The Black Saturday Kilmore East Fire in Victoria, Australia

Home , Black Friday bushfires, Eucalyptus regnans

Forest Ecology and Management xxx (2012) xxx–xxx

Contents lists available at SciVerse ScienceDirect

Forest Ecology and Management

journal homepage: www.elsevier.com/locate/foreco

Anatomy of a catastrophic wildfire: The Black Saturday Kilmore East fire in Victoria, Australia ⇑ M.G. Cruz a, , A.L. Sullivan a, J.S. Gould a, N.C. Sims b, A.J. Bannister c, J.J. Hollis a,d,e, R.J. Hurley a a CSIRO Ecosystem Sciences and CSIRO Climate Adaptation Flagship, GPO Box 1700, Canberra, ACT 2601, Australia b CSIRO Land and Water, Private Bag 10 Clayton Sth, VIC 3169, Australia c Bureau of Meteorology, GPO Box 1636, Melbourne, VIC 3001, Australia d University of New South Wales at the Australian Defence Force Academy, Canberra, ACT 2600, Australia e Bushfire Cooperative Research Centre, Level 5, 340 Albert Street East Melbourne, VIC 3002, Australia article info abstract

Article history: The 7 February 2009 wildfires in south-eastern Australia burned over 450,000 ha and resulted in 173 Received 28 October 2011 human fatalities. The Kilmore East fire was the most significant of these fires, burning 100,000 ha in less Received in revised form 21 February 2012 than 12 h and accounting for 70% of the fatalities. We report on the weather conditions, fuels and prop- Accepted 28 February 2012 agation of this fire to gain insights into the physical processes involved in high intensity fire behaviour in Available online xxxx eucalypt forests. Driven by a combination of exceedingly dry fuel and near-gale to gale force winds, the fire developed a dynamic of profuse short range spotting that resulted in rates of fire spread varying Keywords: between 68 and 153 m minÀ1 and average fireline intensities up to 88,000 kW mÀ1. Strong winds aloft Megafire and the development of a strong convection plume led to the transport of firebrands over considerable Wildland–urban interface Crown fire distances causing the ignition of spotfires up to 33 km ahead of the main fire front. The passage of a wind Eucalyptus change between 17:30 and 18:30 turned the approximately 55 km long eastern flank of the fire into a Spotting headfire. Spotting and mass fire behaviour associated with this wide front resulted in the development Pyrocumulonimbus of a pyrocumulonimbus cloud that injected smoke and other combustion products into the lower stratosphere. The benchmark data collected in this case study will be invaluable for the evaluation of fire behaviour models. The study is also a source of real world data from which simulation studies investigat- ing the impact of landscape fuel management on the propagation of fire under the most severe burning conditions can be undertaken. Ó 2012 Published by Elsevier B.V.

1. Introduction The fires that occurred on 7 February 2009 (colloquially known as ‘Black Saturday’), represent 44% of the fatalities. Of a total of 316 South-eastern (SE) Australia has a combination of climate, fires burning on this date, 13 developed into significant incidents topography and vegetation that makes it prone to severe wildfires. (Fox and Runnalls, 2009) and five resulted in 173 fatalities. The Kil- Fires occur in most years but are generally most extensive and se- more East fire was the most significant of these, resulting in 70% of vere following extended drought, typically associated with El-Nino the fatalities on the day. It burnt nearly 100,000 ha and destroyed events (Sullivan et al., 2012). This region has a long history of severe over 2200 buildings in the first 12 h alone. The fire eventually fire events, some of which have significantly influenced wildland merged with the Murrindindi fire, burning a combined area of fire control and land management policy. In the past seven decades approximately 400,000 ha over a period of 3 weeks. catastrophic fire events (defined here as fire in which at least a sin- Understanding the development and behaviour of the Kilmore gle day of high intensity fire behaviour occurs and generally results East fire is important for a number of reasons. It is a critical step in large area burned with significant destruction of infrastructure in identifying the factors that led to the scale of this catastrophic fire and loss of life) have impacted SE Australia in 1939 (Black Friday), and its unprecedented impact on lives, livelihoods and ecosystem 1983 (Ash Wednesday), 2003 (Canberra) and 2009 (Black Saturday). components. Despite the diverse adaptation of Australian ecosys- These four fire events have burnt 7.68 Mha of land and caused 390 tems to fire (Gill, 1981a,b), large-scale fires can have detrimental fatalities, predominantly in the state of Victoria. impacts on ecological values. Such a fire converts biodiversity-rich, fine-scale mosaics at a range of seral states into a less diverse landscape, both in terms of species composition and vegetation

⇑ Corresponding author. Tel.: +61 2 6246 4219; fax: +61 2 6246 4096. structure (Adams and Attiwill, 2011). The sustainable management E-mail address: [email protected] (M.G. Cruz). of SE Australian ecosystems requires a landscape level approach to

0378-1127/$ - see front matter Ó 2012 Published by Elsevier B.V. http://dx.doi.org/10.1016/j.foreco.2012.02.035

Please cite this article in press as: Cruz, M.G., et al. Anatomy of a catastrophic wildfire: The Black Saturday Kilmore East fire in Victoria, Australia. Forest Ecol. Manage. (2012), http://dx.doi.org/10.1016/j.foreco.2012.02.035 2 M.G. Cruz et al. / Forest Ecology and Management xxx (2012) xxx–xxx fuel management that mitigates the impact of catastrophic fire events. Sound knowledge of fire potential under extreme fire In the absence of any break in fuel or topography, isolated med- weather conditions is necessary to understand the effectiveness of ium-range spot fires are generally overrun by the main fire different land management strategies under similar burning front.When a pattern of concentrated medium range spotting conditions. develops, pseudo flame fronts (McArthur, 1967) lead to an Fire in eucalypt forests exhibit unique behaviour because of the immediate large increase on the overall rate of fire spread. Con- contribution of bark as a fuel (McArthur, 1967). Under extreme fire centrated medium range spotting can produce mass fire or fire- weather conditions bark fuels allow for profuse spotting dynamics storm effects (Luke and McArthur, 1978). In this situation a large (Box 1) that lead to very high rates of spread and occasionally fire- number of coalescing fires causes strong turbulent inflow circu- storm effects (Cheney and Bary, 1969). Quantitative knowledge of lation that results in high intensity burning.Long-distance spot- the processes driving these phenomena is lacking. Field-based fire ting (>5000 m) results from extended flight paths associated behaviour research in Australia has produced a number of models with significant lofting in a well-developed convection column (e.g., McArthur, 1967; Sneeuwjagt and Peet, 1985; Cheney et al., and long burn out times. This class of spotfire generally creates 1998; Gould et al., 2007a; Burrows et al., 2009) that are currently an isolated ignition that develops as a separate fire. Long range used to support forest and fire management activities. Nonethe- spotting of approximately 30 km has been authenticated in sev- less, data from these research programs only cover the lower 10% eral occasions in eucalypt forests (Hodgson, 1967; McArthur, of the fire intensity spectrum observed in wildfires (Cheney, 1967). Transport of firebrands over such distances requires À1 1991). Case studies documenting the propagation of large fires of- upper level wind speeds in the vicinity of 90–100 km h (Luke fer an opportunity to quantitatively describe fire behaviour at the and McArthur, 1978).The dynamics of high intensity fire propa- extreme end of the fire intensity spectrum (Alexander and Thomas, gation in eucalypt forests with fibrous barked species is charac- 2003). The objective of the present study was to describe the prop- terised by a self-sustained process with profuse short range agation and behaviour of the Kilmore East fire and gain insights spotting preceding the arrival/formation of a solid flame front into the physical processes determining the propagation of high involving the whole fuel complex.The distribution of ignitions, intensity fires in eucalypt forests. with density decreasing with distance from the flame front and rapid coalescence by interacting spot fires results in a grada- tion of area involved in combustion at any given moment. The Box 1.High intensity fire propagation in eucalypt forests. closer to the firebrand source the larger the area on fire. As the spot fires coalesce and more ignitions are generated the area There are notable differences in the mechanisms driving burning approaches a limit where the amount of released pyro- high intensity fire propagation in eucalypt forests to other lizates mixes with the air at a critical mixture that result in the fuel types such as conifer forests or shrublands. These differ- formation of a continuous flame front. The concept of a discrete ences arise from particular fuel complex characteristics in eucalypt forests. Most temperate eucalypt forests have fairly flame front separating unburned from burning fuels as observed open canopies (McArthur, 1967, Gill, 1997) that allow the in other fuel types such as conifer forest and shrublands is not development of an understorey layer of dominated trees, directly applicable in this case. What is observed is a continuous shrubs and/or herbaceous vegetation that provide vertical increase on the proportion of area burning until a continuous fuel continuity. In eucalypt fuel complexes the presence of flame front is formed. tree species with fibrous bark (e.g., Eucalypt obliqua, E. marginata and E. macrorrhyncha) is a key factor driving fire behaviour. Fibrous bark particles are easily ignited and dislodged allowing simultaneously for vertical fire propagation and pro- 2. Methods fuse spot fire ignitions. Eucalypt species with smooth decorticating bark (e.g., Eucalyptus viminalis, E. globulus and E. delegatensis) provide aerodynamically efficient, firebrand 2.1. Fuels material that can be transported over considerable distances. Spotting is the dominant fire propagation process in high The general region of the fire (Fig. 1) encompasses a wide range intensity fires in temperate eucalypt forests (Cheney and of vegetation types, from grassland (Cheney and Sullivan, 2008) Bary, 1969). Spotting can be classed into three categories and woodland (Gill, 1981a) in the lower elevations, tall wet-sclero- based on distance and density distribution. phyll eucalypt forests (Ashton and Attiwill, 1984) in the upper ele- Short-distance spotting (including ember showers) in- vations of the Hume range, to temperate rainforest (Busby and cludes all spotfires up to 750 m and is generally the result Brown, 1984) in localised gullies. Ashton (2000) provides a com- of embers and firebrands blown directly ahead of the fire with prehensive description of topography, climate and vegetation of little to no lofting. Short-range spotting density tends to de- the region impacted by the fire. Ten distinct fuel types (Table 1) crease with distance from the fire front (Cheney and Bary, were identified from a regional ecological vegetation classification 1969). In moderate to high intensity (3000–5500 kW mÀ1) experimental fires in eucalypt jarrah forests Gould et al. (Parkes et al., 2003) based on key characteristics of the forest and (2007b) observed spotting densities ranging between 3 and understorey that influence fire behaviour: available fuel for com- 8 firebrands per square meter occurring 50 m ahead of the bustion, fuel dryness, understorey wind reduction, and contribu- flame front. At distances of approximately 100 m from the fire tion of bark to spotting. Quantitative fuel complex characteristics front the firebrand density was still around 1 firebrand per were assessed in representative unburned areas surrounding the square meter. Under drier and windier burning conditions fire area. Of the ten fuel types in Table 1, visual fuel assessment higher spotting densities are expected as litter fuels are more was conducted for five forest fuel types: (i) dry sclerophyll forest susceptible to ignition and more firebrands are transported in with low understorey; (ii) dry sclerophyll forest with dense under- flatter trajectories. storey; (iii) mixed dry-wet sclerophyll forest; (iv) wet sclerophyll Medium-distance spotting (1000–5000 m) results from em- forest; and (v) eucalypt plantations. bers and firebrands being lofted briefly in the convection column, blown directly out of tree tops from a ridge without Understorey and bark fuels were described at representative being lofted or from the collapse of the convection column sites one year after the fire using the methodology in Gould et al. at a break in fuel or topography (Luke and McArthur, 1978). (2007a, 2011). Fuel quantity was estimated based on the equiva- lent fuel loads for the observed fuel hazard ratings (Gould et al.,

Fig. 1. General ﬁre location with topography, Bureau of Meteorology Automatic Weather Station (AWS) locations, and wind change isochrones.

Table 1 Vegetation and fuel description of the main fuel types present in the Kilmore East ﬁre area.

Vegetation/fuel types Description Grassland/open woodland Grassland and open woodland forest cover Eucalypt plantation Industrial plantations of fast growing species eucalypt species for pulping. Main species found in the study area were Eucalyptus globulus and other eucalyptus species. Variable age, but mostly under 10 years Pine plantation Industrial plantations of Pinus radiata. Variable age Dry sclerophyll-low Forest consisting of a medium to tall overstorey of mixed eucalypt trees with heights varying between 20 and 25 m. Main overstorey understorey trees species are peppermints (E. dives, E. radiata) and stringybarks (E. obliqua, E. macrorhyncha). Understorey tree layer is absent. Understorey consisting of sparse to moderate cover of shrubs and a high cover of herbs and grasses Dry sclerophyll-dense Forest consisting of mixed eucalypt trees up to 25 m tall occupying exposed aspects. Overstorey tree composition includes peppermints understorey (E. dives, E. radiata) and gums (E. mannifera sp.). Understorey tree layer is absent. A well-developed medium height to tall shrub layer is present Riparian forest dense Tall forest (up to 30 m) occupying fertile and moist riparian areas. Overstorey consists of messmate stringybark (E. obliqua) and manna understorey gum (E. viminalis). A secondary tree understorey of wattles (Acacia ssp.) is present. Lower understorey has a dense cover of tall shrubs, ferns and grasses/herbs Mixed dry–wet sclerophyll Tall and dense eucalypt tree overstorey up to 30 m height of mixed species from the wet and dry forest types. Dominant tree species are forest mountain grey-gum (E. cypellocarpa), messmate stringybark (E. obliqua) and eurabbie (E. globulus spp. Bicostata). A dense understorey of tall shrubs is also present above scattered herbs and grasses Wet sclerophyll forest Tall eucalypt forest (up to 40 m) dominated by mountain ash (E. regnans) with upper understorey of trees. A tall shrubby layer occupies the lower understorey below which a moist, fern-dominated layer developed. This fuel type is mostly restricted to higher elevation areas or protected gullies Rainforest Dense, multi-storey non-eucalypt forest occurring in high rainfall areas in the absence of ﬁre. Dominant tree species are myrtle beech (Nothofagus cunninghamii) and southern sassafras (Atherosperma moschatum). An understorey layer is dominated by ferns. This fuel type is found mostly in deep gullies Residential A mix of low spread potential fuels characterised by lack of spatial continuity and low fuel load intermixed with areas devoid of fuel. Examples of low spread potential fuels encountered in this fuel complex are irrigated gardens and overgrazed paddocks close to houses. The broad scale spread potential in this area is nil due to fuel discontinuity

2011 and CSIRO unpublished data). Downed woody (d (diameter) (200, 500, 700 and 1000 m). Hourly maps of FDFMC were calcu- >6 mm) fuel load was determined using the linear intersect meth- lated. A detailed description of this modelling approach is given od (Van Wagner, 1968) at four sites (Hollis et al., 2011a). Fuel in Sullivan and Matthews (submitted for publication). structure for grassland and radiata pine (Pinus radiata) plantation was derived from Cheney et al. (1998) and Cruz and Plucinski 2.4. Fire propagation and behaviour (2007), respectively. Understorey fuels were not assessed in riparian forest or rainforest areas. Instead it was assumed that the Data sources used to reconstruct the propagation and behaviour understorey fuel characteristics of riparian forests were similar to of the fire included infrared (IR) linescans, digital photographs, vi- those of mixed dry-wet sclerophyll forest, and those of rainforest deo footage, witness statements to the Victorian Police, and inter- areas were similar to wet sclerophyll forest. The resulting loss in views with eyewitnesses and fire/land management agency staff. precision was considered negligible given other sources of varia- tion and the marginal contribution of these fuel types to fire 2.4.1. Infra-red linescans propagation. An aircraft carrying an infra-red linescanner flew over the fire to Post-fire fuel consumption was estimated at four sites at which map the active fire perimeter. Two line-scans were obtained during the full canopy was burnt. Three sites were in dry sclerophyll forest the early afternoon on 7 February, with one passage at 12:46 and with low understorey and one site was in mixed dry-wet sclero- another at 12:55. Between 22:00 and 22:40 a number of linescans phyll forest. Downed woody fuel consumption was determined were taken on the northern perimeter of the fire. No linescans were through comparison of paired burnt (as a crown fire) and unburnt undertaken between 13:00 and 22:00. Post-processed and geo-rec- sites in close proximity within the same fuel type of similar slope, tified line-scans were made available by the Department of Sus- aspect and stand structure. Details of the fuel consumption meth- tainability and Environment (DSE), Victoria. odology are given in Hollis et al. (2011a). 2.4.2. Eye-witness interviews Interviews were conducted with a number of people involved in 2.2. Weather the suppression activities on 7 February, including firefighters, fire managers, fire lookout operators and air attack supervisors from Fire behaviour was interpreted in relation to weather observa- the Country Fire Authority (CFA), DSE, Parks Victoria and Melbourne tions from Commonwealth Bureau of Meteorology automatic Water. Over 110 individuals were interviewed. Each interview was weather stations (AWS) at Melbourne Airport, Kilmore Gap and conducted following an established protocol aimed at collecting Coldstream. Two datasets were collated: (1) daily (12:00 EDST1) information on a range of topics including: fuels influencing fire weather data (air temperature, relative humidity, wind speed and propagation; level of fire activity (e.g., flaming characteristics, sur- direction, and precipitation (accumulated until 09:00 as per World face fire, crowning, smoke and plume characteristics); spotting Meteorological Organization standard) for a period of 2 years pre- occurrence and characteristics; time of relevant events and weather ceding the fire and (2) half-hourly weather data (air temperature, conditions. Associated with these interviews, a large number of relative humidity, precipitation and wind speed and direction) for photographs and video footage was collected. Other sources of the duration of the fire. Additional data sourced from the Bureau information included witness statements provided to the Victoria of Meteorology included the Melbourne Airport radiosonde observa- Police, DSE and CFA radio logs, a summary of emergency calls with tion collected at 11:00, weather radar data from the Melbourne ra- relevant information, and fire incident maps sketched by firefight- dar, and a wind change map derived from Doppler radar and AWS ing personnel. observations (Bureau of Meteorology, 2009). The daily data were used to characterise the antecedent weath- 2.4.3. Fire isochrones er conditions leading to the fire. The Keetch–Byram Drought Index, The information gathered from each source was geo-referenced KBDI (Keetch and Byram, 1968) and the Drought Factor (McArthur, and classified according to its reliability (see Table A1) and rele- 1967; Noble et al., 1980) were calculated to characterise the fire vance. This geospatial information was then used to delineate fire season dryness. isochrones that describe the fire perimeter growth. The isochrones are considered approximate and represent the formation of a con- 2.3. Fine dead fuel moisture content tinuous flame zone (see Box 1).

Fuel moisture is the most dynamic component of fuels and 2.5. Fire severity plays a significant role in determining fire behaviour. No measurements of fine dead fuel moisture content (FDFMC) were taken on Three Landsat Thematic Mapper (TM 5) images captured on 31 the 7 February. A processed-based model of fine surface and profile January (pre-fire), 16 February (early post-fire) and 21 April 2009 forest fuel moisture content (Matthews, 2006) was used in con- (late post-fire) were acquired for burn severity assessment. These junction with an empirical model of grassland moisture content images were corrected to top-of-atmosphere reflectance using (McArthur, 1966; Cheney et al., 1998) to provide an indication of the Landsat TM pre-processing algorithm in ENVI 4.5 image pro- the likely moisture content of fine dead fuel during the 7 February. cessing software (ITT Industries, 2008), which reduces the influ- Two categories of forest litter fuel (dry woodland and tall wet for- ence of solar illumination conditions and detector sensitivity est) and one of grass (assuming full curing) were employed. variations on pixel values between images. Weather data from the Kilmore Gap AWS were used as the basis Several methods for burn severity mapping were tested. These for this simulation. included creating normalised burn ratio images (NBR; Key and FDFMC models are not inherently spatial, although the forest Benson, 2002), NBR difference images calculated by subtracting litter model does include spatial variables such as solar radiation. the postfire NBR image from the prefire NBR image, unsupervised To apply the models to the landscape of the fire, it was necessary classification of the NBR and NBR difference images, appending a to categorise the landscape according to aspect (45°, 135°, 225° normalised difference vegetation index (NDVI) image to the NBR and 315°), degree of slope (0°,5°,15° and 30°) and elevation images before classification, unsupervised classification of the early and late post fire images and classification of various other 1 All times in this report are Australian Eastern Daylight Savings Time (UTC + 11 h). combinations of the pre- and post-fire images. A range of

Table 2 Remote sensed burn severity class description.

Burn Description Fire type severity class 1 75–100% Crown defoliation/ Active crown fire consumption 2 25–75% Crown defoliation/ Intermittent crown consumption fire 3 <25% Crown defoliation/consumption High intensity or >60% crown scorch surface fire 4 30–60% Crown scorch; nil crown Moderate intensity consumption surface fire 5 0–30% Crown scorch Low intensity surface fire 6 Burnt grass Surface fire 7 Unburnt No fire

Table 3 Summary of understorey and bark fuel load and height for dry sclerophyll forest – low Fig. 2. Seasonal trend in the Keetch–Byram Drought Index (KBDI) in 08–09 for understory (DSFL), dry sclerophyll forest-dense understorey (DSFD), mix dry–wet Coldstream AWS compared with 40-year average and 1982/83 fire season traces. sclerophyll forest (MDWSF), wet sclerophyll forest (WSF), and eucalypt plantations Data sourced by Bureau of Meteorology (2009). (EP). Fuel loads representative of long unburned stands, i.e., wildfire or prescribed fire has not occurred in the last 25 years.

Fuel component Fuel type DSFL DSFD MDWSF WSF EP corded the development of the fire convection plume. This weather radar has a high-powered microwave radio transmitter operating Fuel layer load (kg mÀ2) Litter/duff 1.1 1.1 1.36 1.96 0.32 at a 10 cm wavelength. The reflected energy (echoes) from the ra- Near-surface 0.43 0.61 0.24 0.65 0.23 dar narrow beam was used to measure a mean reflectivity over a Elevated 0.03 0.6 0.1 0.57 0 sample volume of variable dimension. Echoes from smoke particles Bark 0.51 0.41 0.61 0.47 0.3 observed in the radar provided an estimate of plume density and Woody (d < 7.5 cm) 1.21 1.21 1.12 NA 0 Woody (d > 7.5 cm) 1.15 1.15 11.35 NA 0 height. Our analysis of plume development relied on 3D-Rapic, a Total fuel load 4.43 5.08 14.78 3.65 0.85 software for display of volumetric weather radar data (Purdam, Fuel layer depth (m) 2007). Litter 0.025 0.026 0.033 0.043 0.013 Near-surface 0.25 0.24 0.30 1.00 0.26 3. Results Elevated 1.0 3.1 1.8 3.4 0.6

3.1. Vegetation/fuels supervised classification methods, in which the classification was based on field data describing burn severity across the burn area, Dry sclerophyll eucalypt forest comprised about 47% (40% low were also tested on individual and combined-date image products. understorey, 7% dense understorey) of the area burnt within the Classification accuracy was assessed by comparing classification fire perimeter at 24:00 on 7 February. Other significant vegetation results to the burn severity field data at 646 points on a random types were grasslands/open woodland and mixed dry-wet sclero- grid generated over a high spatial resolution airborne post-fire im- phyll eucalypt forest, covering 20.5% and 16.5%, respectively of age of the burn area. Data points were labelled from 1 to 7 indicat- the area burned. The fire also spread over wet sclerophyll eucalypt ing the level of crown and understorey burn (Table 2). Map forest (4%), riparian forest (3.7%), pine plantation (2.3%) and euca- accuracy was calculated using an error matrix (Congalton and lypt plantation (1%). Understorey and bark fuel load in forest fuel Green, 1999) comparing the classified image to burn severity field complexes varied between 0.85 kg mÀ2 in eucalypt plantations data. This was conducted on an independent set of 10 validation and 14.78 kg mÀ2 in mixed dry-wet sclerophyll eucalypt forest (Ta- points per image class. These points were identified using the same ble 3). The dry sclerophyll forest total fuel load varied between method used to identify the classification training points but were 4.43 and 5.08 kg mÀ2, respectively for the low and dense understo- not used in the classification process. Error matrix analysis indi- rey types. Fine fuels, the fuels typically consumed in flaming com- cated that the most suitable map was produced by a supervised bustion, made approximately 50% of the total fuel load in these two classification of the 6-band (1, 2, 3, 4, 5 and 7) early post-fire Land- fuel types, but less then 20% in the mixed dry-wet forest. Live foli- sat TM image using the maximum likelihood algorithm. age biomass in the overstorey canopy was not directly quantified The remotely sensed burn severity classes indicate the degree of in this study. Published values indicate a range of canopy biomass physical damage to vegetation observed after a fire has passed varying between 0.4 kg mÀ2 for dry forest with moderate canopy (Hammill and Bradstock, 2006). They do not aim to describe the ef- cover and 0.9 kg mÀ2 for well-stocked stands of mixed dry-wet for- fect of fire on ecosystem components such as plant species and soil est (Attiwill, 1979; Baker and Attiwill, 1985; Ash and Helman, properties. Burn severity is a function of fire intensity and the sus- 1990). ceptibility of vegetation to fire and is used here as a surrogate for Ungrazed pastures in southern Australia typically have a fuel fire activity level (Table 2). load of 0.45 kg mÀ2 (McArthur, 1966). However, pasture condition (load and continuity) in the fire area was highly variable due to dif- 2.6. Plume development ferent grazing activities in certain areas, ranging from eaten-out to ungrazed. Curing was higher then 95% for most of these pastures The Bureau of Meteorology weather radar located at Laverton, although localised lower curing levels were observed at higher ele- approximately 65 km southwest from the fire ignition location re- vations and in moist sites along rivers and gullies (Cruz et al., 2010).

Fig. 3. Meteorological observations for the 7 February 2009 for Kilmore Gap AWS (left column) and Coldstream AWS (right column). Source: Bureau of Meteorology.

Pine plantations (P. radiata) in the fire area were well-stocked mid- and 6 and 7 February. In the first period most districts in south- dle-aged plantations (sensu Douglas, 1964). Full canopy closure oc- eastern Australia exceeded previous maximum temperature re- curred some years prior to the fire and a well-developed litter layer cords. Melbourne experienced a record three consecutive days was present. Indicative fuel loads for radiata pine plantations of with maximum temperatures exceeding 43 °C. During the second similar structure based on work by Williams (1976), Forrest and shorter period, many of these new records were broken again. Ovington (1969), Madgwick (1983) and Cruz and Plucinski (2007) The KBDI trace for Coldstream AWS (Fig. 2) shows the evolution are: litter 0.4 kg mÀ2; duff and woody fuels: 0.9 kg mÀ2; canopy: of soil and fuel dryness on the 2008/09 fire season compared to 1.0 kg mÀ2. that of 1982/1983 and the long-term average. Coming out of a dry spring, heavy precipitation in December reduced the KBDI to 3.2. Weather around 50 (scale from 0 to 200 mm), the average summer value in December. The graph shows a rapid increase in KBDI from the 3.2.1. Antecedent weather conditions end of December to early February, a result of the lack of significant The fire area is located within a temperate climate region with rainfall and above-average temperatures in January associated moderate winter rainfall. Annual mean precipitation in the vicinity with the extended heatwaves. The rapid increase in KBDI suggests of the fire was: Kilmore Gap AWS: 695 mm (1995–2011); Cold- that the heatwaves had a significant effect in stressing live vegeta- stream AWS: 745 mm (1995–2011); Melbourne Airport AWS: tion and drying coarse woody fuels, increasing the availability of 549 mm (1971–2011); station locations in Fig. 1. The 2008/09 fire these fuels for combustion (Gill and Moore, 1990). On 7 February season in Victoria was preceded by more than a decade of signifi- the KBDI was 123 in Coldstream, and the drought factor (McArthur, cant rainfall deficit and above average temperatures. In central Vic- 1967) was 10, implying all fine fuels were available for combustion toria the preceding twelve-year rainfall total was 10–20% below (Luke and McArthur, 1978). Dryness values at weather stations the 1961–1990 average. South-eastern Australia experienced 12 surrounding the fire area were slightly higher. KBDI values at Mel- consecutive warmer-than-average years prior to 2009 (Bureau of bourne and Eildon AWS were respectively 148 and 133. Meteorology, 2009). Considering the months preceding the fire, November and December saw above average monthly rainfall 3.2.2. Weather on 7 February 2009 (161 and 155 mm, respectively at Kilmore Gap and Coldstream A strong stationary high-pressure system located in the Tasman AWS) followed by a dry January and early February (3 and Sea and a system of low pressure cells located over northern 14 mm rainfall, respectively at Kilmore Gap and Coldstream AWS). Australia conditioned the weather in south-eastern Australia just Two heatwave periods developed over south-eastern Australia prior to 7 February (Bureau of Meteorology, 2009). This promoted from late January to early February: between 27 and 31 January the development of a deep pool of very hot air over central

Australia. The atmospheric circulation around the high pressure curred that the dead fuel moisture contents in forests increased system brought this hot air into Victoria. On 7 February a strong significantly. cold front moving south of Australia created very strong hot north-westerly winds ahead of the frontal passage in western 3.4. Fire propagation and central Victoria. Overnight temperatures and relative humidity at Kilmore Gap AWS remained above 25 °C and below 20%, respec- The propagation of the fire was characterised by two broad tively (Fig. 3). The 11:00 am Melbourne Airport radiosonde showed spread periods. The first period, spanning from the start of the fire a mixed layer depth of 3.5 km with stable inversion layer from the at approximately 11:45 to the arrival of a cold front around 18:00, surface to 1 km above ground level (AGL) (Bureau of Meteorology, was driven by gale force dry winds that resulted in a 50–55 km 2009). The temperature profile indicated that by the time the max- long, 27,000 ha fire footprint. The passage of the cold front and imum temperature (peaking above 45 °C) was reached the mixed associated wind change turned the fire’s eastern flank into a broad layer depth had grown to 5 km. The vertical wind profile showed headfire with widespread short and medium range spotting; a total wind speeds up to 93 km hÀ1 from 1 to 2 km AGL. of approximately 48,000 ha were burned in the two hour period As the morning progressed and surface heating eroded the low- following the wind change. What follows is a detailed description er layer, the dry upper geostrophic winds were able to mix down, of fire propagation for hourly burn periods. Weather data and bringing the hot, dry and fast winds to the surface. At 11:00, 10-m FDFMC for each burn period are provided in Table A2. open wind speeds were averaging 52 km hÀ1 (winds are 10 min averages) gusting at 72 km hÀ1 at Kilmore Gap AWS. Between 3.4.1. Burn period 1: 11:45–12:00 12:00 and 14:00 average wind speeds varied between 46 and The fire is believed to have started at approximately 11:45 as a 68 km hÀ1, with gusts up to 91 km hÀ1 (Fig. 3). This weather situa- result of arcing from a broken power line on private farmland tion prevailed for most of the afternoon until the arrival of the cold (Teague et al., 2010). Topography in the area was undulating with front (Fig. 1). The frontal transition zone is a convectively unstable fuel cover being predominantly heavily-grazed grass with patches area characterised by abrupt pressure fluctuations and associated of open woodland, particularly in gullies. A fire watchtower first gusty winds. A relatively sudden change in wind direction, from reported the fire at 11:47. The fire developed under a very strong north-westerly to south-westerly occurred with the passage of northerly wind and spread in a southerly direction with a length the cold front (Bell, 1985). The arrival of the frontal transition zone to breath ratio (L:B) of 6. Flame heights in grassland fuels were at a given location was apparent in the rapid change in measured low (<1 m), possibly the result of low fuel loads in eaten out pas- weather variables at 17:48 at Coldstream and 18:13 at Kilmore tures and strong winds. Gap (Fig. 3). The change was associated with an increase in wind strength, 3.4.2. Burn period 2: 12:00–13:00 both mean and gust, at Coldstream but with less of an increase The fire continued to spread in a southerly direction through at Kilmore Gap, where gust strength increased but mean wind grassy paddocks and scattered open woodland (Fig. 4) with a 300 strength decreased (Fig. 3). Temperature and relative humidity to 400 m flame front wide and an elongated shaped perimeter similarly show rapid changes as a result of the passage of the cold (L:B of 5). Throughout this burning period, the fire also burned into front. At Kilmore Gap, the temperature dropped from 40.1 to radiata pine plantation logging slash and denser patches of dry 28.8 °C in a period of 30 min. Relative humidity increased from woodland fuels that increased the overall energy released. Fuel 10% to 34% over the same period. At Coldstream the temperature accumulation under denser woodland areas initiated individual dropped from 44.6 to 32.8 °C. Relative humidity increased from tree torching. The rate of spread of the main flame front was ob- 8% to 30% (see Bureau of Meteorology (2009) for a detailed descrip- served to decrease in denser woodland areas and on the lee side tion of the cold front arrival). Behind the cold front winds became of hills, but increased spotting allowed the fire to burn over these light and variable and there was a progressive decrease in temper- areas without slowing its overall propagation. Spotting up to ature (approximately 25 °C at Kilmore Gap and Coldstream at 700 m ahead of flame front was observed. The average forward rate 22:00) and increase in relative humidity (approximately 50% at Kil- of fire spread for this period was 71 m minÀ1. more Gap and Coldstream at 22:00) resulting in a marked decrease in fire spread potential. 3.4.3. Burn period 3: 13:00–14:00 During this burning period the 10-m open wind speed averaged 52 km hÀ1, gusting to 85 km hÀ1 at the nearby Kilmore Gap AWS 3.3. Fuel moisture content and estimated FDFMC varied between 2% and 6% (Table A2). The fire propagated through a combination of pine plantation logging The morning of the 7 February was preceded by low overnight slash, scattered eucalypt plantations and a mixture of grazed and relative humidity over the fire area (Fig. 3), limiting the recovery of ungrazed grassland. The spatial discontinuity of fuels made FDFMC. Predicted maximum FDFMC for forest litter early on 7 Feb- short-range spotting in light fuels the dominant fire propagation ruary was between 12% and 15%. Maximum overnight FDFMC in process, resulting in multiple isolated fires spreading forward but grasslands was 12%. Increasing temperature and decreasing rela- failing to merge into a broad cohesive flame front. At this time sev- tive humidity during the morning caused forest litter moisture eral well-separated zones of fire activity (some 500–700 m apart) content to drop below 6% by midday. During the peak burning per- formed the leading edge of the fire and the southern flank began iod, between 15:00 and 18:00, the lowest FDFMC values were to impact home and other building structures along the northern reached with all of the landscape at less than 5% (Table A2 in edge of Wandong township (Fig. 4). Fire behaviour ranged from Appendix; Sullivan and Matthews, submitted for publication). flames less than 1 m tall in grazed paddocks to active crown fire The FDFMC in the wet tall forests at higher elevations of the Hume propagation in the eucalypt and pine plantations with flame range were higher than the lower elevation fuels. heights 10–20 m above the canopy. The average forward rate of fire The arrival of the change had an almost immediate effect on the spread for this burning period was 73 m minÀ1. FDFMC in grasslands, with a much slower uptake of moisture in forested lands (Sullivan and Matthews, submitted for publication). 3.4.4. Burn period 4: 14:00–15:00 It was not until well after midnight when a brief rain shower After 14:00, the fire’s leading edge moved out of the farmland/ (0.8 mm at Kilmore Gap AWS; 0.2 mm at Coldstream AWS) oc- scattered woodland and peri-urban areas into more complex

Fig. 4. Reconstruction of fire spread with evolution of perimeter between 12:00 and 14:00. Red circles are spotfires identified between 12:00 and 13:00. Red triangles are spotfires identified between 13:00 and 14:00. Fire perimeter reliability rating is: 12:00 – 3; 13:00 – 1; 14:00 – 3. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.) topography with steep slopes and dense forest cover associated ering the transport distance and the wind speeds aloft (75– with the foothills of the Hume Range (Fig. 5). The main fuel type 90 km hÀ1 within the 1–4 km AGL layer), we estimate that these in its path was dry eucalypt forest with low understorey and scat- fuel particles were lifted/injected into the convection plume tered areas of mixed dry-wet eucalypt forest. As the fire burned around 14:00 when the main fire front entered heavier fuels in into long unburned forests dominated by species with loose fibrous eucalypt forest approximately 41 km from where they were land- bark (e.g., Eucalypt obliqua and Eucalypt macrorhyncha) its dynamics ing. Meteorological radar images indicate strong pulses in plume changed, with concentrated short distance spotting becoming the development between 14:20 and 15:00, after which there was a main fire propagation mechanism. Profuse spotfires lighting up substantial increase in plume density. At the end of this burning to 500 m ahead of the solid flame front allowed the fire to travel period the L:B was 5. at speeds substantially higher than expected in the absence of spotting. 3.4.5. Burn period 5: 15:00–16:00 Medium and long distance spotting became an important Between 15:00 and 16:00 fire activity escalated. As the fire mechanism of fire propagation once the fire encountered forests spread along the south-western faces of Mt. Disappointment, it containing eucalypt species with ribbon bark (e.g., Eucalyptus vim- burned into heavier fuels characteristic of mixed dry-wet sclero- inalis, Eucalyptus globulus spp. Bicostata). Isolated medium range phyll forest resulting in a sustained increase in the energy released spotfires occurred up to 2 km ahead of the main fire front. Between by the fire. 14:15 and 14:45 partially charred fuel particles (extinct firebrands) Profuse short-range spotting continued to dominate the main were falling (but failing to ignite spotfires) at several locations 35– fire propagation. After 15:15 isolated long-range spotfires were 40 km SSE from where the main fire front was at the time. Consid- occurring up to 30–35 km ahead of the main fire (Fig. 5). The

Fig. 5. Reconstruction of fire spread with evolution of perimeter between 14:00 and 16:00. Red circles are spotfires identified between 14:00 and 15:00. Red triangles are spotfires identified between 15:00 and 16:00. Fire perimeter reliability rating is: 15:00 – 3; 16:00 – 2. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

farthest confirmed ignitions within this period occurred in open 3.4.7. Burn period 7: 17:00–18:00 paddocks approximately 40 km from its probable source. These Between 17:00 and the arrival of the wind change at around isolated spotfires were quickly extinguished by fire crews. Due to 18:00 fuel discontinuity, including a mosaic of grazed paddocks long-range spotting, the length of fire activity along the fire’s major intermixed with agricultural crops, and active fire suppression, axis went from approximately 14–54 km in one hour. constrained the south-easterly propagation of the fire. Concur- Along this fire ‘footprint’ there were a significant number of rently, high intensity, erratic fire behaviour occurred as the fire medium-range spotfire ignitions. Spotfires developing in forest filled in large unburned woodland and forest covered areas within fuels built-up quickly with flames climbing up the bark and ignit- the broad fire perimeter. Along the eastern flank, upslope passive ing the crowns. These fires then grew independently from the main crowning in eucalypt forests and a pattern of short-range spot fires zone of fire activity. Aided by short- and medium-range spotting was responsible for flank propagation. As the fire progressed into the main flame front propagated with an average forward spread the wet sclerophyll forest around Mt. Disappointment it spread rate of 153 m minÀ1 during this burning period and the fire elon- mostly as a high intensity surface fire with isolated torching trees. gated its shape with an L:B of 6. This decrease in fire intensity resulted from the lower flammability of this forest type compared to the dry or mixed sclerophyll euca- 3.4.6. Burn period 6: 16:00–17:00 lypt forest. Characteristics such as a high stand height, high canopy At around 16:00 a number of subsidiary fires started by long dis- base height and the smooth bark of Eucalyptus regnans restricted tance spotting were spreading in forest fuels largely independent of fire propagation to the understorey layer. the main zone of fire activity. These fires, spreading in broken topog- Between 17:00 and 18:00 the cold front approached the fire raphy had fully developed fire fronts with widths up to 750 m and area (Fig. 1) at a rate of travel of approximately 58 km hÀ1. The were spreading over an area extending up to 25 km. The distribution wind change was felt at Kangaroo Ground fire tower (Fig. 1)at of medium- and long-range spotting between 14:00 and 17:00 re- 17:43. Reports from the fire ground indicate that the change in this sulted in an extended fire perimeter (Fig. 6) with an L:B of 8 and cov- area was rapid, accompanied by strong gusts from the southwest. ering an area of approximately 17,500 ha. An unknown proportion of At Coldstream AWS the wind direction moved 70°, from 320° to this area was still unburnt at 17:00. 250° in a 14-min period.

Fig. 6. Reconstruction of fire spread with evolution of perimeter between 16:00 and 18:00. Red circles are spotfires identified between 16:00 and 17:00. Red triangles are spotfires identified between 17:00 and 18:00. Fire perimeter reliability rating is: 17:00 – 2; 18:00 – 2. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

3.4.8. Burn period 8: 18:00–19:00 namely due to the existence of large blocks of young fuels associ- The immediate impact of the wind change on the fire was to ated with recent fuel reduction burns, a 2006 wildfire, peri-urban convert the eastern flank into a broad, fast moving front and areas and incompletely cured grassland areas. to concurrently throw a large number of firebrands in a north-easterly direction. The timing of this event along the eastern flank 3.4.10. Burn period 10: 20:00–24:00 varied as the wind change extended across the fire (Fig. 6). After Limited fire propagation occurred between 20:00 and 24:00. the passage of the wind change the fire developed a 55 km long Shortly after 20:00 sections of fire spreading in a north-easterly headfire. The southern half of this front burned into a mosaic of direction reached open grazed paddocks. A combination of low fuel grazed paddocks, low curing levels (40–80%) and segregated agri- continuity and rapid increase in the moisture content of grass fuels cultural blocks. As the post-change fire front propagated into this (from 10–12% to 18–19%, see Table A2) and drop in overall wind landscape it fragmented into smaller fires, opportunistically burn- speed restricted fire propagation. Localised rain showers along ing with high intensity in corridors of mixed dry-wet eucalypt for- the northern perimeter extinguished fire in grassland fuels and est fuels. The northern section of the fire propagated as a high damped fire propagation in the forests. intensity crown fire driven by profuse short range spotting through dry and mixed dry-wet eucalypt forest. This front propagated at an 3.5. Fire severity average forward rate of spread of 127 m minÀ1. At 19:00 the area affected by fire was approximately 63,000 ha, more than double The burn severity map for the fire propagation between ignition the 27,180 ha burned until 18:00. and 24:00 h on the 7 February is shown in Fig. 8. Burn severity by area was distributed as: Class 1 (full crown fire): 31%; Class 2 3.4.9. Burn period 9: 19:00–20:00 (intermittent crown fire): 14%; Class 3 (high intensity surface fire The burn dynamics between 19:00 and 20:00 were similar to in forest): 22%; Class 4 (moderate intensity surface fire in forest): those of the previous burn period. The northern section of the fire 5%; Class 5 (low intensity surface fire in forest): 9%; and Class 6 continued to burn as a crown fire aided by short range spotting on (burnt grassland): 16%. Six percent of the area within the fire a 20–25 km extended front in dry eucalypt forest. The overall aver- perimeter at 24:00 was unburnt (Class 7). Fig. 9 shows the distri- age rate of fire spread for this period was 90 m minÀ1. The propa- bution of burn severity by fuel type. The dry-sclerophyll – low gation of the southern section of the post change flame front understorey fuel type was the predominant fuel, covering 40% of continued to be constrained by large scale fuel discontinuity, the area burnt on the 7 February. 63% of the area occupied by this

Fig. 7. Reconstruction of fire spread with evolution of perimeter between 18:00 and 24:00. Red circles are spotfires identified between 18:00 and 19:00. Red triangles are spotfires identified between 19:00 and 20:00. Fire perimeter reliability rating is: 19:00 – 3; 20:00 – 4. 24:00 – 3. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)

fuel type burned as crown fire. Comparable crown fire activity was culations are associated with the main fuel type in each burning observed in the dry sclerophyll-dense understorey and mixed dry- period. Average fireline intensity was greatest (88,000 kW mÀ1) wet eucalypt forest (Fig. 9). The fuel types with the highest level of between 15:00 and 16:00 as the main flame front approached canopy consumption were the eucalypt and pine plantations, the south-western slopes of Mt. Disappointment. Comparable respectively with 67% and 76% of their area burned as a crown fire. intensities were estimated for the two burning periods following The riparian and wet sclerophyll forests burned predominantly as a the passage of the cold front, 73,000 and 83,000 kW mÀ1, respec- high to moderate intensity surface fire. tively for the periods 19:00–20:00 and 20:00–21:00. Fireline intensity was not calculated for the period between 3.6. Fuel consumption and fire intensity 16:00 and 18:00 (burning period 6 and 7). During this period the overall fire propagation was determined by multiple medium to Post-fire sampling in dry sclerophyll forest and mixed dry-wet long-range spotting ignitions that resulted in a number of dis- eucalypt forest indicated that litter, shrub, bark and downed woo- jointed fires spreading and eventually coalescing within a broad dy fuels with diameters up to 7.5 cm were fully consumed. In these area. The concept of fireline intensity, describing the rate of energy forest types consumption of woody fuels with diameters larger release rate of a moving flame front, is not applicable to this situ- than 7.5 cm varied between 80% and 100% (Hollis et al., 2011b). ation (McArthur and Cheney, 1966). An alternative metric to de- Estimated total fuel consumption in the dry sclerophyll forest scribe the intensity of the fire under these conditions is to varied between 4.5 and 6.1 kg mÀ2. Fuel consumption in the mixed quantify the rate of energy release per burning period. Energy re- dry-wet eucalypt forest was estimated to be 13.4 kg mÀ2. Visual leased per burn period integrates: (i) the average amount of fuel inspection of eucalypt and pine plantations indicated full consumed per fuel type, (ii) the area burnt, and (iii) the severity consumption of understorey fuels suggesting overall fuel con- of burn. Before the wind change an extreme value of 933 GW sumption in the headfire region of 1.7 and 2.3 kg mÀ2, respectively. was calculated for burning period 5 (Table 4). The increase in fire Table 4 provides estimates of hourly average fireline intensity area during burning period 6 and 7 resulted in a rate of energy re- (as per Byram, 1959; assuming heat content of 18600 kJ kgÀ1; lease of 2153 and 1794 GW, respectively. Post-wind change Burrows, 1994) and rate of energy released for the 10 burning propagation led to a peak in the energy released (8555 GW) during periods described in the preceding section. Fireline intensity cal- burning period 8, followed by a decrease thereafter.

Fig. 8. Burn severity map for the area burnt by the Kilmore East ﬁre on the 7 February 2009.

3.7. Plume development plume along a SW–NE axis at 19:30. Fig. 10c is a cross-section of the plume along a NW–SE axis, i.e., perpendicular to the direction The smoke echoes observed in the radar data provide indications of fire spread after the passage of the flame front. This image shows of the top height of the convection column. The plume was noticed in a broad flame front extending for approximately 60 km. From the radar scan as early as 12:00. Between 13:00 and 14:30 the plume 18:30 the pyro-cumulonimbus cloud was reported to have had a top height varying between 4 and 5 km and an elongated shape generated lightning strikes at several locations within and in the extending over 50 km, the hallmark of a wind-driven fire. Fig. 10a vicinity of the fire area (Bureau of Meteorology, 2009). provides the profile of the plume through a NNW–SSE axis at 14:00. Radar scans were interrupted between 14:36 and 18:06 due to radar malfunction caused by the extreme heat (46.4 °C at its loca- 4. Discussion tion). Between 18:30 and 19:30 the plume top peaked at 13 km. These plume heights are considered conservative as they were esti- 4.1. Accuracy of reconstruction mated from the midpoint of the control volume (as illustrated in Fig. 10b). Given the uncertainty associated with the large control Reconstructing the propagation of a high intensity forest fire volume at this radar elevation angle and that the radar is tuned to necessarily involves a high degree of uncertainty. The myriad of sense particles larger than cloud water drops, the plume heights information collected from diverse sources had a broad range of during this period could have extended to 15 km. reliability. In order to combine the information from these dispa- The strength of this pyro-cumulonimbus cloud (i.e., fire induced rate sources, a reliability index (Table A1) from 1 (most reliable) convective cloud with considerable vertical development) is obser- to 5 (least reliable) was devised and used to assess the relative vable in Fig. 10b and c. Fig 10b provides a cross-section of the quality of each datum. The reliability of the fire perimeters in Figs.

1969). The presence of fibrous bark provides further vertical con- nectivity between fuel layers facilitating the transition from a surface fire to one involving the full fuel complex. The process of short-range spotting was exacerbated in the weather conditions driving the fire. The strong winds affecting the fires forced the transport of a profusion of burning embers through flat (rather than lofting) trajectories, delivering numerous firebrands up to 500 m ahead of the main fire front. The deep flaming fronts that arise from the coalescence of multiple short-range spot fires resulted in extensive crowning and further generation and transport of burning embers. McArthur (1967) describes this process as key to how a fire maintains overall rates of spread much higher than expected in the absence of spotting. Key components for the maintenance of this process are the presence of high surface fuel loads (to induce ignition of bark fuels), long unburnt (>25 years since last fire in dry sclerophyll forest) eucalypt forest with a significant number of species with fibrous bark (the primary firebrand material), high wind speeds (causing flat ember trajectories) and low fuel moisture contents (increasing the likelihood of spotfire ignition). With fuel moisture contents <4%, the likelihood of spotfire ignitions increase significantly as the heat requirements for ignition are reduced. In this situation even tiny glowing particles had sufficient energy to start new spotfires (Albini, 1979; Ellis, 2011). The observed short range spotting behaviour is consistent with McArthur’s (1967) description of ‘‘showers of burning embers land- ing up to 800 m of the main fire front’’. A quantitative understanding of short range spotting dynamics, namely firebrand density distri- Fig. 9. Distribution of burn severity classes by selected fuel types. bution with distance from the fire front and how distinct fires coalesce in a high turbulent environment, is lacking. Field-based research into short range spotting such as that carried out in Project Vesta (Gould et al., 2007a; Box 1), provide benchmark data that can 4–7 is given in each figure caption. Fuel moisture estimates were be used to parameterize firebrand transport and density models considered to possess a rating of 3. Fuel characteristics data used (e.g., Cheney and Bary, 1969; Sardoy et al., 2007). The validity of was given a rating of 2. scaling up Gould et al. (2007a) observations to conditions similar The number of medium and long distance spot fires given in to those driving the Kilmore East fire is nonetheless unknown. Figs. 4–7 should be seen as conservative. These were spotfires that were authenticated by one or more sources. A number of spot fires were not located and it is believed that there were a larger number 4.3. Spotting – long range of spot fires that were not detected. Spotfires were located with a precision of 100–500 m. For the scale of the fire, this precision cor- The occurrence of long-range spot fires (>5 km) contributed to responds to an error less than 10% for long distance spotting. the rapid extension of the fire zone. At 16:00 there were at least five spotfires occurring up to 33 km south-east of the main fire, approximately 40 km from its probable source. Comparable spot- 4.2. Spotting – short range ting distances have been verified on particular occasions in fires in southern Australia eucalypt forests (Hodgson, 1967; Cheney Eucalypt stringybark species (e.g., E. obliqua, E. marginata and E. and Bary, 1969; McArthur, 1969). The process of long-range macrorhyncha) are notable for their fibrous rough bark that can be spotting can be seen as distinct from short-range spotting, easily ignited and decorticate from the trunk providing an opti- requiring the presence of a specific different set of conditions. mum firebrand source for short-range spotting (Cheney and Bary, The firebrands responsible for long-range spotting are long

Table 4 Average rate of fire spread, area burned, fireline intensity (Byram, 1959) and heat release for selected burning periods (BP) in the Kilmore East fire.

BP Main fuel Fire perimeter Average rate of ﬁre spread Hourly area burned Average ﬁreline intensity Energy released type (km) (m minÀ1) (ha) (kW mÀ1) (GW) 1 Grass 1.3 – – 1.1 2 Grass 10 71 239 6603 25.6 3 Grass, DSFL 21 73 1534 – 152.3 4 DSFL 30 68 1311 39,209 227.7 5 DSFL 50 153 4286 88,220 933.0 6 DSFL, MDWSF 106 – 10,068 – 2153.7 7 DSFL, MDWSF 143 – 9752 – 1794.4 8 DSFL 217 127 35,802 73,228 8554.8 9 MDWSF 311 90 12,626 82,677 2605.4 10 DSFL 487 – 6129 – 860.9 nd, not determined; DSFL, dry sclerophyll forest, low understory, MDWSF, mix dry-wet sclerophyll forest.

contributed with higher available fuel loads that resulted in a peak in fireline intensity (85,000–90,000 kW mÀ1) and strengthened the plume. This escalation in fire activity between 15:00 and 16:00 extended into the subsequent burning periods with dramatic effects. From 16:00 to 17:00 the fire increased in area from 7400 to 17,400 ha, corresponding to an elongation of the main axis of fire propagation from 23 to 47 km. Nonetheless, this was not due to the spread of a single flame front but the result of multiple new ignitions due to medium and long range spotting. At 17:00 substantial areas remained unburnt behind the leading edge of the fire but in a broad sense, considering fire spread at the landscape scale, one could say that the fire spread 24 km in an hour. The passage of the wind change was followed by very high rates of fire spread in the northeast direction, 127 m minÀ1 between 18:00 and 19:00 and 90 m minÀ1 between 19:00 and 20:00. Such fast spread rates were maintained by profuse short-range spotting dynamics in dry and mixed dry-wet sclerophyll forest. Fireline intensities associated with this post wind change period varied between 70,000 and 85,000 kW mÀ1. This extreme fire behaviour cannot be explained by the prevailing weather conditions at the Kilmore Gap and Coldstream AWS where the Forest Fire Danger In- dex (FFDI; McArthur, 1967) dropped to values below 20 (Table A2). Although the post-change cool and moist air mass caused an overall drop in fire potential at the various AWSs outside the fire area, it is unknown how much this air mass was able to influence the fire environment, particularly the moisture content of fine forest fuels, downwind of the fire perimeter. As the southwest winds pushed Fig. 10. Melbourne weather radar view of Kilmore East fire plume. (a) Profile of the over the broad fire area (Fig. 10c), advection of heated products plume through a NNW–SSE axis at 14:00. (b) Cross-section of the plume along a SW–NE axis at 19:30. (c) Cross-section of the plume along a SE–NW axis at 19:30. from residual combustion likely resulted in hotter and drier air in the fire path than recorded at the AWS peripheral locations. This advected heat mechanism allowed the maintenance of vigorous fire behaviour in the immediate hours following the wind change. streamers of decorticating bark that normally hang from the The observed rates of fire spread and associated fireline intensi- upper branches in certain smooth-barked eucalypt species such ties are at the top end of the known fire behaviour spectrum in dry as E. viminalis, E. globulus, E. delegatensis (Cheney and Bary, sclerophyll eucalypt forests (Cheney, 1991; Gill and Moore, 1990). 1969). The bark strips curl into hollow tubes that when ignited To the authors’ knowledge, there are no other published detailed at one end can burn for as long as 40 min (Hodgson, 1967). case studies in eucalypt forests showing comparable rates of The long combustion times coupled with its good aerodynamic spread and fireline intensities. This is not to say that other fires properties (Luke and McArthur, 1978; Ellis, 2011) allows these in the past did not exhibit similar or more severe fire behaviour. firebrands to be a viable ignition source even when transported The growth of past fires such as the 1952 Mangoplah fire over long distances. Long-range spotting also requires an intense (300,000 ha in 8 h) and the 1965 Chatsbury fire (260,000 ha in a fire that maintains a strong upward motion in the buoyant 72 km long run in 8–10 h; McArthur, 1969) are clear evidence that plume to transport relatively large fuel particles several kilome- commensurate fire behaviour have occurred in the past. tres above the ground and high winds aloft to transport firebrands for extended distances downwind. 5. Conclusions

On the 7 February 2009, a date later named ‘Black Saturday’, 4.4. Rate of fire spread and intensity history repeated itself. Under a synoptic situation typical of extreme fire weather potential in south-eastern Australia (Sullivan The fire exhibited three distinct spread phases between 12:00 et al., 2012) the Kilmore East fire exhibited rates of fire spread and the arrival of the wind change around 18:00, even though and intensity at the top end of the fire behaviour spectrum. An area average fire weather conditions stayed relatively constant (Table of approximately 100,000 ha burned in less then 12 h and a total of A2). These changes in spread dynamics were driven by fuels, 121 people were killed during this period. Spotting dynamics topography and fire-atmosphere interactions. Between 12:00 and dominated fire propagation. Prolific short range spotting linked 15:00 the fire spread in a mosaic of fuel types, ranging from grazed with crown fire propagation in eucalypt forest allowed for exceed- paddocks intermixed with open woodland to dry sclerophyll euca- ingly fast rates of spread, typically higher then 70 m minÀ1 and lypt forest, and attained average rates of fire spread varying be- peaking at 150 m minÀ1. The extreme fireline intensities, varying tween 68 and 73 m minÀ1 (Table 4) and fireline intensities between 70,000 and 88,000 kW mÀ1 throughout most of the after- approaching 40,000 kW mÀ1. In the next burning period, 15:00– noon, combined with the strong winds aloft allowed for the devel- 16:00, the rate of fire spread doubled to an average of opment of a strong convection plume and the transport of 153 m minÀ1. This increase could be attributed partially to a grad- firebrands at long distances. Spot fires occurred 33 km ahead of ual decrease in fine dead fuel moisture content (Table A2) but also the flame front. These spotting distances corresponded to firebrand to the fact that the fire transitioned into more complex topography transport distances of approximately 41 km. with a mix of dry and wet sclerophyll forests. Upslope runs in this The high amounts of energy released by this fire resulted in a area might also have contributed to an increase in the overall rate pyrocumulonimbus cloud with a top height of at least 13 km. This of fire spread. The denser and more productive forests of this area resulted in the injection of smoke and other combustion products

Table A1 Reliability rating for weather, fuel and ﬁre spread observations for wildﬁre case studies. Adapted from Cheney et al. (1998).

Rating Weather Fine fuel moisture content Fuel complex Rate of spread 1 Nearby (<25 km) meteorological station or direct Point measurements made at Fuel characteristics inferred Direct timing of fire spread measurements in the field with high quality time of fire and extrapolated to from a fuel age function measurements i.e. IR scans, aerial instruments, and/or validated modelled wind field fire area taking into account developed for the particular observations, observed reference topographic effects fuel type/area points with photographs 2 Meteorological station within 50 km of the fire Single point measurements Fuel characteristics inferred Reliable timing (within ± 15 min) of with no local effects (i.e. terrain, vegetation) on the within or in the vicinity of fire from a visual assessment or fire spread by field observations wind field, and/or partially validated modelled area measurements of nearby with general reference points wind field unburnt forest 3 Meteorological station within 50 km of the fires Estimation of fuel moisture Fuel characteristics inferred Reconstruction of fire spread with but there are local effects on the wind field or the through validated models taking from a fuel age curve for a numerous cross references data not representative of the fire area. into account topographic effects forest type of similar Meteorological station >50 km of the fire, structure reconstruction of wind speed for fire site. Unvalidated modelled wind field 4 Spot meteorological observation near the fire Estimated fuel moisture for Fuel characteristics typical of Doubtful reconstruction of fire nearly (<25 km) meteorological equilibrium level in the spread stations representative fuel type 5 Distant meteorological observations at locations Estimated fuel moisture for Qualitative fuel type Anecdotal or conflicting reports of very different to fire site distant (>50 km) meteorological description fire spread stations

Table A2 Hourly weather data, estimated fuel moisture content and Forest Fire Danger Index (FFDI) at Kilmore Gap and Coldstream AWS; and observed rate of ﬁre spread and maximum spotting distances for the Kilmore East ﬁre.

Time Location Air temp. (°C) Relative humidity (%) Wind speed Wind direction FFDI Fuel moisture content (%) Mean/gust (km hÀ1) Forest/grass 12:00 Kilmore Gap 36.6 15 56/72 N 94 5–7/3–5 Coldstream 41.9 13 39/63 N 81 13:00 Kilmore Gap 39.6 11 69/82 N 162 4–6/2–4 Coldstream 42.9 10 46/67 NNE 110 14:00 Kilmore Gap 42 10 46/69 NNW 106 4–5/1–3 Coldstream 43.8 9 37/56 N 95 15:00 Kilmore Gap 41.4 10 63/91 NNW 155 3–4/1–3 Coldstream 44.3 9 41/56 N 106 16:00 Kilmore Gap 41.6 9 54/80 NW 131 3–4/1–3 Coldstream 43.2 8 30/59 N 82 17:00 Kilmore Gap 40.6 10 44/69 NW 97 4–5/2–3 Coldstream 43 9 28/37 N 75 18:00 Kilmore Gap 40.1 10 37/61 NW 81 5–6/2–3 Coldstream 32.8 30 48/69 SW 41 19:00 Kilmore Gap 27.9 38 26/43 SSW 16 7–8/8–9 Coldstream 30 35 20/37 ESE 16 20:00 Kilmore Gap 27.7 36 32/43 SSW 19 8–9/8–9 Coldstream 30.4 34 15/22 SE 15 21:00 Kilmore Gap 25.8 44 24/32 S 11 10–11/10–11 Coldstream 28.2 41 7/11 SSW 9 22:00 Kilmore Gap 24 52 20/26 S 7 11–12/11–12 Coldstream 26.3 47 9/13 S 7 23:00 Kilmore Gap 24.5 48 22/28 SSE 9 12–13/11–12 Coldstream 26.8 46 9/15 NW 7 24.00 Kilmore Gap 17.3 92 22/37 S 2 14–16/18–19 Coldstream 22.8 67 7/9 NW 3

References Cheney, N.P., Sullivan, A.L., 2008. Grass Fires: Fuel, Weather and Fire Behaviour, second ed. CSIRO Publishing, Collingwood, Australia, pp. 160. Cheney, N.P., Gould, J.S., Catchpole, W.R., 1998. Prediction of fire spread in Adams, M., Attiwill, P.M., 2011. Burning Issues, Sustainability and Management of grasslands. International Journal of Wildland Fire 8 (1), 1–15. Australia’s Southern Forests. CSIRO Publishing, Collingwood, pp. 144. Congalton, R.G., Green, K., 1999. Assessing the Accuracy of Remotely Sensed Data: Albini, F.A., 1979. Spot fire distance from burning trees – a predictive model. USDA Principles and Practices. Lewis Publishers, London. For. Serv. Gen. Tec. Rep. INT-56, Intermountain For. and Range Exp. Stn., Ogden Cruz, M.G., Plucinski, M.P., 2007. Billo road fire – analysis if fire phenomena and Utah, pp. 31. suppression activities. Bushfire CRC Report No.A.07.02. Bushfire Cooperative Alexander, M.E., Thomas, D.A., 2003. Wildland fire behavior case studies and Research Centre, Melbourne, Victoria, Australia, pp. 95. analyses: value, approaches and practical uses. Fire Management Today 63 (3), Cruz, M.G., Sullivan, A.L., Hurley, R., Sims, N., Gould, J.S., Hollis, J.J., 2010. The 4–8. Kilmore East: progression and behaviour. Report No. CSE/BDA-001, CSIRO Ash, J., Helman, C., 1990. Floristics and vegetation biomass of a forest catchment, Sustainable Ecosystems, Crace, ACT. Kioloa, south coastal New South Wales. Cunninghamia 2, 167–182. Douglas, D.R., 1964. Some characteristics of major fires in coniferous plantations. Ashton, D.H., 2000. The Environment and Plant Ecology of the Hume Range, Central Australian Forestry 28, 119–124. Victoria. Proceedings of the Royal Society of Victoria 112, 185–278. Ellis, P.F., 2011. Fuelbed ignition potential and bark morphology explain the Ashton, D.H., Attiwill, P.M., 1984. Tall open-forests. In: Groves, R.H. (Ed.), Australian notoriety of the eucalypt messmate ‘stringybark’ for intense spotting. Vegetation. Cambridge University Press, New York, pp. 157–196. International Journal of Wildland Fire 20 (7), 897–907. Attiwill, P.M., 1979. Nutrient cycling in a Eucalyptus oblique (L’Herit.) forest. III. Forrest, W.G., Ovington, J.D., 1969. Organic matter changes in an age series of Pinus Growth, biomass, and net primary production. Australian Journal of Botany 27, radiata plantations. Journal of Applied Ecology 7, 177–186. 429–458. Fox, J., Runnalls, R., 2009. Operational debrief report 2008/09 fire season, Victorian Baker, T.G., Attiwill, P.M., 1985. Above ground nutrient distributions and cycling in Government Report, Melbourne, Vic. August 2009. Pinus radiata D Don and Eucalyptus obliqua L’Herit. forests in southeastern Fromm, M., Tupper, A., Rosenfeld, D., Servranckx, R., McRae, R., 2006. Violent pyro- Australia. Forest Ecology and Management 13, 41–52. convective storm devastates Australia’s capital and pollutes the stratosphere. Bell, A., 1985. When a cold front sweeps across. Ecos 42, 22–26. Geophysical Research Letters 33, L05815. Bradstock, R., Price O., 2010. Fire severity patterns in the Victorian Bushfires of 7 Fromm, M., Lindsey, D.T., Servranckx, R., Yue, G., Trickl, T., Sica, R., Doucet, P., Godin- February 2009: influences of weather, terrain and land use history. A report Beekmann, S., 2010. The untold story of pyrocumulus. Bulletin of the American prepared for the 2009 Victorian Bushfires Royal Commission, pp. 21. Meteorological Society 91, 1193–1209. Bureau of Meteorology, 2009. Meteorological aspects of the 7 February 2009 Gill, A.M., 1981a. Patterns and processes in open-forests of Eucalyptus in southern Victorian fires, an overview. Bureau of Meteorology report for the 2009 Australia. In: Groves, R.H. (Ed.), Australian Vegetation. Cambridge University Victorian Bushfires Royal Commission, Bureau of Meteorology. Press, New York, pp. 197–226. Burrows, N.D., 1994. Experimental development of a fire management model for Gill, A.M., 1981b. Adaptive responses of Australian vascular plant species to fires. In: jarrah (Eucalyptus marginata Donn ex Sm.) forest. Aust. Natl. Univ., Canberra, Gill, A.M., Groves, R.H., Noble, I.R. (Eds.), Fire and the Australian Biota. Australian Capital Territory. Ph.D. thesis, pp. 293. Australian Academy of Science, Canberra, p. 582. Burrows, N.D., Ward B., Robinson A., 2009. Fuel dynamics and fire spread in spinifex Gill, A.M., 1997. Eucalypts and fires: interdependent or independent. In: Williams, grasslands of the western desert. In: Tran, C. (Ed.), The Proceedings of the Royal J.E., Woinarski, J.C.Z. (Eds.), Eucalypt ecology: individuals to ecosystems. Society of Queensland – Bushfire 2006 conference, Brisbane 6–9 June 2006, pp. Cambridge University Press, Cambridge. 69–76. Gill, A.M., Moore, P.H.R., 1990. Fire intensities in eucalyptus forests of Southeastern Busby, J.R., Brown, M.J., 1984. Southern rainforests. In: Groves, R.H. (Ed.), Australian Australia. In: Proceedings of International Conference on Forest Fire Research, Vegetation. Cambridge University Press, New York, pp. 131–1155. Coimbra, Portugal, 19–22 November 1990, pp. B24:1–12. Byram, G.M., 1959. Combustion of forest fuels. In: Davis, K.P. (Ed.), Forest Gould, J.S., McCaw, W.L., Cheney, N.P., Ellis, P.F., Knight, I.K., Sullivan, A.L., 2007a. Fire: Control and Use. McGraw-Hill Book Company, New York, pp. 61–89, Project vesta: fire in dry eucalypt forest: fuel structure, fuel dynamics and fire 554–555. behaviour. Ensis-CSIRO, Canberra ACT and Department of Environment and Cheney, N.P., 1991. Predicting forest fire behaviour – the Australian experience. In: Conservation, Perth WA. Proceedings of the Workshop Integrating research on hazards in fire-prone Gould, J.S., McCaw, W.L., Cheney, N.P., Ellis, P.F., Matthews, S., 2007b. Field guide- Fuel environments. The U.S. Man and the Biosphere Program. assessment and fire behaviour prediction in dry eucalypt forest. (Ensis-CSIRO, Cheney, N.P., Bary, G.A.V., 1969. The propagation of mass conflagrations in a Canberra ACT and Department of Environment and Conservation, Perth WA). standing eucalypt forest by the spotting process. Paper A6 in Collected Papers, Gould, J.S., McCaw, W.L., Cheney, N.P., 2011. Quantifying fine fuel dynamics and Mass fire Symposium, vol. 1. Common W. Aust., Defence Standards Lab, structure in dry eucalypt forest (Eucalyptus marginata) in Western Australia for Maribyrnong, Victoria, pp. 18. fire management. Forest Ecology and Management 262, 531–546.

Hammill, K.A., Bradstock, R.A., 2006. Remote sensing of fire severity in the Blue McArthur, A.G., Cheney, N.P., 1966. The characterization of fire in relation to Mountains: influence of vegetation type and inferring fire intensity. ecological studies. Australian Forestry 2, 36–45. International Journal of Wildland Fire 15 (2), 213–226. McArthur, A.G., Cheney, N.P., Barber, J., 1982. The fires of 12 February 1977 in the Hodgson, A., 1967. Fire management in Eucalypt forest. In: Proceedings of 6th western district of Victoria. CSIRO Division of Forest Research and Country Fire Annual Tall Timbers Fire Ecology Conference, March 6-7, 1967, Tallahassee, Authority of Victoria, Melbourne, Vic, pp. 73. Florida. Charied by E.V. Komarek, Tall Timbers Research Station, Tallahassee, McCaw, W.L., 2010. Report on the effect of fuel reduction burning on the behaviour Florida. and severity of the Victorian bushfires of 7 February 2009. A report prepared for Hollis, J.J., Anderson, W.R., McCaw, W.L., Cruz, M.G., Burrows, N.D., Ward, B., the 2009 Victorian bushfires Royal Commission, pp. 22. Tolhurst, K.B., Gould, J.S., 2011a. Using fireline intensity to predict woody fuel Noble, I.R., Bary, G.A.V., Gill, A.M., 1980. McArthur’s fire-danger meters expressed as consumption in Australian southern eucalypt forest fires. Australian Forestry equations. Australian Journal of Ecology 5, 201–203. 74, 81–96. Parkes, D., Newell, G., Cheal, D., 2003. Assessing the quality of native vegetation: Hollis, J.J., Matthews, S., Anderson, W.R., Cruz, M.G., Burrows, N.D., 2011b. Behind The ‘habitat hectares’ approach. Ecological Management and Restoration 4 (S1), the flaming zone: predicting woody fuel consumption in Australian southern S29–S38. eucalypt forest fires. Forest Ecology and Management 261, 2049–2067. Purdam, P.J., 2007. 3D-Rapic—The Australian radar visualisation system. Preprints, ITT Industries, 2008. ENVI: The Environment for Visualising Images. ITT Visual 33rd AMS Conf. Radar Meteorology, Cairns, Australia, 6–10 August 2007, Amer. Information Solutions, Boulder, Colorado. Meteor. Soc. Keetch, J.J., Byram, G.M., 1968. A drought index for forest fire control. Southeast Rawson, R.P., Billing, P.R., Duncan, S.F., 1983. The 1982–83 forest fires in Victoria. Forest Exp. Sta., USDA Forest Service Res. Pap. SE-38, pp. 32. Australian Forestry 46 (3), 163–172. Keeves, A., Douglas, D.R., 1983. Forest fires in South Australia on 16 February 1983 Sardoy, N., Consalvi, J.L., Porterie, B., Fernandez-Pello, A.C., 2007. Modeling transport and consequent future forest management aims. Australian Forestry 46 (3), and combustion of firebrands from burning trees. Combustion and Flame 150, 148–162. 151–169. Key, C.H., Benson, N.C., 2002. Post-fire assessment by remote sensing on National Siddaway, J.M., Petelina, S.V., 2011. Transport and evolution of the 2009 Australian Park Service Lands. Poster abstract. In: Coffelt, J.L., Livingston, R.K. (Eds.), U.S. Black Saturday bushfire smoke in the lower stratosphere observed by OSIRIS on Geological Survey Wildland Fire Workshop, Los Alamos, NM, October 31– Odin. Geophysical Research Letters 116, D06203. November 3, 2000. USGS Open-File, Report 02–11, pp. 56. Sneeuwjagt, R.J., Peet, G.B., 1985. Forest fire behaviour tables for western Australia. Luke, R.H., McArthur, A.G., 1978. Bushfires in Australia. Aust. Goverm. Publishing Department of Conservation and Land Management, pp. 59. Serv., Canberra, ACT, pp. 359. Sullivan, A.L., Matthews, S. Determining landscape fine fuel moisture content of the Madgwick, H.A.J., 1983. Seasonal changes in the biomass of a young Pinus radiata Kilmore East ‘Black Saturday’ wildfire using spatially-extended point-based stand. New Zealand Journal of Forestry Science 13, 25–36. models. Submitted to Environmental Modelling and Software. Matthews, S., 2006. A process-based model of fine fuel moisture. International Sullivan, A.L., McCaw, W.L., Cruz, M.G., Matthews, S., Ellis, P.F., 2012. Fuel, fire Journal of Wildland Fire 15 (2), 155–168. weather and fire behaviour in Australian ecosystems. In: Bradstock, R.A., Gill, McArthur, A.G., 1966. Weather and grassland fire behaviour. Commonw. Aust., A.M., Williams, R.D. (Eds.), Flammable Australia. CSIRO Publishing, Collingwood. Forest and Timber Bureau, For. Res. Inst., Canberra, Leaflet 100, pp. 23. Teague, B., McLeod, R., Pascoe, S., 2010. 2009 Victorian Bushfires Royal Commission McArthur, A.G., 1967. Fire behaviour in eucalyptus forests. Commonw. Aust., Dep. Final Report. State of Victoria, Melbourne. Nat. Devel., Forest and Timber Bureau, Canberra, Leaflet No. 107, pp. 36. Van Wagner, C.E., 1968. The line intersect method in forest fuel sampling. Forest McArthur, A.G., 1969. The fire control problem and fire research in Australia. In: Science 14 (1), 20–26. Proceedings of the 1966 Sixth world forestry congress, vol. 2. Spanish Minister Williams, D.F., 1976. Forest fuels in unthinned radiata pine stands. Australian of Agriculture, Madrid, pp. 1986–1991. Forestry 39, 238–244.