter, litter_class); %normalise to create probability of a fuel load in a given area: fuelprob= fuelfreq./sum(fuelfreq); %plot the histogram of fuel load frequency bar(litter_weight, fuelprob,1); xlabel('litter weight (t/ha)'); ylabel('probability'); title('probability of a fuel load'); %create a cumulative histogram of fuel load frequency: fuel_counter= 0; for i=1:length(fuelprob) cum_fuel(i)= fuelprob(i)+fuel_counter; fuel_counter= cum_fuel(i); end clear fuel_counter; % figure % plot(litter_weight, cum_fuel) % xlabel('litter weight (t/ha)'); % ylabel('cumulative probability'); % title('cumulative probability of a fuel load occurring'); %-NUMBER OF FIRES------------------------------------------------------------------------------%create a histogram of number of fires/year %firstly convert string cell array into numerical form for i=1:length(fire_date) date_matrix(i,:)= dateconverter(fire_date(i)); end %determine the number of fires with an area greater than or equal to 10ha in each year area_10plus= zeros(length(sum_area),1); for i=1:length(sum_area) if sum_area(i)>=10 %this is the place we set the arbitrary 10ha area limit area_10plus(i)=date_matrix(i,3); end end %create a histogram then drop all the zero values (0 = fire had an area less than 10ha) 107 Appendices %define the length of the record for the histogram: yearspan_1= 1980:1993; yearspan_2= 2000:2003; yearspan= [0,yearspan_1,yearspan_2]; [ann_freq1, ann_class1]=hist(area_10plus,yearspan); %drop the zero values for i=1:length(ann_freq1)-1 ann_freq(i)= ann_freq1(i+1); ann_class(i)= ann_class1(i+1); end clear ann_freq1; clear ann_class1; %allocate consecutive numbers to yearspan and plot number of fires in each year year= 1:length(yearspan)-1; % bar(year, ann_freq); % xlabel('year number') % ylabel('number of fires') %now need to plot the probability that, in a given year, more than 'n' fires occur that are greater than 10ha. %assume that number of fires is Poisson distributed- determine the sample mean number of fires mu_number= sum(ann_freq)/length(year); % %check using built-in MATLAB function % p_mean= poissfit(ann_freq) %probabilities are given by (exp(-mean)*mean^year)/year! %e.g. prob of 36 fires in a year p_36= (exp(-mu_number)*mu_number^36)/factorial(36) for i=1:max(ann_freq) prob_number(i)= (exp(-mu_number)*mu_number^i)/factorial(i); end %now obtain the probability that the number of fires in a year is at least n: =1- P(X<n) %where P(X<n)= sum(p(i))*delta(x(i)). Since delta(x)=1, then this is just sum(p(i)) number_counter= 0; % create a cumulative histogram to get P(X<n): for i=1:length(prob_number) p_lessthan10(i)=prob_number(i)+number_counter; number_counter= p_lessthan10(i); end clear number_counter; %obtain P(X>n), the probability of obtaining n fires with an area greater than 10ha in a given year p_greaterthan10= 1-p_lessthan10; figure plot(p_greaterthan10); title('Annual number of fires with an area greater than 10ha') xlabel('number of fires with an area exceeding 10ha'); ylabel('probability of occurrence in a given year'); clear area_10plus; clear fire_date; %SMC ------------------------------------------------------------------------------%no info on these variables from CALM- need to sample from a uniform distribution %SMC (varies 3-25% table 4 Beck) SMC_class= 3:25; 108 Appendices SMCprob= (1/(length(SMC_class)-1)).*(ones(1, length(SMC_class))); figure plot(SMC_class, SMCprob, 'LineWidth', 3) axis([0, max(SMC_class)+2, 0,0.1]) title('Probability distribution function for Surface Moisture Content') xlabel('Surface Moisture Content (%)') ylabel('Probability') Dateconverter.m Dateconverter.m is a simple function m-file used by pdf.m to format the date field of the data files into the desired format when determining the pdf for the number of fires per year. function [day, month,year]= dateconverter(datestring) % dateconverter.m converts strings of dates from the form dd/mm/yy to date, month, and year %numeric values if length(char(datestring)) == 9 datestring1= char(datestring); ch_day= datestring1(1); ch_month= strcat(datestring1(3),datestring1(4)); ch_year= strcat(datestring1(6), datestring1(7), datestring1(8), datestring1(9)); elseif length(char(datestring)) == 10 datestring1 = char(datestring); ch_day= strcat(datestring1(1), datestring1(2)); ch_month= strcat(datestring1(4),datestring1(5)); ch_year= strcat(datestring1(7), datestring1(8), datestring1(9), datestring1(10)); end char(ch_day); day= [str2num(ch_day), str2num(ch_month), str2num(ch_year)]; Monte_carlo.m Monte_carlo.m is the function m-file used by CALManalysis.m to create simulated fire intensities by random simulation for the desired number of years of simulated record. function [sim_num, sim_FDI, sim_fuel, sim_SMC, FQCF, ROS, sim_Inten]= monte_carlo(yrs, p_greaterthan10, cum_FDIprob, FDI_bins, cum_fuel, litter_class, SMC_class) 109 Appendices %monte_carlo.m completes the monte carlo simulation for any number of years (input number of years in 'yrs') %-MONTE CARLO SIMULATION------------------------------------------------------------------------%randomly generate number of fire years %generate 75 years % yrs=75; N = rand([1, yrs]); %convert these random numbers (probabilities) to number of fires from the distribution above for i=1:length(N) for j=1:length(p_greaterthan10) if N(i)>p_greaterthan10(j) sim_num(i)=j; %sim_num gives the number of fires for each simulation year break end end end %randomly generate FDI and fuel values for each fire in each year sim_FDI= zeros(max(sim_num), length(N)); %initialise a matrix to store FDI values sim_fuel= zeros(max(sim_num), length(N)); %initialise a matrix to store fuel values sim_SMC= zeros(max(sim_num), length(N)); %initialise a matrix to store SMC values for i=1:length(N) M= rand([1, sim_num(i)]); %generate random probability of each fire occurring in a given year H = rand([1, sim_num(i)]); L = rand([1, sim_num(i)]); for j=1:length(M) %FDI for m= 1:length(cum_FDIprob) if M(j)<cum_FDIprob(m) %use pdf to get corresponding FDI sim_FDI(j,i)= FDI_bins(m); break end %end 'if' loop %Fuel for n= 1:length(cum_fuel) if L(j)<cum_fuel(n) sim_fuel(j,i)=litter_class(n); break end end %SMC sim_SMC(j,i)= min(SMC_class) + (max(SMC_class)-min(SMC_class)).*H(j); %assume uniformly dist end end end %end 'm' loop %end 'j' loop %end 'i' loop %Calculate FQCF (fuel quantity correction factor) since fuel load not standard 8t/ha Jarrah (Beck 1995) for i=1:max(sim_num) for j=1:length(N) 110 Appendices if sim_fuel(i,j) <= 8 & sim_SMC(i,j) <=26 & sim_SMC(i,j) >=3 FQCF(i,j)= 1.02/(1+7266.83*exp(-1.36*sim_fuel(i,j)))+0.1; elseif (sim_SMC(i,j) <=9) & (sim_fuel(i,j) > 8) & sim_SMC(i,j)>3 FQCF(i,j)= (6.03+5.81*sim_fuel(i,j))/53.44; elseif (sim_SMC(i,j)>9) & (sim_SMC(i,j) <= 18) & (sim_fuel(i,j) > 8) FQCF(i,j)= (11.19+2.92*sim_fuel(i,j))/35.02; elseif (sim_SMC(i,j) > 18) & (sim_fuel(i,j) > 8) FQCF(i,j)= (0.055+0.0023*sim_fuel(i,j))/0.074; elseif sim_SMC(i,j)==0; FQCF(i,j)=0; end end end %obtain the ROS (m/h)for a non-sloping terrain (Beck 1995) ROS= FQCF.*sim_FDI; %can now obtain intensities for all the fires: I= HwR= 0.47*ROS*Weight units are kW/m sim_Inten= 0.47.*ROS.*sim_fuel; Correlations.m Correlations.m creates plots showing the correlations between FDI, burnt area and fuel load. The correlation coefficient between each variable is calculated. %correlations determines the correlations between fuel, burnt area and FDI and creates %plots of the correlations clear all; %FULL DATA SET CORRELATION-----------------------------------------------------------------------------%read in the data [fire_date, cause, FDI, sum_area] = textread('date-cause-FDI-area-update.txt', '%s %d %d %f', 'delimiter', '\t'); for i=1:length(fire_date) date_matrix(i,:)= dateconverter(fire_date(i)); end month= date_matrix(:,2); clear fire_date; %create a matrix of the variables A= [FDI, sum_area, cause, month]; %now correlate the variables using the covariance 111 Appendices covariance_matrix= cov(A); %correlation between the variables correlation_matrix= corrcoef(A); %FUEL CORRELATIONS------------------------------------------------------------------------------------%don't have a complete fuel load data set- correlate with the existing fuel age data set: [fuel_date, fuel_area, fuel_FDI, fuel_cause, fuel_years] = textread('fuel_load_info.txt', '%s %f %d %d %d', 'delimiter', '\t'); for i=1:length(fuel_date) fuel_date_matrix(i,:)= dateconverter(fuel_date(i)); end fuel_month= fuel_date_matrix(:,2); clear fuel_date_matrix; clear fuel_date; %ensure fuel load is <= 30 for i=1:length(fuel_years) if fuel_years(i)>30 fuel_years(i)=30; end end %create a matrix of the variables B= [fuel_FDI, fuel_area, fuel_cause, fuel_month, fuel_years]; %now correlate the variables using the covariance fuel_covariance_matrix= cov(B); %correlation between the variables fuel_correlation_matrix= corrcoef(B); clear i A B %check the correlations between variables by plotting %FDI and Area burnt loglog(sum_area, FDI, '*'); xlabel('Area burnt (ha)') ylabel('FDI') title('FDI vs. Area burnt') 112 Appendices figure loglog(fuel_years, fuel_FDI, '*') xlabel('Fuel age (yrs)') ylabel('FDI') title('Fuel age vs. FDI') figure loglog(fuel_years, fuel_area, '*') xlabel('Fuel age (yrs)') ylabel('Area burnt (ha)') title('Fuel age vs. Area burnt') 113