View
1
Download
0
Category
Preview:
Citation preview
Time submitted: 20/03/2020 07:59:07 PM
Submission Number: NND.001.00076
Submission Of: Professor David Lindenmayer
Your Details
Email address:Phone:Preferred means of contact: Email What is your submission based on? I am making this submission based on my professional knowledge, qualifications or experience or on behalf of a group or organisationWhat is your area of professional expertise? If you are lodging your submission on behalf of a group or organisation, what is the name of the group or organisation?
Your Submission
In your experience, what areas of the bushfire emergency response worked well?Communications from RFS and CFA and Victorian Government
Public support for those firefighters
News coverage from ABC24In your experience, what areas of the bushfire emergency response didn’t work well?Australian Government statements ‐ the Federal Government did not co‐ordinate well with State Governments ‐ who by and large did a good job under the circumstancesIn your experience, what needs to change to improve arrangements for preparation, mitigation, response and recovery coordination for national natural disaster arrangements in Australia?My written attachment spells out my concerns regarding
1. How forestry operations can make forests more prone to high‐severity fire2. The critical need for long‐term monitoring spanning multiple fires to truly understand fire dynamics and patterns of post‐fire ecosystem and wildlife recovery. 3. The need to adopt the science of fire impacts and house loss ‐ hazard reduction burning works, in part, when conducted near houses and when done frequently. It does not work under extreme fire weather. Is there anything else you would like to tell the Royal Commission?I have written a detailed submission with accompanying attachments that are based on my 37 years of experience in working on fires, forests and ecosystem responses. Do you agree to your submission being published? Yes I agree to my submission being published in my nameSupporting material provided:Lindenmayer Bushfires Royal Commission submission combined.pdfLindenmayer Bushfires Royal Commission submission.pdfLindenmayer Bushfires Royal Commission submission.docx2017 Please do not disturb ecosystems further ‐ NatureEcolEvol.pdf2012 Land management practices associated with house loss in wildfires ‐ PLoSOne.pdf2009 Effects of logging on fire regimes in moist forests ‐ ConsLett.pdf2014 Nonlinear effects of stand age on fire severity ‐ ConsLett.pdf
Page 1 of 1
1
SUBMISSION TO THE BUSHFIRES ROYAL COMMISSION
Professor David Lindenmayer AO
March 2020
To whom it may concern,
I am a fire and forest ecologist who has worked on the impacts of fires on the environment, house loss and ecosystem resilience for more almost 37 years. I am widely regarded as one of the world’s leading ecologists having received numerous international awards for my work. I have the world’s highest citation record for any forest ecologist (63,000 citations and an H-index of 123). I was elected to the Australian Academy of Science in 2008 and made an Officer of the Order of Australia (AO) in 2014.
My insights on fire come from working in the wet forests of Victoria, the softwood plantations of southern New South Wales, the coastal forests and heathland of New South Wales, and the temperate woodlands of inland Victoria, New South Wales and southern Queensland. Fire regimes and impacts have been a fundamental component of all my long-term research and monitoring.
My submission relates to several key areas associated with the targeting of prescribed burning to protect human lives and infrastructure, the effects of forest management on fire severity, and the damage caused by post-fire logging on the environment. I have summarized key findings from our work as a series of dot points in each section. I have provided citations to peer-reviewed science that is the quantitative evidence for the statements that I have made. I am happy to provide copies of this material to Royal Commission is required. I am also more than happy to be called before the commissioners to provide further insights to the material summarized in this submission.
Yours sincerely,
Professor David Lindenmayer AO BSc, Dip Ed, PhD, DSc, FAA, FESA
NND.001.00076.01_0001
2
LOGGING EFFECTS ON SUBSEQUENT FIRE SEVERITY
A series of major studies has shown that the severity of wildfires is influenced by the condition of the forest at the time of the fire (1, 2).
• A detailed and thorough empirical study after the 2009 Black Saturday wildfires in Victoria showed that logged forests that had been regenerated after harvesting were at 7 times greater risk of burning at high severity than older, unlogged forests (2).
• The elevated high severity fire risk in forests that have been logged and regenerated lasts for at least three decades after timber harvesting. The shape of the fire response curve is distinctly non-linear.
• Work by pro-forestry advocates claiming there is no link between forestry operations and fire risk (e.g. (3)) is flawed (4). This is because it did not analyze the data in an appropriate way and failed to check for non-linear patterns (as shown by (2)).
• The likely reasons for elevated fire severity in logged forests is the extensive amount of logging slash that is left behind (which contributes to forest fuel), the loss of mesic understorey plants such as tree ferns in logged areas (5, 6) (which leads to a drying of the forest), and the densely stocked stands created by reseeding after logging.
• Recent studies have found more evidence of logging-related fire risks in the areas burned in the 2019-2020 wildfires (Taylor et al., 2020, scientific article in re-review).
• The logging-triggered increases in fire risk are evident in forests globally (1).
• Expert statistical science is needed to detect non-linear fire risk effects such as those detected by (2). The work by Taylor et al. in 2014 (2) and 2020 (above) was overseen by leading statistical experts. The work conducted by Attiwill et al. (3) was not underpinned by anyone with statistical expertise and is widely acknowledged to be flawed.
In summary, logging adds significantly to fire risks in Australian forests. We have recommended a nominal distance of 2 km for logging exclusion zones around regional settlements. However, to date, Victorian Government agencies such as VicForests have largely ignored the science on this important topic. Indeed, as of 2019, the latest iteration of the Timber Release Plan (7) shows that logging coupes have been planned to be harvested near regional settlements in Central and Eastern Victoria. This poses an unnecessary and unacceptable added fire risk to these regional communities.
NND.001.00076.01_0002
3
THE IMPACTS OF POST-FIRE “SALVAGE” LOGGING
Many forest ecosystems were extensively damaged by the wildfires that occurred in 2019-2020. States like NSW have embarked on post-fire logging (sometimes called salvage logging). Extensive work from around the world (e.g. (8, 9)) shows that post-fire logging has major impacts on ecosystems. The key effects are as follows:
• Bird populations are severely reduced in salvage logged areas (10).
• Soils remain highly depleted of key soil nutrients for up to 80 years (and possibly longer) (11).
• Plant communities are radically altered, with moist forest elements like tree ferns severely depleted (5, 12).
• The recovery of natural forest vegetation is impeded (5, 12, 13).
• Habitat suitability for threatened cavity-dependent mammals is impaired for up to 170 years (14).
• Populations of insects and other key forest biota are detrimentally affected (9, 15).
• Salvage logged and regenerated areas can be highly prone to further fire (13).
In summary, all available data indicate that post-fire salvage logging is THE most highly detrimental and impactful form of forest logging.
NND.001.00076.01_0003
4
PRESCRIBED BURNING, HOUSE LOSS AND PROTECTION OF HUMAN LIVES
Prescribed burning is a key part of the management approaches to protect human lives and property. There is a strong body of research on house loss (16). That work shows:
• Prescribed burning will have some effect on fire risk reduction for house loss if it is done close to houses and done frequently (16). That is, the quality (rather than the quantity) of burning is important.
• Given the dot point above, extensive prescribed burning in remote areas will not prevent house loss or damage to infrastructure.
• The main driver of fire behaviour is climate and weather. Therefore, property damage will be inevitable under extreme conditions.
• Extensive house loss and loss of life was apparent at Marysville in the 2009 wildfires where prescribed burning was done around 8 weeks prior to the February 7 2009 conflagration.
• Some forest ecosystems, such as tall wet forests, sub-tropical rainforests, and southern conifer forests, should not be subject to any prescribed burning, otherwise the ecosystems will be lost completely.
• Calls for more prescribed burning of recently burned forests are misguided and wrong. Many forests, such as those in East Gippsland, have been subject to up to 4 fires in the past 25 years (when they should burn no more than once every 50-100 years or even longer) (Lindenmayer and Taylor, 2020, in press). These ecosystems need less (not more) fire if they are to recover.
• Efforts to reduce property damage and loss during wildfires must extend beyond hazard reduction burning. They must include (among others) more considered urban planning with houses not established in highly fire-prone areas, better house design, and clearing of vegetation within 30 m of homes (16, 17).
• There are useful lessons in the application of fire management that can be gained from exploring indigenous uses of fire (18). However, the notion of indigenous fire management is complex because different First Nations people burnt country in different ways and for different purposes (and often not to control subsequent wildfires) (17). Moreover, the widespread existence of even-aged forests prior to European settlement shows that extreme wildfires occurred even during the era of indigenous people as the sole land managers of the Australian continent.
• Simple metrics for targets for the area of a jurisdiction to be burned can have a highly perverse effect. For example, following the Royal Commission into the 2009 Black Saturday fires in Victoria, a recommendation was made to burn 5% of the State every
NND.001.00076.01_0004
5
year. The only way to achieve such a high target was to burn remote areas with limited human infrastructure. Yet the areas that need to be subject to regular hazard reduction burning are those close to human settlements (16). This kind of perverse corruption of flawed metrics is well known in many fields (from economics to education as well as fire management) and is called Goodhart’s Law. It is most commonly expressed as “When a measure becomes a target, it ceases to be a good measure” (19).
In summary, prescribed burning is useful but it needs to be conducted close to human infrastructure and conducted frequently. Many ecosystems do not need more fire (given their recent exposure to fire, including many fires in the past 25 years). Some ecosystems should not be subject to any prescribed fire at all. Australian governments at all levels must investment more in natural resource management agencies so that they can conduct more hazard reduction burning in areas where it matters – close to the peri-urban interface.
NND.001.00076.01_0005
6
THE CRITICAL NEED FOR A LONG-TERM MONITORING PROGRAM TO UNDERSTAND POST-FIRE OF ECOSYSTEMS
A key issue is that understanding the response of ecosystems to wildfire demands long-term environmental monitoring. This is fundamentally important because the effects of fire and the strength of post-fire recovery is influenced by conditions prior to the fire (e.g. vegetation age) and the number of previous fires that have occurred in a given ecosystem. Long-term monitoring spanning multiple fires is critical to understanding many key aspects of fire dynamics and post-fire recovery. However, Australia axed its long-term ecological monitoring capability in 2017.
• Australia used to have a Long-term Ecological Research Network (LTERN). It cost $1.2m per annum to run (as there was major co-investment from many partners) and fire monitoring was a fundamental part of LTERN’s activities.
• Inexplicably, LTERN was axed by its parent body in 2017 and Australia no longer has a Long-term Ecological Research Network that would have been fundamentally important for understanding post-fire recovery.
• This is a major deficiency that could be readily remedied with a limited level of investment. Indeed, such an investment is critical if Australia is to have any hope of developing reliable environmental prediction systems (20).
In summary, Australian governments at all levels must invest in a Long-term Ecological Research and Monitoring Network so that we will have the capability to document how environments recover following wildfire.
NND.001.00076.01_0006
7
THE IMPACT OF SEVERE FIRES ON THE ECONOMY IN URBAN, REGIONAL, RURAL AND REMOTE AREAS
Wildfires can have a significant negative impact on a wide range of economic values.
• Forests in water catchments should be maintained at the oldest age possible because old forests generate the most water for human consumption and allied human activities such as irrigation for agriculture (21-24). Wildfires will reduce the age of the forest in many forest types (because those fires can be what are called stand-replacing fires). Fires will therefore the amount of water available from those catchments which have been burnt. The costs of alternative sources of water when yields are reduced by disturbances in forests (such as by wildfire or logging) are substantial. For example, the costs of desalinated water in Victoria to offset losses in water yield due to fire and logging in the water catchments that supply Melbourne (and areas north of the Great Divide) are estimated to be $1650 per ML more than conventional water catchment supplied water (25).
• The impacts of wildfire on timber yields has been poorly accounted for by forest management agencies in various states. For example, in Victoria, the State logging agency, VicForests, has consistently failed to account for the loss of wood associated with wildfires – even though it manages some of the most fire-prone forest types on earth (26). This has led to significant overcutting of remaining unburned forests (because of unmodified harvest rates from a smaller remaining unburned forest estate). The key lesson here is that forest management agencies and other agencies need to far better account and plan for the wood and other resource losses that are inevitable in highly fire-prone environments like those that characterize much of Australia.
NND.001.00076.01_0007
8
References 1. Lindenmayer, D. B., Hunter, M. L., Burton, P. J., Gibbons, P., Effects of logging on
fire regimes in moist forests. Conservation Letters 2, 271-277 (2009). 2. Taylor, C., McCarthy, M. A., Lindenmayer, D. B., Non-linear effects of stand age on
fire severity. Conservation Letters 7, 355-370 (2014). 3. Attiwill, P. M., et al., Timber harvesting does not increase fire risk and severity in wet
eucalypt forests of southern Australia. Conservation Letters 7, 341-354 (2014). 4. Bradstock, R., Price, O. F., Logging and fire in Australian forests: errors by Attiwill et
al. (2014). Conservation Letters 7, 419-420 (2014). 5. Blair, D., McBurney, L., W., B., Banks, S., Lindenmayer, D. B., Disturbance gradient
shows logging affects plant functional groups more than fire. Ecological Applications 26, 2280-2301 (2016).
6. Ough, K., Murphy, A., The effect of clearfell logging on tree-ferns in Victorian wet forest. Australian Forestry 59, 178-188 (1996).
7. VicForests, "Approved Timber Release Plan 2019." Available at: http://www.vicforests.com.au/planning-1/timber-release-plan-1/approved-timber-release-plan-december-2019. Accessed 5 May 2019.
8. Lindenmayer, D. B., Burton, P. J., Franklin, J. F. Salvage Logging and its Ecological Consequences. (Island Press, 2008)
9. Thorn, S., et al., Impacts on salvage logging on biodiversity: a meta-analysis. Journal of Applied Ecology 55, 279-289 (2018).
10. Lindenmayer, D. B., McBurney, L., Blair, D., Wood, J., Banks, S. C., From unburnt to salvage logged: quantifying bird responses to different levels of disturbance severity. Journal of Applied Ecology 55, 1626-1636 (2018).
11. Bowd, E. J., Banks, S. C., Strong, C. L., Lindenmayer, D. B., Long-term impacts of wildfire and logging on forest soils. Nature Geoscience 12, 113-118 (2019).
12. Bowd, E. J., Lindenmayer, D. B., Banks, S. C., Blair, D. P., Logging and fire regimes alter plant communities. Ecological Applications 28, 826-841 (2018).
13. Donato, D. C., et al., Post-wildfire logging hinders regeneration and increases fire risk. Science 311, 352 (2006).
14. Lindenmayer, D. B., Ough, K., Salvage logging in the montane ash eucalypt forests of the Central Highlands of Victoria and its potential impacts on biodiversity. Conservation Biology 20, 1005-1015 (2006).
15. Leverkus, A. B., Lindenmayer, D. B., Thorn, S., Gustaffson, L., Salvage logging in the world’s forests: Interactions between natural disturbance and logging need recognition. Global Ecology and Biogeography 27, 1140-1154 (2018).
16. Gibbons, P., et al., Land management practices associated with house loss in wildfires. PLOS One 7, e29212 (2012).
17. Cary, G., Lindenmayer, D. B., Dovers, S. eds (2003) Australia Burning: Fire Ecology, Policy and Management Issues. (CSIRO Publishing, Melbourne).
18. Bowman, D. M. Bushfires: a Darwinian perspective. Australia burning: Fire ecology, policy and management issues, eds G. Cary, D. B. Lindenmayer, & S. Dovers (CSIRO Publishing, Melbourne). (2003)
19. Koehrsen, W., "Unintended consequences and Goodhart's Law." Available at: https://towardsdatascience.com/unintended-consequences-and-goodharts-law-68d60a94705c. Accessed 19 March 2020.
20. Lindenmayer, D. B., Developing accurate prediction systems for the terrestrial environment. BMC Biology 16, Art 16:42 (2018).
21. Langford, K. J., Moran, R. J., O'Shaughnessy, P. J. The Coranderrk Experiment - the effects of roading and timber harvesting in mature Mountain Ash forest on streamflow
NND.001.00076.01_0008
9
and quality. The First National Symposium on Forest Hydrology, eds E. M. O'Loughlin & L. J. Bren (Institution of Engineers, Canberra), pp 92-102. (1982)
22. Langford, K., Change in yield of water following a bushfire in a forest of Eucalyptus regnans. Journal of Hydrology 29, 87-114 (1976).
23. Taylor, C., Blair, D., Keith, H., Lindenmayer, D. B., Modelling water yields in response to logging and Representative Climate Futures. . Science of the Total Environment 688, 890-902 (2019).
24. Viggers, J., Weaver, H., Lindenmayer, D. B. Melbourne's Water Catchments: Perspectives on a World-Class Water Supply. (CSIRO Publishing, 2013)
25. Vardon, M., May, S., Keith, H., Lindenmayer, D. B., Accounting and valuing the ecosystem services related to water supply in the Central Highlands of Victoria, Australia. Ecosystems Services 39, 101004 (2019).
26. Lindenmayer, D., Halting natural resource depletion: Engaging with economic and political power. The Economic and Labour Relations Review 28, 41-56 (2017).
NND.001.00076.01_0009
POLICY PERSPECTIVE
Effects of logging on fire regimes in moist forestsDavid B. Lindenmayer1, Malcolm L. Hunter2, Philip J. Burton3, & Philip Gibbons1
1 Fenner School of the Environment and Society, The Australian National University, Canberra, ACT, 0200, Australia2 Department of Wildlife Ecology, University of Maine, Orono, ME, USA3 Canadian Forest Service and University of Northern British Columbia, 3333 University Way, Prince George, BC, V2N 4Z9, Canada
KeywordsBiodiversity; conservation; forest management;
logging; wildfire.
CorrespondenceDavid Lindenmayer, Fenner School for
Environment & Society, WK Hancock Building
[43], The Australian National University, Biology
Place, Canberra, ACT, 0200. Tel: +61 2 6125
0654; fax: +61 2 6125 0757. Email:
david.lindenmayer@anu.edu.au
Received: 12 August 2009; accepted 29
September 2009.
doi: 10.1111/j.1755-263X.2009.00080.x
Abstract
Does logging affect the fire proneness of forests? This question often arises af-ter major wildfires, but data suggest that answers differ substantially amongdifferent types of forest. Logging can alter key attributes of forests by changingmicroclimates, stand structure and species composition, fuel characteristics, theprevalence of ignition points, and patterns of landscape cover. These changesmay make some kinds of forests more prone to increased probability of ignitionand increased fire severity. Such forests include tropical rainforests where firewas previously extremely rare or absent and other moist forests where naturalfire regimes tend toward low frequency, stand replacing events. Relationshipsbetween logging and fire regimes are contingent on forest practices, the kindof forest under consideration, and the natural fire regime characteristic of thatforest. Such relationships will influence both the threat of fire to human lifeand infrastructure and biodiversity conservation. We therefore argue that con-servation scientists must engage in debates about fire and logging to providean environmental context to guide considered actions.
Introduction
Does logging reduce the fire proneness of forests? Thisquestion is often posed after major wildfires, especiallythose marked by substantial loss of human life or infras-tructure, such as occurred in February 2009 in south east-ern Australia, the worst fires in Australia’s history withthe loss of 173 lives and more than 3000 homes. In thewake of fires such as these, calls for forests to be logged toprevent major wildfires have been made by senior publicofficials (Tuckey 2001) and by a key lobby group (Na-tional Association of Forest Industries 2009a,b,c). Similararguments have also characterized fire and forest man-agement debates in western North America (DellaSalaet al. 2004; Odion et al. 2004). For example, Aber et al.(2000, p. 12) noted that “conversion of old growth forestsin the Pacific Northwest [of the USA] has sometimesbeen justified on grounds that it reduced the potentialfor catastrophic fire.” They further stated that perceptionsthat managed (logged) landscapes are less susceptible towildfire than unmanaged ones are “an article of faith.”Indeed, the opposite may be the case in some forests as
we show in this article through a brief examination ofrelationships between logging and several aspects of fireregimes. This is an important issue because it could haveprofound consequences for how forests are managed, in-cluding some that are currently reserved. As a conse-quence, the issue has been raised in post fire commissionsof inquiry in places like Australia and Canada. Potentialchanges in natural fire regimes underpinned by rapid cli-mate change (Cary 2002; Lenihan et al. 2003; Wester-ling et al. 2006; Flannigan et al. 2008; Cochrane & Bar-ber 2009) and hence interactions between managementpractices and altered climate further underscore the im-portance of this issue.
Our focus is on relationships between industrial loggingpractices in native forests (i.e., not plantations) and alter-ations to natural fire regimes (sensu Gill 1975) that mightinclude (among others) changed susceptibility to ignition,altered fire severity, altered fuel loads and fuel condi-tion, and changed fire frequency. Altered fire regimes canhave significant negative effects on biodiversity in moistforests (Holdsworth & Uhl 1997; Brown et al. 2004; Nosset al. 2006b; Lindenmayer et al. 2008), especially those
Conservation Letters 2 (2009) 271–277 Copyright and Photocopying: c©2009 Wiley Periodicals, Inc. 271
NND.001.00076.01_0010
Logging effects on fire regimes D. B. Lindenmayer et al.
forest types where wildfires are extremely rare or even anovel kind of major natural disturbance (e.g., some kindsof tropical rainforest, Uhl & Kauffman 1990; Cochrane &Barber 2009).
We consider industrial logging and forest managementto include the array of activities associated with the har-vesting of timber and pulpwood from a forest includingthe construction of road networks, the cutting of trees,and postharvesting stand regeneration. We do not dis-cuss in detail the broader issues of forest fire managementas this is a vast literature. Similarly, we do not exam-ine the already extensively explored topic of promotinggreater congruence between natural disturbance regimesand human disturbance regimes, notably logging (Hunter2007). However, we note that natural fire regimes cannotsimply be replaced with regulated disturbance by logging(Hunter 2007). This is because, in part, many elementsof forest flora and fauna depend on particular fire returnintervals and associated habitat features (Saint-Germainet al. 2004). Logging operations also do not provide thediversity of habitats and micro site conditions found af-ter wildfires (Haeussler & Kneeshaw 2003; Lindenmayeret al. 2008). We also of course recognize, but do not dis-cuss, an extreme response to managing forest fires, whichis to remove forests and the fuel they support, altogether.Forests and their susceptibility to fire are characterized bya continuum of precipitation and humidity ranging fromrelatively moist to relatively dry; we focus primarily onfire in relatively moist forests where fires naturally occurat a lower frequency relative to dry forests. Relationshipsbetween some kinds of logging practices (e.g., thinningoperations) and fire regimes may differ between moistforests and dry forests (e.g., Covington 2003; Noss et al.
2006a) and we briefly discuss this issue toward the endof this article.
This article is based on our past experience in workingin different forest types coupled with a recent (3 August2009) search of the fire, fire management, forest manage-ment, and conservation biology literature. Although oursearch was extensive and encompassed more than 650articles, we fully acknowledge that it was not comprehen-sive and we only touch on key points rather than exam-ine each in detail. However, to the best of our collectiveknowledge, there has been to date no detailed publishedreview of the how industrial logging policies and practicescan alter fire regimes.
Logging and fire regimes
Logging can change forests in at least five interrelatedways that could influence wildfire frequency, extentand/or severity. These include changing: (1) microcli-
mates, (2) stand structure and species composition, (3)fuel characteristics, (4) the prevalence of ignition points,and (5) patterns of landscape cover (Figure 1)
Changes in microclimate
The removal of trees by logging creates canopy openingsand this in turn alters microclimatic conditions, especiallyincreased drying of understorey vegetation and the for-est floor (Ray et al. 2005; Miller et al. 2007). As with theinfluence of forest edges (Harper et al. 2005), microcli-mate effects (including fuel drying) associated with for-est harvesting can be expected to be greatest where theunmodified forest is moist. Work in tropical rainforestssuggests that when microclimatic conditions are alteredby selective logging, the number of dry days needed tomake a forest combustible is reduced (Kauffman & Uhl1991; Holdsworth & Uhl 1997; Malhi et al. 2009). In onestudy, uncut native forest would generally not burn after>30 rainless days but selectively logged forest would burnafter just 6–8 days without precipitation (Uhl & Kauffman1990). Similarly, Nepstad et al. (1999) estimated that log-ging increased the flammability of tropical rainforest by14–50%.
Changes in stand structure and plant speciescomposition
Many studies document how logging alters the structureand species composition of forest (reviewed by Hunter1999; Lindenmayer & Franklin 2002). Such changes notonly alter microclimatic conditions as described above,but also can change stocking densities and patterns oftrees, inter crown spacing, and other forest attributessuch as plant species composition. These changes can,in turn, influence fire regimes (e.g., Ray et al. 2005).For example, logging in some moist forests in southeastern Australia has shifted the vegetation composi-tion toward one more characteristic of drier forests thattend to be more fire prone (Mueck & Peacock 1992).Research in western North America indicates that log-ging related alterations in stand structure can increaseboth the risk of occurrence and severity of subsequentwildfires through changes in fuel types and conditions(Thompson et al. 2007). Similarly, in Asian rainforests,post fire salvage logging changed the vegetation com-position towards more fire-prone grassland taxa, whichin turn damaged fire sensitive remnant rainforest stands(van Nieuwstadt et al. 2001).
272 Conservation Letters 2 (2009) 271–277 Copyright and Photocopying: c©2009 Wiley Periodicals, Inc.
NND.001.00076.01_0011
D. B. Lindenmayer et al. Logging effects on fire regimes
Figure 1 Hypothesized changes to variables that affect: (A) fire risk at the
scale of a single stand (i.e., microclimatic dryness, fine fuels, the preva-
lence of fire-prone species) and (B) fire risk at the scale of a landscape
(i.e., ignition points, amount of forest edge) in which the same number of
stands are logged per annum. The y-axis is relative abundance with zero
equal to pre-logging levels. The x-axis is time since first logging, that is, it
is assumed that no logging has occurred at time zero.
Changes in fuel characteristics
Logging can alter fire regimes by changing the amount,type, and moisture content of fuels (Perry 1994;Weatherspoon & Skinner 1995; Thompson et al. 2007;Krawchuk & Cumming 2009). As an example, work inwestern North America has highlighted how post firesalvage logging created additional fine fuels and led toelevated short-term risks of subsequent fires (Donatoet al. 2006). Whelan (1995) noted that clearfelling ofmoist forests in southern Australia led to the develop-ment of dense stands of regrowth saplings that createdmore available fuel than if the forest was not clear-felled. Large quantities of logging slash created by har-vesting operations can sustain fires for longer than fuelsin unlogged forest and also harbor fires when conditionsare not suitable to facilitate flaming combustion or thespread of fire (Cochrane & Schulze 1999). Holdsworth
& Uhl (1997) quantified increased fuel drying in selec-tively logged Amazonian rainforest, and these effects de-clined with increasing time since logging as openingsregenerated.
Change in ignition points
The road networks required for logging operations cre-ate an increased number of ignition points for wildfires.A substantial increase in ignitions and fire frequency inRussian boreal forests (Achard et al. 2006) has been at-tributed, in part, to roads built for logging and mining(Dienes 2004; Bradshaw et al. 2009). Even natural light-ning initiated ignition points may be influenced by log-ging. In Canadian mixedwood boreal forests, fire initia-tion following lightning strikes is more likely to occur inharvested areas because of increased fine fuels resulting
Conservation Letters 2 (2009) 271–277 Copyright and Photocopying: c©2009 Wiley Periodicals, Inc. 273
NND.001.00076.01_0012
Logging effects on fire regimes D. B. Lindenmayer et al.
from logging slash and this effect can remain for 10–30years following logging (Krawchuk & Cumming 2009).
Change in the spatial pattern of stands
Logging operations change natural patterns of spatial jux-taposition of different kinds of forests stands (i.e., pat-terns of landscape heterogeneity) (Franklin & Forman1987). This, in turn, can change spatial contagion inthe spread of wildfire through landscapes (Whelan 1995;Bradshaw et al. 2009) with some areas traditionally char-acterized by an absence of fire becoming more suscepti-ble to being burned by fires that spread from adjacent,more flammable, logged areas (Holdsworth & Uhl 1997;Perry 1998; Nepstad et al. 1999; Malhi et al. 2009). Simi-larly, forest edges created by logging and by logging roadscan become sites for fire incursions into adjacent forests(Cochrane & Laurance 2002). Empirical analysis in Cana-dian forests (Arienti et al. 2006) has failed to support thepresumed efficacy of road networks in facilitating wildfirecontainment and prompt fire suppression.
An alternative perspective fromdry forests
Relationships between logging and the frequency, extentand severity in some kinds of dry forests can differ frommoist forests (Noss et al. 2006b). These include forestswhere prolonged fire suppression activities have alterednatural fire regimes by increasing fuel loads and therebyelevated the risk of uncharacteristic high severity wild-fires (Harrod et al. 2009). Examples include ponderosapine (Pinus ponderosa) forests of the south western UnitedStates (Covington 2003; Noss et al. 2006a), dry east sideconiferous forests of the Pacific Northwest (Spies et al.2006; Harrod et al. 2009) and the pine forests of the southeastern United States (Phillips & Waldrop 2008). In theseforests, tree removal can be employed as an appropriaterestoration technique if thinning is aimed at removingunnaturally high fuel loads, thereby reducing the like-lihood of inappropriate high severity wildfires (Noss et al.2006b; Spies et al. 2006). Nevertheless, if thinnings areleft on site rather than taken out of the forest for disposal,these operations too can elevate the risk of unplanned ig-nitions (Schroeder et al. 2006). Carefully prescribed treeremovals also can be the first step in recreating a modern(although somewhat crude) analogue of past fire regimes(Covington 2003; Noss et al. 2006a). In dry forests that donot require restoration, the key questions are likely to be:how does logging affect the amount and condition of fueland the likelihood of ignition events?
Concluding remarks and policyimplications
The likelihood of human caused ignitions and the ac-cumulation of dry fuels are the basis for longstandingforestry practices such as “closing” forests to industrialoperations during extreme fire weather, and widespreadprescriptions for slash disposal, respectively. It has beenargued by some that, “industrial logging was a source ofalmost unprecedented holocausts. . .” (Pyne 1982, p. 182)in the past. Contrary to claims by some commentators(e.g., National Association of Forest Industries 2009a,b,c),industrial logging is likely to make some kinds of forestsmore, not less, prone to an increased probability of ig-nition (Krawchuk & Cumming 2009) and increased fireseverity and/or fire frequency (Uhl & Kauffman 1990;Thompson et al. 2007; Bradshaw et al. 2009; Malhi et al.2009). Such places include tropical rainforests where firewas previously extremely rare or absent (Uhl & Kauff-man 1990; Barlow & Peres 2004; Malhi et al. 2009),and other moist forests where natural fire regimes tendtoward low frequency, stand replacing events (Whelan1995; Odion et al. 2004; Bradshaw et al. 2009). These al-tered fire regimes can, in turn, have significant negativeeffects on a range of elements of forest biodiversity (Uhl& Kauffman 1990; Lindenmayer & Franklin 2002; Barlow& Peres 2004; Cochrane & Barber 2009).
Relationships between industrial forest managementand fire regimes are contingent on the kind of forest un-der consideration and the natural fire regime characteris-tic of that forest (Brown et al. 2004; Noss et al. 2006b). De-spite the importance of understanding such relationships,studies directly examining them are not particularly com-mon in the majority of forest ecosystems (but see Uhl &Kauffman 1990; Odion et al. 2004; Thompson et al. 2007)and this suggests an important role for additional researchin many parts of the world. These investigations could in-clude post hoc studies of fire ignition and severity in forestssubject to different management regimes (e.g., Weather-spoon & Skinner 1995; Thompson et al. 2007; Krawchuk& Cumming 2009). Such additional studies are essentialfor at least three key reasons. First, relationships betweenindustrial logging and wildfire are likely to be importantin many kinds of forests worldwide—as suggested by theexamples touched on in this article. Hence, it is criticalto identify and then manage the factors that may exacer-bate problems associated with altered fire regimes (Malhiet al. 2009). Second, climate change is likely to drive sub-stantial changes in fire regimes (Cary 2002; Westerlinget al. 2006; Flannigan et al. 2008; Pittock 2009). If indus-trial logging changes fire proneness, then interactions be-tween logging and climate change could lead to cumu-lative negative impacts, including those on biodiversity.
274 Conservation Letters 2 (2009) 271–277 Copyright and Photocopying: c©2009 Wiley Periodicals, Inc.
NND.001.00076.01_0013
D. B. Lindenmayer et al. Logging effects on fire regimes
Conversely, recent work in Amazonia suggests that somekinds of forest may have some inherent resilience to cli-mate change through maintaining mesic microclimateconditions if other agents such as logging are left undis-turbed (Malhi et al. 2009). Third, a better understand-ing of relationships between logging and wildfire willimprove policy making and forest management. For ex-ample, in moist forests there may be a case to createbuffer areas adjacent to human settlements. In addition,there may be a strong case to exclude logging from thoseareas where past human disturbances (like timber har-vesting) have been limited (Cochrane & Barber 2009).This is because logging induced alterations in landscapecover patterns can take prolonged periods to reverse andhence associated changes in fire susceptibility also may belong lived (Perry 1998). More refined studies of relation-ships between industrial logging and wildfire also mightidentify ways to manage post harvesting slash (e.g., pre-scribed burning, biofuel production) to reduce fire risks(Weatherspoon & Skinner 1995).
Perfunctory responses to natural resource managementproblems are commonplace after major natural distur-bance events that have catastrophic effects on humansand on infrastructure (Lindenmayer et al. 2008). Calls tolog forests to save them (Tuckey 2001) are overly simplis-tic. In this case, fire and forest management recipes suit-able in one situation (e.g., for restoring the natural fireregime of a dry forest) might be inappropriate (and evencounter productive) in another (e.g., a relatively moistforest) (Brown et al. 2004). In both situations, manage-ment actions will influence the threat of fire to humanlife and infrastructure and also affect all other aspects ofthe forest (e.g., biodiversity and the provision of ecosys-tem services; Barlow & Peres 2004; Phillips & Waldrop2008). Therefore, conservation scientists must stronglyengage with these issues in public fora. They need to ar-gue that environmental context is critically important toguide considered actions.
Acknowledgments
This work was informed through discussions with J.Beck, R. Bradstock, G. Cary, S. Cumming, D. DellaSala,J. Franklin, A.M. Gill, H. Keith, G. Likens, B. Mackey,F. Schmiegelow, T. Spies, and R. Williams. We thanktwo referees and C. Bradshaw and J. Barlow for insight-ful comments, which improved an earlier version of themanuscript.
References
Aber, J., Christensen N., Fernandez I. et al. (2000) Applying
ecological principles to management of the U.S. National
Forests. Issues Ecol 6, 1–10.
Achard, F., Mollicone D., Stibig H.J. et al. (2006) Areas of
rapid forest-cover change in boreal Eurasia. Forest Ecol
Manage 237, 322–334.
Arienti, M.C., Cumming S.G., Boutin S. (2006) Empirical
models of forest fire initial attack success probabilities: the
effects of fuels, anthropogenic linear features, fire weather,
and management. Can J Forest Res 36, 3155–3166.
Barlow, J., Peres C.A. (2004) Ecological responses to El
Nino-induced surface fires in central Brazilian Amazonia:
management implications for flammable tropical forests.
Phil Trans R Soc Lond 359, 367–380.
Bradshaw, C.J., Warkentin I.G., Sodhi N.S. (2009) Urgent
preservation of boreal carbon stocks and biodiversity.
Trends Ecol Evol 24, 541–548.
Brown, R.T., Agee J.K., Franklin J.F. (2004) Forest restoration
and fire: principles in the context of place. Conserv Biol 18,
903–912.
Cary, G. (2002) Importance of a changing climate for fire
regimes in Australia. Pages 26–48 in R.A. Bradstock, J.E.
Williams, A.M. Gill, editors. Flammable Australia.
Cambridge University Press, Melbourne.
Cochrane, M.A., Barber C.P. (2009) Climate change, human
land use and future fires in the Amazon. Glob Change Biol
15, 601–612.
Cochrane, M.A., Laurance W.F. (2002) Fire as a large-scale
edge effect in Amazonian forests. J Trop Ecol 18, 311–325.
Cochrane, M.A., Schulze M.D. (1999) Fire as a recurrent
event in tropical forests of the eastern Amazon: effects of
forest structure, biomass, and species composition.
Biotropica 31, 2–16.
Covington, W.W. (2003) The evolutionary and historical
context. Pages 26–47 in P. Friederici, editor. Ecological
restoration of Southwestern Ponderosa Pine Forests. Island Press,
Washington, D.C.
DellaSala, D., Williams J.E., Williams C.D., Franklin J.F.
(2004) Beyond smoke and mirrors: a synthesis of fire
policy and science. Conserv Biol 18, 976–986.
Dienes, L. (2004) Observations of the problematic potential of
Russian oil and the complexities of Siberia. Eurasian Geog
Econ 45, 319–345.
Donato, D.C., Fontaine J.B., Campbell J.L., Robinson W.D.,
Kauffman J.B., Law B.E. (2006) Post-wildfire logging hin-
ders regeneration and increases fire risk. Science 311, 352.
Flannigan, M.D., Stocks B.J., Turetsky M.R., Wotton M.
(2008) Impacts of climate change on fire activity and fire
management in the circumboreal forest. Glob Change Biol
14, 1–12.
Franklin, J.F., Forman R.T. (1987) Creating landscape
patterns by forest cutting: ecological consequences and
principles. Landscape Ecol 1, 5–18.
Gill, A.M. (1975) Fire and the Australian flora: a review.
Australian Forestry 38, 4–25.
Harper, K.A., Macdonald S.E., Burton P.J. et al. (2005) Edge
influence on forest structure and composition in
fragmented landscapes. Conserv Biol 19, 768–782.
Conservation Letters 2 (2009) 271–277 Copyright and Photocopying: c©2009 Wiley Periodicals, Inc. 275
NND.001.00076.01_0014
Logging effects on fire regimes D. B. Lindenmayer et al.
Harrod, R.J., Peterson D.W., Povak N.A., Dodson E.K. (2009)
Thinning and prescribed fire effects on overstorey tree and
snag structure in dry coniferous forests of interior Pacific
Northwest. Forest Ecol Manage 258, 712–721.
Haeussler, S., Kneeshaw D. (2003) Comparing forest
management to natural processes. In P.J. Burton, C.
Messier, D.W. Smith, W.L. Adamowicz, editors. Towards
sustainable management of the boreal forest. NRC Research
Press, Ottawa, Canada.
Holdsworth, A.R., Uhl C. (1997) Fire in Amazonian
selectively logged rain forest and the potential for fire
reduction. Ecol Appl 7, 713–725.
Hunter, M.L. (1999) Managing biodiversity in forest ecosystems.
Cambridge University Press, London.
Hunter, M.L. (2007) Core principles for using natural
disturbance regimes to inform landscape management.
Pages 408–422 in D.B. Lindenmayer, R.J. Hobbs, editors.
From principles for practice for the ecological management of
landscapes. Blackwell Publishing, Oxford.
Kauffman, J.B., Uhl C. (1991) Interactions of anthropogenic
activities, fire, and rainforests in the Amazon Basin. Pages
117–134 in J.G. Goldammer, editor. Fire in the Tropical
Biota. Springer-Verlag, New York.
Krawchuk, M.A., Cumming S.G. (2009) Disturbance history
affects lightning fire initiation in the mixedwood boreal
forest: observations and simulations. Forest Ecol Manage
257, 1613–1622.
Lenihan, J.M., Drapek R., Bachelet D., Neilson R.P. (2003)
Climate change effect on vegetation distribution, carbon,
and fire in California. Ecol Appl 13, 1667–1681.
Lindenmayer, D.B., Burton P.J., Franklin J.F. (2008) Salvage
logging and its ecological consequences. Island Press,
Washington, D.C.
Lindenmayer, D.B., Franklin J.F. (2002) Conserving forest
biodiversity: a comprehensive multiscaled approach. Island Press,
Washington, D.C.
Malhi, Y., Aragao L.E., Galbraith D. et al. (2009) Exploring the
likelihood and mechanism of a climate-change-induced
dieback of the Amazon rainforest. PNAS,
doi:10.1073/pnas.0804619106.
Miller, S.D., Goulden M.L., da Rocha H.R. (2007) The effect of
canopy gaps on subcanopy ventilation and scalar fluxes in
tropical forest. Agric Forest Meteor 142, 25–34.
Mueck, S.G., Peacock R. (1992) Impacts of intensive timber
harvesting on the forests of East Gippsland, Victoria. VSP
technical report no. 15. Department of Conservation and
Environment, Melbourne, Australia.
National Association of Forest Industries (2009a) National
Bushfire Summit Urgently Needed [WWW document].
Available from: http://www.nafi.com.au/userfiles/media1/
National%20bushfire%20summit%20urgently%
20needed%20160209.pdf. Accessed 3 August 2009.
National Association of Forest Industries (2009b) Forest
management must be based on science, not outdated ideas
[WWW document]. Available from: http://www.nafi.com.
au/userfiles/media1/Forest%20management%20must%
20be%20based%20on%20science,%20not%20outdated%
20ideas%20240709.pdf. Accessed 3 August 2009.
National Association of Forest Industries (2009c) Bushfires
flame debate over national park management [WWW
document]. Available from: http://www.nafi.com.au/
userfiles/media1/NAFI%20op%20ed%20Bushfires%
20flame%20debate%20over%20national%20park%
20management%20200109.pdf. Accessed 3 August
2009.
Nepstad, D.C., Verissimo A., Alencart A. et al. (1999)
Large-scale impoverishment of Amazonian forest by
logging and fire. Nature 398, 505–508.
Noss, R.F., Beier P., Covington W. et al. (2006a) Integrating
ecological restoration and conservation biology: a case
study for Pondersoa Pine ecosystems of the Southwest.
Restor Ecol 14, 4–10.
Noss, R.F., Franklin J.F., Baker W.L., Schoennagel T., Moyle
P.B. (2006b) Managing fire-prone forests in the western
United States. Frontiers Ecol Environ 4, 481–487.
Odion, D.C., Frost E.J., Strittholt J.R., Jiang H., Dellasala D.A.,
Moritz M.A. (2004) Patterns of fire severity and forest
conditions in the western Klamath Mountains, California.
Conserv Biol 18, 927–936.
Perry, D.A. (1994) Forest ecosystems. Johns Hopkins Press,
Baltimore, Maryland.
Perry, D.A. (1998) The scientific basis for forestry. Ann Rev
Ecol Syst 29, 435–466.
Phillips, R.J., Waldrop T.A. (2008) Changes in vegetation
structure and composition in response to fuel reduction
treatments in the South Carolina Piedmont. Forest Ecol
Manage 255, 3107–3116.
Pittock, A.B. (2009) Climate change: the science, impacts and
solutions. CSIRO Publishing, Melbourne.
Pyne, S.J. (1982) Fire in America: a cultural history of wildland
and rural fire. Princeton University Press, Princeton, New
Jersey.
Ray, D., Nepstad D., Moutinho P. (2005) Micrometeorological
and canopy controls of fire susceptibility in a forested
Amazon landscape. Ecol App 15, 1664–1678.
Saint-Germain, M., Drapeau P., Heebert C. (2004)
Comparison of Coleoptera assemblages from recently
burned and unburned Black Spruce forests of northeastern
North America. Biol Conserv 118, 583–592.
Schroeder, D., Russo G., Beck J., Hawkes B., Dalrymple G.
(2006) Modelling ignition probability of thinned lodgepole
pine stands. FERIC Advantage 7(12), 1–8.
Spies, T.A., Hemstrom M.A., Youngblood A., Hummel S.
(2006) Conserving old-growth forest diversity in
disturbance-prone landscapes. Conserv Biol 20, 351–362.
Thompson, J.R., Spies T.A., Ganio L.M. (2007) Reburn
severity in managed and unmanaged vegetation in a large
wildfire. PNAS 104, 10743–10748.
Tuckey, W. (2001) Protecting our forest. Commonwealth of
Australia, Canberra, Australia.
276 Conservation Letters 2 (2009) 271–277 Copyright and Photocopying: c©2009 Wiley Periodicals, Inc.
NND.001.00076.01_0015
D. B. Lindenmayer et al. Logging effects on fire regimes
Uhl, C., Kauffman J.B. (1990) Deforestation, fire
susceptibility, and potential tree responses to fire in the
Eastern Amazon. Ecology 71, 437–449.
van Nieuwstadt, M.G., Shiel D., Kartawinata D. (2001)
The ecological consequences of logging in the burned
forests of east Kalimantan, Indonesia. Conserv Biol 15,
1183–1186.
Weatherspoon, C.P., Skinner C.N. (1995) An assessment of
the factors associated with damage to tree crowns from
1987 wildfires in northern California. Forest Sci 41,
430–451.
Westerling, A.L., Hidalgo H.G., Cayan D.R., Swetnam T.W.
(2006) Warming and earlier spring increase western U.S.
forest wildfire activity. Science 313, 940–943.
Whelan, R.J. (1995) The ecology of fire. Cambridge University
Press, Cambridge.
Editor: Dr. Jos Barlow
Conservation Letters 2 (2009) 271–277 Copyright and Photocopying: c©2009 Wiley Periodicals, Inc. 277
NND.001.00076.01_0016
Land Management Practices Associated with House Lossin WildfiresPhilip Gibbons1*, Linda van Bommel1, A. Malcolm Gill1, Geoffrey J. Cary1, Don A. Driscoll1, Ross A.
Bradstock2, Emma Knight3, Max A. Moritz4, Scott L. Stephens4, David B. Lindenmayer1
1 The Fenner School of Environment and Society, The Australian National University, Canberra, Australian Capital Territory, Australia, 2 Centre for Environmental Risk
Management of Bushfires, University of Wollongong, Wollongong, New South Wales, Australia, 3 Centre for Mathematics and its Applications, The Australian National
University, Canberra, Australian Capital Territory, Australia, 4 Ecosystem Sciences Division, Department of Environmental Science, Policy and Management, University of
California, Berkeley, California, United States of America
Abstract
Losses to life and property from unplanned fires (wildfires) are forecast to increase because of population growth in peri-urban areas and climate change. In response, there have been moves to increase fuel reduction—clearing, prescribedburning, biomass removal and grazing—to afford greater protection to peri-urban communities in fire-prone regions. Buthow effective are these measures? Severe wildfires in southern Australia in 2009 presented a rare opportunity to addressthis question empirically. We predicted that modifying several fuels could theoretically reduce house loss by 76%–97%,which would translate to considerably fewer wildfire-related deaths. However, maximum levels of fuel reduction are unlikelyto be feasible at every house for logistical and environmental reasons. Significant fuel variables in a logistic regressionmodel we selected to predict house loss were (in order of decreasing effect): (1) the cover of trees and shrubs within 40 mof houses, (2) whether trees and shrubs within 40 m of houses was predominantly remnant or planted, (3) the upwinddistance from houses to groups of trees or shrubs, (4) the upwind distance from houses to public forested land (irrespectiveof whether it was managed for nature conservation or logging), (5) the upwind distance from houses to prescribed burningwithin 5 years, and (6) the number of buildings or structures within 40 m of houses. All fuel treatments were more effectiveif undertaken closer to houses. For example, 15% fewer houses were destroyed if prescribed burning occurred at theobserved minimum distance from houses (0.5 km) rather than the observed mean distance from houses (8.5 km). Ourresults imply that a shift in emphasis away from broad-scale fuel-reduction to intensive fuel treatments close to property willmore effectively mitigate impacts from wildfires on peri-urban communities.
Citation: Gibbons P, van Bommel L, Gill AM, Cary GJ, Driscoll DA, et al. (2012) Land Management Practices Associated with House Loss in Wildfires. PLoS ONE 7(1):e29212. doi:10.1371/journal.pone.0029212
Editor: Rohan H. Clarke, Monash University, Australia
Received September 8, 2011; Accepted November 22, 2011; Published January 18, 2012
Copyright: � 2012 Gibbons et al. This is an open-access article distributed under the terms of the Creative Commons Attribution License, which permitsunrestricted use, distribution, and reproduction in any medium, provided the original author and source are credited.
Funding: The authors received financial support from the Environmental Decisions Hub of the Australian Government’s National Environmental ResearchProgram and the Australian Research Council Centre of Excellence for Environmental Decisions. The funders had no role in study design, data collection andanalysis, decision to publish, or preparation of the manuscript.
Competing Interests: The authors have declared that no competing interests exist.
* E-mail: Philip.Gibbons@anu.edu.au
Introduction
Peri-urban communities in fire-prone regions around the world
are at increasing risk from unplanned fires (wildfires) because of
population growth [1,2,3] and climate change [4,5,6,7]. The
potential consequences of these factors were illustrated by recent
major wildfires in California (26 deaths, 3361 houses lost) [8],
Russia (54 deaths, circa. 2000 houses lost) [9] and Australia (173
deaths, 2133 houses lost) [10]. The behaviour of wildfires is
primarily determined by weather, terrain and fuel [11]. Fuel in
vegetation is often the easiest of these to manipulate [12]. Thus,
there have been moves to increase the area of fuel reduction in
many fire-prone regions [10,13,14].
Common fuel-reduction treatments employed in fire-prone
landscapes are clearing, prescribed burning, grazing and mechanical
removal of biomass (e.g., thinning) [6,12,13,15]. These treatments
are often undertaken at broad-scales and distant from peri-urban
communities. For example, in the United States of America, 89% of
all fuel-reduction treatments undertaken on federal lands were
.2.5km from the wildland urban interface [13]. Fuel treatments can
be expensive [13] and can have undesirable health [16] and
environmental [17] impacts (although not in all cases [18]). Yet,
evidence that these treatments mitigate impacts on peri-urban
communities from wildfires remains extremely limited [1].
Houses are a critical asset to protect during wildfires because
most wildfire fatalities occur among people evacuating late from,
sheltering in, or defending them [19]. Houses are destroyed during
wildfires when exposed to flames in adjacent fuel, radiant heat
from nearby fuel (#40m) [20], or airborne embers and firebrands
originating in nearby and distant fuel (typically,10 km) [21,22].
However, the relative importance of these different fuels—and
therefore the relative effectiveness of different fuel treatments in
protecting houses—have not been examined empirically. This is
because wildfires are a difficult phenomenon to study [1,23].
Large, destructive fires cannot be lit experimentally, house loss
during wildfires is often aggregated, preventing replication of
landscape-scale variables, and adequate pre- and post-fire data are
not always available. Thus, there are few wildfires that lend
themselves to empirical research on the effects of the full range of
fuel treatments on house loss.
PLoS ONE | www.plosone.org 1 January 2012 | Volume 7 | Issue 1 | e29212
NND.001.00076.01_0017
Wildfires in south-eastern Australia in 2009 destroyed a large
population of houses in landscapes with a mix of housing densities,
terrains and fuel types, and occurred in landscapes where there
were adequate pre- and post-fire data. These wildfires therefore
provided a rare opportunity to robustly quantify and compare the
effectiveness of different fuel types and different fuel treatments on
house loss during wildfire.
Results
To quantify the relative effects of different fuels on house loss we
sampled 499 houses and at each house recorded 24 potential
explanatory variables representing the three principal drivers of
fire behaviour (i.e. weather, terrain and fuel) [11]. We sampled
extremes in several of these variables not achieved in previous
studies. For example, the Forest Fire Danger Index (FFDI) [24] at
sampled houses ranged from 5 to 189, slope ranged from 0.3 to
22.6u, the percent of cleared land upwind from houses ranged
from 0 to 100% and the percent of land prescribe-burnt within 5
years upwind from houses ranged from 0 to 36.4%.
A model to predict house lossA logistic regression model we selected to predict house loss
contained eight significant explanatory variables (Table 1). This
model indicated that a greater proportion of houses were lost
where: there was a higher % cover of trees and shrubs within
40 m; the vegetation within 40 m was dominated by remnant
native (rather than planted) vegetation; there were more buildings
within 40 m; groups of trees or shrubs were closer in the upwind
direction; forest burnt within 5 years in the upwind direction was
distant rather than close; and houses were closer to public land
(had less private land) in the upwind direction (Figure 1). In the
best alternative logistic model we identified, variables representing
the amount of land that is not State Forest and the amount of land
that is not National Park replaced the amount of private land
upwind from houses (together the former are negatively correlated
with the latter). This alternative logistic model indicated that, other
things being equal, houses were at similar risk when they occurred
close to either National Park or State Forest. None of the
interactions we tested (see Materials and Methods) were significant
in the selected model.
The selected logistic regression model included several variables
in addition to fuel that also affect fire behaviour. Other things
being equal, weather had a strong effect, with a greater proportion
of houses lost at higher levels of temperature and wind speed and
lower levels of relative humidity (measured using FFDI). We
included an ‘‘autocovariate’’ [25] in the selected logistic regression
model to account for spatial autocorrelation between houses
within 1 km of each other (see Materials and Methods). No
variables representing terrain were significant in the selected
model.
A Hosmer–Lemeshow test for the selected logistic model
indicated that observed house loss was not significantly different
from predicted house loss (P = 0.487). The area under the Receiver
Operating Characteristic Curve (AUC) indicated that the fitted
logistic model correctly discriminated between burnt and unburnt
houses 80% of the time.
Predictions from the fitted logistic model indicated that reducing
fuel could substantially reduce the number of houses destroyed
during severe wildfires. With variables representing fuel held at
observed minimum loads (i.e., 10% cover of planted trees and
shrubs within 40 m from houses, 100 m to the nearest trees and
shrubs in the upwind direction, 500 m to forest burnt #5 years
ago in the upwind direction and no buildings within 40 m); and
other variables fixed at their means (i.e., FFDI, the distance to
public land and the covariate representing spatial autocorrelation),
we predicted that 4.6% (61.9% s.e.m) of all houses would be
destroyed. Thus, under otherwise average conditions observed
during these wildfires, minimizing key fuels at every house could
potentially reduce the percent of houses destroyed from the
observed value of 35.0% to a predicted mean value of 4.6%
(61.9% s.e.m). This equates to a reduction in the number of
houses lost of 76%–97% (95% confidence interval). However, this
level of fuel management is unlikely to be realised at all houses for
reasons we outline in the Discussion.
The relative effects of fuel variables on house lossWe used mean predictions from the selected logistic model to
examine the relative effects of different fuels on house loss
(Figure 1). Predictions for each fuel variable were made with the
other significant explanatory variables held at their means, but
with FFDI held at 100. This is the value for FFDI above which
64% of houses have been destroyed by wildfires in Australia [26]
and the value for FFDI that invokes the highest level of public
warning in fire-prone regions of Australia. We predicted that
reducing remnant native vegetation around houses (within 40 m)
from 90% cover (the observed maximum) to 5% cover reduced the
likelihood of house loss by 43%. That is, every 10% reduction in
remnant native vegetation cover around houses reduced the
likelihood of house loss by approximately 5%. Thirty eight percent
fewer houses were destroyed if surrounded (within 40 m) by
predominantly planted vegetation rather than predominantly
remnant native vegetation. Twenty six percent fewer houses were
lost if further (100 m) from the nearest group of trees or shrubs in
the upwind direction, compared with houses adjacent (0 m) to
groups of trees or shrubs in the upwind direction. Compared with
houses located 10 m from public forest, 14% fewer houses were
lost if 200 m from public forest, and 26% fewer houses were lost if
2 km from public forest (the average distance between houses and
public forest). On average, 15% fewer houses were lost if
prescribed burning within 5 years was undertaken 0.5 km upwind
from houses (the nearest distance between houses and prescribed
Table 1. The selected logistic regression model used topredict the proportion of houses lost during the sampledwildfires.
VariableCoefficient ±s.e.m P(.|z|)
Intercept 25.68761.073 0.000
Tree and shrub cover (%) within 40 meters (m) 0.02260.006 0.000
Log10 (FFDI) 1.06260.3076 0.000
Log10 (amount of land not burnt within 5 years (m)) 0.56560.216 0.001
Vegetation type within 40 m (planted) - -
Vegetation type within 40 m (remnant) 0.72660.246 0.003
Log10 (amount of private land (m)+1) 20.47960.199 0.016
Log10 (distance to nearest trees and shrubs (m)+1) 20.57460.191 0.003
Log10 (buildings within 40 m) 0.96360.483 0.046
Autocovariate (spatial autocorrelation) 4.80061.110 0.000
Significant explanatory variables, their coefficients and P-values in the logisticmodel selected to predict the (logit or log-odds) proportion of housesdestroyed during wildfire. Vegetation type is a categorical variable with‘planted’ being the reference level. The autocovariate represents spatialautocorrelation between neighbouring houses.doi:10.1371/journal.pone.0029212.t001
Land Management and House Loss in Wildfires
PLoS ONE | www.plosone.org 2 January 2012 | Volume 7 | Issue 1 | e29212
NND.001.00076.01_0018
burning), rather than 8.5 km upwind from houses (the average
distance between houses and prescribed burning). We predicted a
3% increase in the number of destroyed houses with every
additional building or shed located within 40 m.
Discussion
We predicted that modifying key fuels could substantially
reduce house loss during wildfires that burn in extreme fire
weather conditions. Many deaths occur among people sheltering
in houses during wildfires (69% of lives lost during the 2009
wildfires examined here were in houses [10]), so managing key
fuels could also save considerable lives.
The relative effects of fuel treatments on house lossWe found that modifying fuel closer to houses was a more
effective way to reduce house loss than modifying fuel distant from
houses. In severe fire weather (FFDI = 100), we predicted that
reducing trees and shrubs from 90% cover to 5% cover within
40 m of houses could potentially reduce house loss by an average
of 43%, making this the single most effective fuel treatment that we
measured. We predicted that conversion from predominantly
remnant to predominantly planted vegetation within 40 m from
houses could reduce house loss by 38%. Increasing the upwind
distance from houses to groups of trees and shrubs from zero to
100 m would reduce the number of houses lost by an average of
26%. The distance between houses and public forest had a similar
effect. Twenty six percent fewer houses were lost 2 km from public
land (the mean observed distance) compared with houses adjacent
to public land. We predicted that an average of 15% fewer houses
were destroyed when prescribed burning was undertaken 0.5 km
from houses (the minimum distance we observed), compared with
8.5 km from houses (the mean distance we observed). One less
building within 40 m from a house reduced the likelihood of
Figure 1. Individual effects (mean ± s.e.m.) of fuel variables in the logistic model used to predict the proportion of houses lostduring wildfire. Each prediction was made with the other significant explanatory variables held at their means and FFDI fixed at 100, which is thevalue above which 64% of houses have been destroyed in wildfires in Australia [26]. Magneta (pink) lines are predictions for vegetation within 40 mof houses that is predominantly remnant native vegetation and cyan lines are predictions for vegetation within 40 m of houses that is predominantlyplanted vegetation.doi:10.1371/journal.pone.0029212.g001
Land Management and House Loss in Wildfires
PLoS ONE | www.plosone.org 3 January 2012 | Volume 7 | Issue 1 | e29212
NND.001.00076.01_0019
destruction by an average of 3%, making this the least influential
fuel variable in the selected logistic regression model.
Our finding that fuel management close to houses was more
effective than fuel management further from houses can be
explained by the behaviour of embers and radiant heat—the
principal causes of house loss during wildfires [20,22,27]. The
density of airborne embers [21] and the amount of radiant heat
[20] are greatest closer to the fuel source, which is consistent with
our results that fuel and fuel treatments closer to houses were more
strongly associated with house loss. The reduction of fuel close to
houses also increases ‘defensible space’, or the area around houses
in which suppression is most likely to be successful [23].
Prescribed burning and house lossPrescribed burning is a widely employed fuel treatment in many
regions [12,13] and many commentators identified this as the key
fuel treatment contributing to house loss in our study area.
Although there was relatively limited prescribed burning in many
parts of our study area, stratifying by this variable enabled us to
sample houses with between 0% and 36.4% of the landscape burnt
within 5 years in the upwind direction to the 2009 wildfire
boundary. Within this range of variation, we found the effect of
prescribed burning within 5 years was greatest closer to houses
(Figure 1). This pattern is consistent with our results across all fuel
variables (Figure 1). It is also noteworthy that prescribed burning
was not a significant explanatory variable in any of the feasible
logistic regression models when it was measured as the percentage
of the landscape treated in the upwind direction from houses to the
nearest 2009 wildfire boundary (rather than upwind distance from
houses to the closest prescribed burning). This suggests that the
proximity to houses of prescribed burning is more important than
the total percentage of the landscape that is prescribe-burnt. These
results are consistent with previous research indicating the effects
of prescribed burning can diminish within a short period of time
(2–6 years) [12,28,29] and in severe fire weather conditions
[8,12,30], which are the conditions when most houses are
destroyed [26]. Our results therefore indicated that prescribed
burning—when executed at the scale observed in this study—was
most effective when undertaken close to houses and at least every 5
years.
It is argued [10], based on relationships between prescribed
burning and changes to the incidence and extent of wildfires
[31,32], that prescribed burning can make control and suppression
of wildfires before they reach houses more effective if executed in
larger units and over a larger percentage of the landscape than
observed in this study. However, it remains untested whether this
strategy is effective in the extreme weather conditions. It is also
important to note that the extent of prescribed burning that is
feasible in many landscapes, including our study area, is restricted
because of the number of days per year in which weather
conditions are suitable and/or the proximity of public land to
infrastructure [12].
ConclusionsDevastating wildfires provide a window into conditions that may
become more common in the future [1,2,3,4,5,6] and therefore
represent important learning opportunities for decision-makers.
The typical response to destructive wildfires is to increase the total
area of land that is fuel-reduced [10,13]. Our results instead
indicate that a shift in emphasis from broad-scale fuel-reduction
treatments to intensive fuel treatments close to houses will more
effectively mitigate impacts from wildfires on houses. This result is
consistent with observations that the density of airborne embers
and amount of radiant heat (the principal causes of house loss
during wildfires) are greatest closer to the fuel source. This suggests
that the actions of private landholders, who manage fuel close to
houses, are extremely important when reducing risks to houses
posed by fuel. Our results are based on data collected at wildfires
in south-eastern Australia. While it has been speculated that these
conclusions apply to other regions around the world [13], the
broader applicability of our results can only be confirmed with
sampling across a broader range of fuel types and climates.
Although our results indicated that risks posed to peri-urban
communities by severe wildfires can be reduced by effectively
managing fuel, these risks cannot be eliminated by managing fuel
alone. Fuel treatments can be expensive [13] and can have
undesirable health [16] and environmental [17] impacts (but not
in all cases [18]). Therefore, intensive fuel-reduction is not always
an appropriate strategy to reduce risk posed by wildfire. Weather
strongly influenced the effect of fuel variables (Table 1), hence
other measures not accounted for here (e.g., architectural
solutions, education of residents, suppression effort, safer places,
early evacuation) [22,33,34,35] must remain part of a strategy to
mitigate increasing risks to communities from wildfires.
Overall our results clearly imply that fuel close to housing plays
a key role in house loss during wildfire, so fuel management should
be considered as part of a strategy to mitigate increasing risks to
peri-urban communities from wildfires. Future impacts from
wildfires will be reduced, and the negative effects of fuel treatments
avoided, if new peri-urban developments in fire prone regions are
restricted to areas where there is adequate separation between
high fuel loads and houses.
Materials and Methods
Study area and stratificationHouses were sampled within the boundaries of three wildfires
that ignited in the State of Victoria, south-eastern Australia on 7
February, 2009. These fires were known as: the Kilmore East fire,
which burnt 125,383ha, destroyed 1242 houses and claimed 119
lives; the Murrindindi fire, which burnt 43,159ha, destroyed 538
houses and claimed 40 lives [10]; and the Churchill fire which
burnt 25,861ha, destroyed 145 houses and claimed 11 lives (Figure
S1). The wildfire boundaries were as mapped in the FIRE_SEV09
GIS shape file provided by the Victorian Department of
Sustainability and Environment (DSE). We stratified the study
area by the three principal drivers of wildfire behaviour: weather,
terrain and fuel [11]. Weather (measured using FFDI), ranged
from 5 to 189. Slope of the terrain at each house ranged from 0.3uto 22.6u. Fuel, measured as % burnt upwind from houses within 5
years to the 2009 wildfire boundary and as the % of the landscape
cleared upwind from houses to the 2009 wildfire boundary, ranged
from 0% to 36% and 0% to 100% respectively.
Response variableOur response was a binary variable representing house loss
(intact or destroyed). To sample houses we allocated 499 points
randomly to the study area in a Geographical Information System
(GIS) in numbers proportional to the area of each stratum. We
then selected the nearest house to each point using fine-scale
(35 cm to 50 cm pixel resolution) orthorectified aerial imagery in
the visible spectrum taken between 1 and 37 months prior to the
fires. Our sampling of houses was blind in the sense that we did not
know which had been destroyed. Several variables were measured
within a circle with a 40 m radius from the centroid of each house
(Table S1). To increase the likelihood of independence between
responses we did not sample houses when these circles overlapped,
instead choosing another random point until a non-overlapping
Land Management and House Loss in Wildfires
PLoS ONE | www.plosone.org 4 January 2012 | Volume 7 | Issue 1 | e29212
NND.001.00076.01_0020
house with 40 m circle was located. We recorded damage to each
sampled house (intact or destroyed) based on a visual inspection of
the house using fine-scale (15 cm pixel resolution) orthorectified
imagery in the visible spectrum taken 17–23 days after the fires.
We judged a house as destroyed if at least part of the roof had
visibly collapsed or incinerated and judged a house as intact if the
roof remained. In all cases this distinction was clear, which is
consistent with on-ground observations by Wilson and Ferguson
[33] who rejected using a continuous or ordinal variable to
categorise house damage because virtually all houses affected by
wildfire in their study had either been destroyed or sustained only
superficial damage.
Potential explanatory variablesWe recorded 24 potential explanatory variables at each house
reflecting the three drivers of fire behaviour (i.e. weather, terrain
and fuel) [11].
Weather conditions were measured with the Forest Fire Danger
Index (FFDI) because house loss in Australian fires exhibits a higher
correlation with this index rather than any of its individual
components (i.e., wind, temperature, relative humidity and drought
factor) [26]. FFDI was calculated using the formula [24]
FFDI~2:0|exp({0:450z0:987 ln(DF){
0:0345RHz0:338Tz0:0234V )
where, DF is drought factor, RH is relative humidity (%), T is air
temperature (uC) and V is wind speed (km hr21). Weather variables
used to calculate FFDI (and to calculate wind direction for some fuel
variables in Table S1) were derived from half-hourly data recorded
at the closest weather station to each house [36]. We were advised
by the Australian Bureau of Meteorology that these were the most
reliable weather data available for our purpose. The estimated time
that fire impacted on each house was estimated from fire
progression maps provided by DSE for the Kilmore East and
Murrindindi fires and a fire progression map prepared by the
Victorian Country Fire Authority for the Churchill fire [10].
Terrain was measured as slope, topographic position and
aspect (Table S1).
Fuel was measured (a) within 40 m of the centroid of each
house, which is the approximate maximum distance that radiant
heat is likely to ignite a wooden structure [20], (b) as a percentage
of the landscape along a single transect in the upwind direction
from each house to the nearest 2009 wildfire boundary, which
ensured there was little overlap between measurements taken for
different houses at this scale, and (c) as the distance from each
house to the fuel variable in the upwind direction (Table S1). We
could not measure the distance from each house to several fuel
variables (public land, previous burning within 10 years, logging
within 30 years, National Park, State Forest) because they did not
always occur between sampled houses and the 2009 wildfire
boundary in the upwind direction. If we excluded houses that did
not have all of these fuel variables in the upwind direction then this
would bias the sample (only 13 of the 499 sampled houses had all
of the measured fuel variables in the upwind direction to the 2009
wildfire boundary). We therefore measured the amount of land
upwind from houses that did NOT contain these fuel variables.
For example, instead of recording the distance from houses to land
burnt within 5 years in the upwind direction, we recorded the
amount of land not burnt within 5 years in the upwind direction
between each house and the 2009 wildfire boundary. This enabled
us to include all randomly sampled houses in the analysis. The %
cover of trees and shrubs within 40 m from the centroid of each
house was estimated visually on the pre-fire aerial imagery by one
person (P.G.). To test the accuracy of this method, we randomly
selected 30 houses and compared our visual % cover estimates with
estimates for the same houses derived by digitising tree and shrub
cover in a GIS. Visual % cover estimates were highly correlated
with % cover estimates derived from digitising trees and shrubs
(r = 0.95, Pearson correlation coefficient). The mean (6 s.e.m.) %
cover of trees and shrubs derived from visual estimates (33.264.4)
was not significantly different from estimates derived from digitising
trees and shrubs (32.164.0) (P = 0.462, 2-tailed t-test).
Exploratory data analysisSummary statistics for the continuous potential explanatory
variables (see Table S1 for definitions) (mean, range) were: FFDI
(48, 5–189), slope (8.5u, 0.3–22.6u), aspect (186u, 25–329u),number of buildings (2, 1–9), % cover of trees and shrubs (30%,
0–90%), distance to nearest tree or shrub (2.6 m, 0–108 m),
upwind distance to nearest trees or shrubs (26 m, 0–686 m),
upwind distance to nearest block of trees (272 m, 0–3021 m),
upwind distance to mapped cleared land (773 m, 0–25121 m),
amount of private land (2145 m, 0–15280 m), % cleared (32%, 0–
100%), amount of land not burnt for #5 years (8848 m, 14–
40041 m), % of land burnt #5 years ago (2.8%, 0–36.4%),
amount of land not burnt for .5–10 years (10985 m, 14–
35157 m), % of land burnt .5–10 years ago (0.4%, 0–37.7%),
amount unlogged (9107 m, 14–36168 m), % logged (1.7%, 0–
32.9%), amount not National Park (7457 m, 14–35157 m) and
amount not State Forest (5501 m, 13–24703 m). The % of houses
with the different measured fuel variables in the upwind direction
within the 2009 wildfire boundary were: trees or shrubs (97%),
block of trees (97%), land burnt within 5 years (25%), land burnt
within .5–10 years (12%), logged within 30 years (26%), mapped
cleared land (94%), public land (75%), National Park (41%) and
State Forest (72%). The majority (86%) of the area burnt within 10
years was from prescribed fire, with the remainder burnt from
wildfire.
We constructed a correlation matrix using the Pearson
correlation coefficient (r) for all pairs of potential explanatory
variables to determine those highly correlated (r$0.7). One pair of
potential explanatory variables was highly correlated: distance
from houses to the 2009 wildfire boundary and amount of land not
burnt for .5 to 10 years (r = 0.83). Only one of each of these
variables was included in any model. Several other pairs of
variables (amount of land not burnt for #5 years, amount
unlogged, amount of land that is not National Park, amount of
land that is private and amount of land that is not State Forest)
were reasonably highly correlated (r = 0.59–0.68), but were
included in all models. The following variables had highly skewed
distributions and were therefore transformed by log10 (x) (or
log10(x+1) for variables with zeros) to give them a more
symmetrical distribution prior to statistical modelling: upwind
distance to nearest trees or shrubs, upwind distance to nearest
block of trees, upwind distance to mapped cleared land, amount of
land upwind from houses not burnt for #5 years and .5–10
years, amount of land upwind from houses that is privately owned,
buildings within 40 m of houses and the amount of land upwind
from houses that is not National Park or State Forest.
Statistical analysisWe initially examined relationships between the response and
uncorrelated potential explanatory variables using mixed-effects
modelling to account for the hierarchical structure of our data (i.e.,
houses were sampled within three separate fires which ignited at
different times of the day and in different regions). That is, there
Land Management and House Loss in Wildfires
PLoS ONE | www.plosone.org 5 January 2012 | Volume 7 | Issue 1 | e29212
NND.001.00076.01_0021
was potential for the effect on houses of the same fire to be more
alike than the effect on houses in a different fire. However, initial
analyses using Generalized Linear Mixed Modelling (GLMM),
implemented using the MGCV library in R [37], indicated that
the variable representing the three different fires had a variance
component approaching zero (,0.001). That is, the response did
not change between fires. Remaining analyses were therefore
undertaken using the more parsimonious Generalized Linear
Modelling (GLM).
We used GLM with a logit link implemented using the MASS
library in R [37] to identify fuel variables that were the best
predictors of house loss. We accounted for the influence of weather
and terrain by fitting FFDI slope, aspect and topographic position
as co-variates during model selection. We included several
interaction terms during model selection. To test whether effects
of fuel variables varied with weather conditions we included
interactions between FFDI and some fuel variables (% of
landscape not burnt in the upwind direction, amount of land
not burnt in the upwind direction, nearest upwind distance to trees
or shrubs, % cover of trees and shrubs #40 m from houses). To
test whether the effect of slope on the proportion of houses
destroyed was influenced by aspect we fit an interaction term
between slope and aspect. To test whether there was an interaction
between defensible space created by clearing close to houses and
broader-scale fuel reduction, we included interaction terms
between the % cover of trees and shrubs #40 m from houses
and the amount of previous burning in the upwind direction from
houses.
We chose a model of ‘best fit’ using stepwise selection [38]. We
first fitted a full model (with all terms) and then dropped terms
sequentially if they did not lower Akaike’s Information Criterion
(AIC). Following Venebles and Ripley [38] (pp. 175–176), we then
dropped any variables from this model if they were not significant
(P#0.05) using the traditional analysis of deviance, thus obtaining
a more parsimonious result. All predictions were made from this
single model of ‘best fit’.
Because many houses in our study area occurred in a clustered
spatial arrangement around towns (Figure S1), it follows that there
is potential for spatial autocorrelation in our data. That is, if one
house is destroyed then neighbouring houses are more likely to be
destroyed, which would violate the assumption of independence in
our fitted GLM. We tested whether residuals from the fitted GLM
were spatially autocorrelated using Moran’s I. This test was
undertaken using the Ape package in R [37], which is based on the
method described by Gittleman and Kot [39]. Moran’s I,
calculated using the residuals from the selected logistic model,
was significantly different (p,0.001) to the expected value of
Moran’s I if the residuals were distributed randomly, leading us to
conclude that there is strong evidence for spatial autocorrelation in
our data.
To account for this spatial autocorrelation we added an
‘‘autocovariate’’ [25] to the fitted GLM, which is a covariate
representing spatial autocorrelation, following the methodology for
non-normally distributed data reported in the appendix to
Dormann et al. [40]. The autocovariate was scaled from zero
(there was no relationship in the response between neighbouring
houses) to 1 (the response was identical between neighbouring
houses). The autocovariate was calculated using a matrix of
neighbours within 1 km of each house. A correlogram of Moran’s
I indicated that most spatial autocorrelation in the residuals
between neighbouring houses occurred within this distance. We
added the autocovariate to the fitted GLM and then, again using
the method of Gittleman and Kot [39], we confirmed that
Moran’s I calculated using the residuals in this new model was no
longer significantly different to expected if distributed randomly
(p = 0.845).
We reported goodness of fit for our selected logistic model using
the Hosmer-Lemeshow test calculated using the Design library in
the R statistical software [37]. We calculated AUC [41] from
observed and predicted values for this model using the package
Rocr in the R statistical software [37] to determine the probability
that true positives rank above false positives. AUC has a value
between 0.5 (a discriminating ability no better than chance) to 1
(perfect discriminating ability).
Supporting Information
Figure S1 Houses sampled in (A) the Kilmore EastMurrindindi wildfires and (B) the Churchill wildfire.Sampled houses that were intact (clear houses) and destroyed (solid
red houses) after the wildfires are illustrated.
(TIF)
Table S1 Potential explanatory variables recorded foreach sampled house.
(DOC)
Acknowledgments
Several data sets were supplied by the Victorian Department of
Sustainability and Environment (DSE) and the Australian Government’s
Bureau of Meteorology. We acknowledge several officers of the
Department of Sustainability and Environment for their cooperation and
advice when seeking and using spatial data. Owen Price (University of
Wollongong) supplied some data and provided helpful advice. Tony
Bannister (Bureau of Meteorology) provided meteorological data and
provided advice on its use. Jeff Wood, Ross Cunningham and Wade
Blanchard (The Australian National University) provided advice on
experimental design and statistical analysis. We thank Rohan Clarke and
two anonymous referees for helpful comments on an earlier version of the
manuscript.
Author Contributions
Conceived and designed the experiments: PG EK AMG DD GC RB DL
LvB MM SS. Performed the experiments: PG LvB. Analyzed the data: PG
Lv . Wrote the paper: PG EK AMG DD GC RB DL LvB MM SS.
References
1. Mell WE, Manzello SL, Maranghides A, Butry D, Rehm RG (2010) The
wildland-urban interface fire problem - current approaches and research needs.
International Journal of Wildland Fire 19: 238–251.
2. Hughes R, Mercer D (2009) Planning to reduce risk: the wildfire management
overlay in Victoria, Australia. Geographical Research 47: 124–141.
3. Radeloff VC, Hammer RB, Stewart SI, Fried JS, Holcomb SS, et al. (2005) The
wildland-urban interface in the United States. Ecological Applications 15:
799–805.
4. Pinol J, Terradas J, Lloret F (1998) Climate warming, wildfire hazard, and
wildfire occurrence in coastal eastern Spain. Climatic Change 38: 345–357.
5. Pausas JG (2004) Changes in fire and climate in the eastern Iberian Peninsula
(Mediterranean basin). Climatic Change 63: 337–350.
6. Westerling AL, Hidalgo HG, Cayan DR, Swetnam TW (2006) Warming and
earlier spring increase western U.S. forest wildfire activity. Science 313:
940–943.
7. Liu Y, Stanturf J, Goodrick S. Trends in global wildfire potential in a changing
climate. Forest Ecology and Management 259: 685–697.
8. Keeley JE, Fotheringham CJ, Moritz MA (2004) Lessons from the October 2003
wildfires in Southern California. Journal of Forestry 102: 26–31.
9. Vasquez T (2011) The Russian inferno of 2010. Weatherwise 62: 20–25.
10. Teague B, McLeod R, Pascoe S (2010) 2009 Victorian Bushfires Royal
Commission. Final Report. Melbourne: Parliament of Victoria.
11. Pyne SJ, Andrews PL, Laven RD (1996) Introduction to Wildland Fire. Second
Edition. New York: John Wiley & Sons.
Land Management and House Loss in Wildfires
PLoS ONE | www.plosone.org 6 January 2012 | Volume 7 | Issue 1 | e29212
B
NND.001.00076.01_0022
12. Fernandes PM, Botelho HS (2003) A review of prescribed burning effectiveness
in fire hazard reduction. International Journal of Wildland Fire 12: 117–128.13. Schoennagel T, Nelson CR, Theobald DM, Carnwath GC, Chapman TB
(2009) Implementation of National Fire Plan treatments near the wildland-urban
interface in the western United States. Proceedings of the National Academy ofSciences of the United States of America 106: 10706–10711.
14. Agee JK, Skinner CN (2005) Basic principles of forest fuel reduction treatments.Forest Ecology and Management 211: 83–96.
15. Bradstock RA, Gill AM (2001) Living with fire and biodiversity at the urban
edge: in search of a sustainable solution to the human protection problem insouthern Australia. Journal of Mediterranean Ecology 2: 179–195.
16. Wotowa G, Trainer M (2000) The influence of Canadian forest fires on pollutantconcentrations in the United States. Science 288: 324–328.
17. Driscoll DA, Lindenmayer DB, Bennett AF, Bode M, Bradstock RA, et al.(2010) Resolving conflicts in fire management using decision theory: asset-
protection versus biodiversity conservation. Conservation Letters 3: 215–223.
18. Stephens SL, McIver JD, Boerner REJ, Fettig CJ, Fontaine JBH, B.R., et al. (inpress) Effects of forest fuel reduction treatments in the United States. BioScience.
19. Haynes K, Handmer J, McAneney J, Tibbits A, Coates L (2010) Australianbushfire fatalities 1900–2008: exploring trends in relation to the ‘Prepare, stay
and defend or leave early’ policy. Environmental Science & Policy 13: 185–194.
20. Cohen JD (2000) Preventing disaster: Home ignitability in the wildland-urbaninterface. Journal of Forestry 98: 15–21.
21. Koo E, Pagni PJ, Weise DR, Woycheese JP (2010) Firebrands and spottingignition in large-scale fires. International Journal of Wildland Fire 19: 818–843.
22. Maranghides A, Mell WE (2011) A case study of a community affected by theWitch and Guejito wildland fires. Fire Technology 47: 379–420.
23. Gill AM, Stephens SL (2009) Scientific and social challenges for the
management of fire-prone wildland-urban interfaces. Environmental ResearchLetters 4: 1–10.
24. Noble IR, Bary BAV, Gill AM (1980) McArthur’s fire-danger meters expressedas equations. Australian Journal of Ecology 5: 201–203.
25. Augustin NH, Mugglestone MA, Buckland ST (1996) An Autologistic Model for
the Spatial Distribution of Wildlife. Journal of Applied Ecology 33: 339–347.26. Blanchi R, Lucas C, Leonard J, Finkele K (2010) Meteorological conditions and
wildfire-related house loss in Australia. International Journal of Wildland Fire19: 914–926.
27. Chen K, McAneney J (2004) Quantifying bushfire penetration into urban areasin Australia. Geophysical Research Letters 31: L12212.
28. Cary GJ, Flannigan MD, Keane RE, Bradstock RA, Davies ID, et al. (2009)
Relative importance of fuel management, ignition management and weather for
area burned: evidence from five landscape-fire-succession models. International
Journal of Wildland Fire 18: 147–156.
29. Bradstock RA, Hammill KA, Collins L, Price O (2010) Effects of weather, fuel
and terrain on fire severity in topographically diverse landscapes of south-eastern
Australia. Landscape Ecology 25: 607–619.
30. Moritz MA, Keeley JE, Johnson EA, Schaffner AA (2004) Testing a basic
assumption of shrubland management fire: how important is fuel age? Frontiers
in Ecology and the Environment 2: 67–72.
31. Boer MM, Sadler RJ, Wittkuhn RS, McCaw L, Grierson PF (2009) Long-term
impacts of prescribed burning on regional extent and incidence of wildfires-
Evidence from 50 years of active fire management in SW Australian forests.
Forest Ecology and Management 259: 132–142.
32. Finney MA, Seli RC, McHugh CW, Ager AA, Bahro B, et al. (2007) Simulation
of long-term landscape-level fuel treatment effects on large wildfires. Interna-
tional Journal of Wildland Fire 16: 712–727.
33. Wilson AAG, Ferguson IS (1986) Predicting the probability of house survival
during bushfires. Journal of Environmental Management 23: 259–270.
34. Abt R, Kelly D, Kuypers M (1987) The Florida Palm Coast fire: an analysis of
fire incidence and residence characteristics. Fire Technology 23: 186–197.
35. Stephens SL, et al. (2009) Urban wildland fires: how California and other
regions of the US can learn from Australia. Environmental Research Letters 4:
014010.
36. Bureau of Meteorology (2009) Meteorological Aspects of the 7 February 2009
Victorian Fires. An Overview. Canberra: Australian Government, Bureau of
Meteorology.
37. R Development Core Team (2010) R: A Language and Environment for
Statistical Computing. Vienna, Austria: R Foundation for Statistical Computing.
38. Venables WN, Ripley BD (2002) Modern Applied Statistics with S. New York:
Springer.
39. Gittleman JL, Kot M (1990) Adaptation: statistics and a null model for
estimating phylogenetic effects. Systematic Zoology 39: 227–241.
40. Dormann CF, McPherson JM, Araujo MB, Bivand R, Bolliger J, et al. (2007)
Methods to account for spatial autocorrelation in the analysis of species
distributional data: a review. Ecography 30: 609–628.
41. Pearce J, Ferrier S (2000) Evaluating the predictive performance of habitat
models developed using logistic regression. Ecological Modelling 133: 225–245.
Land Management and House Loss in Wildfires
PLoS ONE | www.plosone.org 7 January 2012 | Volume 7 | Issue 1 | e29212
NND.001.00076.01_0023
REVIEW
Nonlinear Effects of Stand Age on Fire SeverityChris Taylor1, Michael A. McCarthy2, & David B. Lindenmayer3
1 Melbourne Sustainable Society Institute, University of Melbourne, Parkville, Victoria 3051, Australia2 School of Botany, University of Melbourne, Parkville, Victoria 3051, Australia3 Fenner School of Environment and Society, The Australian National University, Canberra, ACT 0200, Australia
KeywordsFire; fire severity; logging; forests; stand age;
probit regression; south-eastern Australia.
CorrespondenceDavid Lindenmayer, Fenner School of
Environment and Society, The Australian
National University, Canberra, ACT 0200,
Australia.
Tel: 02 61250654; fax: 02 61250746.
E-mail: david.lindenmayer@anu.edu.au
Received13 February 2014
Accepted29 June 2014
doi: 10.1111/conl.12122
Abstract
We quantify the relationship between forest stand age and fire severity using adetailed case study of Mountain Ash (Eucalyptus regnans Muell) forest burned insouth-eastern Australia in 2009. We focused on two important areas of Moun-tain Ash forest that feature a range of growth stages and disturbance histories.Using probit regression analysis, we identified a strong relationship betweenthe age of a Mountain Ash forest and the severity of damage that the forestsustained from the fires under extreme weather conditions. Stands of Moun-tain Ash trees between the ages of 7 to 36 years mostly sustained canopy con-sumption and scorching, which are impacts resulting from high-severity fire.High-severity fire leading to canopy consumption almost never occurred inyoung stands (<7 years) and also was infrequent in older (>40 years) standsof Mountain Ash. We discuss the significant forest conservation and manage-ment implications of these results for Mountain Ash forests as well as othersimilar biomes, where high-severity fire is a common form of disturbance.
Introduction
Fire is a major ecological process in ecosystems world-wide (Bowman et al. 2009). It can influence the type ofvegetation cover (Whelan 1995), levels and patterns ofbiodiversity (Lindenmayer et al. 2014a), nutrient cycling(Raison 1980), and carbon stocks (Keith et al. 2009). Italso affects human communities through loss of life andproperty (Bowman et al. 2011). The sequence and charac-teristics of fires in an area is shaped by multiple biotic, abi-otic, and anthropogenic factors that are not stable in spaceand time and create pervasive uncertainty (Dovers 2003).These uncertainties make predicting the spread and im-pacts of fires difficult (Leonard et al. 2014). The rates offire spread, fire intensities, and other properties of firevary widely (Collins et al. 2007; Gill 2012). The resultingimpact of fire on vegetation, that is, the fire severity, isalso variable (Keeley 2009).
Fire severity is defined as the extent of loss or con-sumption of the vegetation and other biomass as a resultof fire. It describes how fire intensity affects ecosystems,where direct information on fire intensity can be absentand effects are often quite variable within and between
different ecosystems (Keeley 2009). Many, often inter-related, natural factors influence fire severity. These in-clude: (1) vegetation type (Bradstock et al. 2012); (2) theamount of fuel (Sullivan et al. 2012); (3) fire weather con-ditions (e.g., temperature, relative humidity, wind speed,and drought factor); and (4) timing (time of day, seasonand time elapsed since the preceding fire) (Collins et al.2007; Price & Bradstock 2010).
Another key factor which can influence fire sever-ity is past anthropogenic disturbance, such as logging(Thompson et al. 2007; Lindenmayer et al. 2011). This hasbeen observed in tropical, temperate, and boreal forests,where anthropogenic disturbance and modification of na-tive ecosystems has not acted in isolation, but has in-teracted with natural ecosystem disturbance (Thompsonet al. 2007; reviewed by Lindenmayer et al. 2009). How-ever, the risk that anthropogenic disturbance can influ-ence fire severity is controversial (Ferguson & Cheney2011; Attiwill et al. 2013). This makes it critical to ex-amine interactions between fires and the environmentalvariables that influence them.
Here, we quantify relationships between prior anthro-pogenic disturbance and fire severity using a detailed case
Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc. 355
NND.001.00076.01_0024
Nonlinear stand age effects on fire severity C. Taylor et al.
study of the Mountain Ash (Eucalyptus regnans Muell)forests in south-eastern Australia. Parts of these forestswere burned by large wildfires in February 2009 (Cruzet al. 2012). Indeed, these fires were one of Australia’sworst natural disasters, where 173 human lives were lostand over 3,000 properties destroyed (Gibbons et al. 2012).Several studies have analyzed the behavior and impactof this fire at a broad landscape scale (Cruz et al. 2012;Price & Bradstock 2012). Our study expands on theseprevious studies by focusing exclusively on fire severityimpacts in Mountain Ash forests at the stand level. Apaucity of stand-level analysis is a key knowledge gap,given the prevalence of logged and regenerated standsthroughout not only Mountain Ash ecosystems, but alsoin other high productivity wet forest types elsewhere inAustralia (Wood et al. 2014) and overseas (e.g., Franklinet al. 2002). We addressed this important knowledge gapby addressing the question: How does stand age influence theseverity of fire in Mountain Ash forest under extreme fire weather
conditions?Major forest disturbances, including large high-severity
wildfires, are likely to increase as a consequence of cli-mate change (Bachelet et al. 2005; Cary et al. 2012). Wetherefore argue it is critical to better quantify the relation-ships between anthropogenic disturbances and fire sever-ity and, in turn, consider their implications for how forestecosystems are conserved and managed.
Methods
Study Area
Our case study focused on Mountain Ash forests affectedby the February 2009 fires in the Central Highlands ofVictoria, south-eastern Australia (Figure 1). These forestsform visually spectacular stands containing trees that canexceed 100 m in height (Ashton 1976). They providehabitat for a number of threatened and endemic species,such as the Leadbeater’s Possum (Gymnobelideus leadbeat-
eri McCoy) (Lindenmayer et al. 2013) and the Baw BawFrog (Philoria frosti Spencer) (Hollis 2004).
Forests dominated by Mountain Ash are located inwater catchments critical for supplying water to thecity of Melbourne and surrounding rural communities(Viggers et al. 2013). These forests support the largestknown stores of terrestrial ecosystem carbon, with somesites containing over 1,800 t/C/ha (Keith et al. 2009).Mountain Ash forests have been a major source of pulp-wood and sawlogs for industry since the 1930s (Lutze etal. 1999).
The main form of natural disturbance in MountainAsh forests is wildfire, which typically occurs following
dry summer conditions, particularly periods of prolongeddrought (Mackey et al. 2002). Mountain Ash forests haveexperienced many fires including those in 1851, 1898,1926, 1939, and 1983 (Griffiths 2001). The impacts of fireon these forests has been variable, and range from high-severity conflagrations resulting in stand replacement, tofires of lower severity that leave biological legacies (sensuFranklin et al. 2002) such as two or more age cohorts oftrees (i.e., multiaged stands) (McCarthy & Lindenmayer1998). Extensive salvage logging following the 1939 firesremoved many of these biological legacies, creating ex-tensive areas of even-aged stands after that conflagration(Mackey et al. 2002).
Fires of February 2009
The months of January and February 2009 were charac-terized by intense fire activity across the state of Victo-ria, with 825 wildfires occurring and collectively burning437,000 hectares (DSE 2009a). Fire activity peaked on7 February, with extremely intense fire behavior andrapid rates of spread in the lead up to, and directly fol-lowing, the passage of a cold front (Cruz et al. 2012). Atthis time, the Forest Fire Danger Index (FFDI), whichcorrelates with fire behavior and depends on tempera-ture, wind velocity, relative humidity and the Keeck–Byram Drought Index (Noble et al. 1980) (see Figure S1),reached unprecedented levels (Gellie et al. 2013). (SeeMethods section in Supporting Information for a descrip-tion of how the FFDI was derived). It was during thistime that the fire spread was at its most rapid and thefire was at its greatest intensity (Cruz et al. 2012; Gellie et
al. 2013). Overall, 169 fires were burning across the statein the days following 7 February 2009 (DSE 2009a). Twoof these, the Kilmore East and Murrindindi fires, burnedlarge areas of Mountain Ash forest (Cruz et al. 2012; Price& Bradstock 2012).
The Kilmore East and Murrindindi fires were ignitedaround 11:45 and 14:55 hours, respectively. Leading upto the wind change driven by an approaching cold front,the Kilmore East and Murrindindi fires traveled approx-imately 55 and 33 km, respectively (Gellie et al. 2013).The cold front passed across the fires between 18:10 and18:35 hours, resulting in the long north-eastern flank ofthe fire becoming the front (Cruz et al. 2012). In the hoursfollowing the south-westerly wind change, fire weatherconditions moderated and the spread of the fire slowed.However, the fire continued to burn for another 26 daysbefore it was contained. This was after efforts in controlline construction and back burning to stop the spread ofthe fire (Gellie et al. 2013).
356 Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc.
NND.001.00076.01_0025
C. Taylor et al. Nonlinear stand age effects on fire severity
Figure 1 Location of study area and extent of fire impacted area in the study region.
Fire Progression
We mapped the progression of the 2009 fires, from thepoint of ignition to their full extent, using a series ofisochrones (Gellie et al. 2013). These isochrones mark thespatial extent of the fires at 30–60 minute intervals to aspatial accuracy of 200–500 m and a temporal accuracy of5 minutes. They were interpreted from fire severity pat-terns on postfire aerial photography or remote sensingand supplemented by videos, photographs and eyewit-ness accounts obtained from local people and staff fromvarious Victorian Government organizations responsiblefor fire control. However, where such evidence was de-ficient, such as in remote areas, we used isochrones oflower confidence (Gellie et al. 2013) (Refer to Figure 2).
Sample Stands
We determined the extent of Mountain Ash forest frommaps of the “wet forest” ecological vegetation class (DSE2005). We stratified this ecological vegetation class intosix age classes at the time of the fire: 4–9 years, 10–14
years, 15–24 years, 25–40 years, 70 years, and 300 years(see Figures 3 and 4). Stand age determination was basedon multiple datasets, with logging history data informingstand ages between 0 and 40 years (DSE 2011), State For-est Resource Inventory (SFRI) data (DSE 2007a) and datapresented in Mackey et al. (2002) informing forest age of70 years, and modeled old growth informing the forestage of 300 years (DSE 2007b). Although, the modeled oldgrowth dataset does not describe stand age, the determi-nation of 300 years was based on the findings presentedby Lindenmayer et al. (2000a) (see Figures 5 and 6).
The type of disturbance resulting in the establishmentof the stands prior to the fire, varied across the age classes.Stands in age classes between 0 and 40 years were mostlyestablished following clearfell logging (DSE 2011). Standsin the 70 year age class were mostly established follow-ing the 1939 fires (Griffiths 2001). Extensive areas of thisage class were subjected to postfire (salvage) logging for20 years following the 1939 fires (Mackey et al. 2002).However, no historical records were kept on the timing,location, spatial extent and intensity of these past logging
Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc. 357
NND.001.00076.01_0026
Nonlinear stand age effects on fire severity C. Taylor et al.
Figure 2 Extent of fire affected forest before the wind change (Zone 1) and for the two hours following the wind change (Zones 2, 3, and 4). The extent
of Mountain Ash forest is shown as dark green. Sample subregions show the areas of Mountain Ash Forest targeted for analysis. R1 shows the Mountain
Ash stands at Mount Disappointment; R2 show Mountain Ash at Paradise Plains and Deep Creek.
operations, which made it difficult to determine whetherstands were established after fire or following logging.The majority of stands in the 70 year age class were even-aged across the sites at the time of the fires. This in-dicates that postfire logging had occurred because mostunlogged stands feature multiple cohorts of age classes(Lindenmayer et al. 2000a). Stands in the 300-year ageclass are located entirely within those closed water catch-ments where logging had been prohibited (Viggers et al.2013).
We identified areas of Mountain Ash forest burnt bythe 2009 fires under similar weather conditions. Ac-cording to Price & Bradstock (2012), the magnitude andspread of the 2009 fires were strongly influenced by theseverity of fire weather. We controlled for the effects offire weather by constraining our analyses to areas burntunder similar fire weather conditions at similar times,thereby enabling us to quantify the effects of other vari-
ables that can influence fire severity such as slope, aspect,and stand age (Mackey et al. 2002).
We focused our study on areas burned under extremefire weather conditions (determined as places where theFFDI exceeded an index value of 75 using the CFA FireDanger Rating System) and for the period immediatelyfollowing thereafter. The two reasons for selecting thisperiod of the fire were first, to examine the impact of thefire on Mountain Ash forest under extreme weather con-ditions, when tree death is most likely, and second, tomaintain consistency in weather conditions and the typeof fire in our analyses. Human attempts to control thefire were conducted when fire weather conditions mod-erated for several hours following the wind change. Thissignificantly alters the behavior of the fire, because back-burning is widely applied throughout native vegetationwith the aim of protecting assets, such as property andinfrastructure.
358 Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc.
NND.001.00076.01_0027
C. Taylor et al. Nonlinear stand age effects on fire severity
Figure 3 Fire zones and stand age of Mountain Ash forest across R1 (Mount Disappointment).
We mapped the spatial extent of the area burned whenthe FFDI was extreme and for the period immediately fol-lowing thereafter using the fire isochrones (Gellie et al.
2013). To facilitate our analysis, we identified two sub-regions dominated by Mountain Ash forest. One was lo-cated on the plateau surrounding Mount Disappointment(R1) and the other on the plateau encompassing ParadisePlains and Deep Creek (R2) (Figure 2). Subregions R1and R2 are 5,984 and 3,569 hectares, respectively, andapproximately 56 km apart (see Figure 2; and Figures S2and S3).
Within each subregion, we defined four time periodsrelative to the time of the wind change, which we referto as zones (Figure 2; Figures S2 and S3). For the KilmoreEast fire, which covered the R1 subregion, the zoneswere: (1) areas burnt between ignition and the windchange at 17:50 hours; (2) areas burnt between the windchange and 18:35 hours; (3) areas burnt between 18:35and 19:15 hours; and (4) areas burnt between 19:15 and20:25 hours. For the Murrindindi fire, which covered theR2 subregion, the four zones were: (1) areas between
time of ignition and the wind change at 18:45 hours; (2)areas burnt between the wind change and 19:30 hours;(3) areas burnt between 19:30 and 20:30 hours; and (4)areas between 20:30 and 21:30 hours. Zone (1) was burntunder the highest FFDI. Zones (2), (3), and (4) wereburnt under declining FFDIs in relation to time follow-ing the wind change (refer to Figure S1).
Fire Severity Data
We obtained fire severity maps for Mountain Ash forestin our two study regions from the Victorian Government(DSE 2009c). These maps were based on a fire severityindex developed through digital processing of the pre-fire and postfire imagery acquired by the SPOT 4 and 5Satellite. These data comprised the Short Wave Infrared(SWIR) and Near Infrared (NIR) bands that were ana-lyzed through a Transform Vegetation Index (TVI). Fireseverity was scaled as an index gauging the magnitude ofcanopy scorch resulting from fire (DSE 2009b). This indexwas then compared against ground-based information
Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc. 359
NND.001.00076.01_0028
Nonlinear stand age effects on fire severity C. Taylor et al.
Figure 4 Fire zones and stand age of Mountain Ash forest across R2 (Paradise Plains and Deep Creek).
and split into five classes (DSE 2009c). This process de-termined the statistical association between fire severityobserved on the ground and that derived from satellitedata. These five classes ranged from a canopy consumingfire (the most severe) to no canopy scorch (the lowest)(see Table S1).
We assumed that the top two severity classes causedwidespread tree mortality in Mountain Ash forest. Thehighest severity class was canopy consumption in which70–100% of the canopy was burnt. The second most se-vere class was canopy scorch, where 60–100% of euca-lypt and noneucalypt canopies were scorched, but theleaves remained on the branches immediately followingthe fire. We refer to these two classes as “high-severity”fire. The assumption that these two fire severity classesresulted in tree mortality was based on studies by Smith& Woodgate (1985) and Vivian et al. (2008). In a studyof fire damage following the 1983 “Ash Wednesday” firesin Mountain Ash forest that was conducted in an area lo-cated approximately 25 km south of subregion 2 in ourstudy, Smith & Woodgate (1985) found that fire severityclasses of complete canopy consumption, canopy scorch
and 75–99% canopy scorch resulted in widespread treemortality. Trees subject to lower severity fire were con-sidered capable of surviving. In Alpine Ash (Eucalyptus del-egatensis Barker) trees (which is a similarly fire-sensitivespecies to Mountain Ash), Vivian et al. (2008) found thatfire severity classes of canopy consumption and severecanopy scorch resulted in 98.9% tree mortality and lowerseverity classes resulted in only 17.5% tree mortality,with much of the prefire stand surviving.
Data Sampling
We sampled fire severity data using a square grid at 100 mintervals. The points of the grid formed our sites. Ourresponse variables were the fire severity classes that re-sulted in the mortality of Mountain Ash trees. Theseclasses were represented as either present (1) or absent(0) for each site in our analysis. We calculated thesevariables using Geostatistical Analyst software (ArcGIS10) (ESRI 2011) for fire severity patterns in the fourzones.
360 Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc.
NND.001.00076.01_0029
C. Taylor et al. Nonlinear stand age effects on fire severity
Figure 5 Fire zones and disturbance history in Mountain Ash forest across R1 (Mount Disappointment).
Our procedure generated data for 9,934 sites. Wepooled the presence and absence of canopy consump-tion and the combined canopy consumption and scorchwithin each zone and across the zones. We displayedthe results for each zone separately, which allowed usto compare the different subregions and time zones. Wealso displayed the overall average of the presence and ab-sence of canopy consumption and the combined canopyconsumption and scorch across all the zones, to drawcomparisons of fire severity between the zones. In theKilmore East fire, 1,256 sites of Mountain Ash forestburnt prior to the wind change at 17:50 hours (Kil-more East Zone 1), 2,061 sites burnt from the windchange up to 18:35 hours (Zone 2), 207 sites burnt be-tween 18:35 and 19:15 hours (Zone 3), and 300 sitesburnt between 19:15 and 20:25 hours (Zone 4). For theMurrindindi fire, 1,582 sites of Mountain Ash forestburnt prior to the wind change at 18:45 hours (Mur-rindindi Zone 1), 1,069 sites burnt from the wind changeup to 19:30 hours (Zone 2), 2,249 sites burnt between19:30 and 20:30 hours (Zone 3), and 961 sites burnt be-tween 20:30 and 21:30 hours (Zone 4). The total number
of sites was more than could be analyzed using our mul-tivariate probit modeling approach. We therefore tooka random sample of points, stratified on stand age andzone defining time of fire impact, for subsequent statis-tical analysis. Stratifying the selection by age and zoneensured good coverage of the area and previous distur-bance history of Mountain Ash forest affected by the fires.Where the requisite number of sites for an age class wasnot available in a subset, we selected all sites in thatage class. In total, we analyzed data from 633 sites (seeTable S2). We assumed that sites in Kilmore East andMurrindindi were independent of each other given thedistance between them (�56 km). This was supported bythe estimated correlations, which were very close to zeroat distances >1 km.
Statistical analyses
We analyzed our data with a multivariate probit regres-sion model that accounted for correlation in the occur-rence of fires among sites (Chib & Greenberg 1998). Ac-counting for correlation is important because not doing so
Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc. 361
NND.001.00076.01_0030
Nonlinear stand age effects on fire severity C. Taylor et al.
Figure 6 Fire zones and disturbance history in Mountain Ash forest across R2 (Paradise Plains and Deep Creek).
would lead to unmodeled pseudo-replication, with con-sequent overestimation of model precision and possiblebias of parameter estimates.
Probit regression is a form of generalized linear model-ing that uses the cumulative distribution function of thestandard normal distribution to convert a linear predictorto a probability (McCullagh & Nelder 1989). We modeledthe occurrence of a particular fire severity class as a func-tion of three explanatory variables: stand age; aspect; andslope. Because fires burn more intensively when travel-ing uphill, we measured aspect as the cosine of the an-gle between the direction of the prevailing wind and theaspect. We calculated slope in degrees. Due to the possi-ble interaction between slope and aspect on fire behav-ior, we included an interaction term for these two mea-sures. McCarthy et al. (2001) suggested that, in Moun-tain Ash forests, the probability of fire might first increasewith time since fire and then decline. More specifically,McCarthy et al. (1999) also suggested a particular func-tional form for this relationship, although it was based onexpert judgment rather than empirical data. To allow themodel to fit a functional form similar to that suggested by
McCarthy et al. (1999), we used an additive function ofthe reciprocal of stand age 1/t and the square of this term1/t2. Thus, the linear predictor was for site i:
mi = β0 + β1/ti + β2/t2i + β3ϕi + β4 cos(θi − ψi )
+β5ϕi cos(θi − ψi ), (1)
where ti is the age of the stand at the time of the fire,ϕi and θ i are the slope aspect of the site, ψ i is the di-rection from which the wind was blowing and β0–β5 areregression coefficients that were estimated. The effects oftopography (β3, β4, and β5) appeared to be weak. For ex-ample, parameter estimates for β3 and β4 were almostexactly zero, while the parameter estimate for β5 sug-gested a small and uncertain increase in risks for sitesfacing toward the prevailing wind (estimate and 95%CI of 0.18 [−0.007, 0.043]) for the canopy consump-tion model. These models with topographic effects alsoestimated almost identical effects of stand age as modelswithout these effects, so we report the results of a simpli-fied statistical model that excluded these terms:
mi = β0 + β1/ti + β2/t2i (2)
362 Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc.
NND.001.00076.01_0031
C. Taylor et al. Nonlinear stand age effects on fire severity
We used a model in which correlation in the occur-rence of high-severity fire was a declining function of thedistance between sites. The correlation in the latent nor-mal distribution between two sites i and j was:
ρi j = exp(−ϕdi j ) (3)
where dij is the distance between sites i and j, and ϕ con-trols the rate at which the correlation declines with dis-tance. Thus, the correlation approaches 1 as the distancebetween sites approaches 0 and it approaches 0 as thedistance increases. This is equivalent to the model thatMcCarthy & Lindenmayer (1998) used to simulate fire inmountain ash forest. Here, we use the same basic modelto estimate the probability of fire occurrence and the na-ture of the correlation.
We fitted the models using Bayesian MCMC methodsin OpenBUGS (Spiegelhalter et al. 2012). Vague priorswere used to ensure the results were driven by the data.For the regression coefficients (β j), normal distributionswith mean of 0 and standard deviation of 1,000 wereused. For the rate at which the correlation decline withdistance (ϕ), we used a uniform prior on the interval [0,20]. We took 50,000 MCMC samples after discarding thefirst 100,000 as a burn-in. This burn-in was more thansufficient to ensure the samples were drawn from the sta-tionary distribution, which was assessed by inspection ofhistory plots in OpenBUGS as well as from running twoMarkov chains with different initial conditions.
Results
Our analyses revealed the occurrence of canopy con-sumption was strongly correlated with the age of thestands (β1 and β2) (Figure 7). In contrast, the effects of to-pography (β3, β4, and β5) appeared to be weak. Canopyconsumption rarely, if ever, occurred in stands youngerthan 7 years. However, the probability of canopy con-sumption increased rapidly with age up to approximately15 years (mi � 0.8). In stands older than 15 years, theprobability of canopy consumption decreased with age,such that it rarely occurred in stands aged around 300years (mi � 0.1). Previous disturbance history prior to thefires was variable, with stands between 0 and 40 yearsbeing established following clearfell logging, stands of 70years being established following the 1939 fires, and ex-tensive postfire logging and stands of 300 years previouslyestablished following past fires.
The relationship between probability of canopy con-sumption and stand age was similar across the four zonesdefining time of fire impact, suggesting that the reducedprobability of canopy consumption in young stands lessthan 7 years and those aged around 300 years wasnot simply an artifact of subsampling from our dataset.
Figure 7 Probability of canopy consumption versus stand age based
on data from the Kilmore East and Murrindindi Fires during the periods
before and after the wind change on 7 February 2009. These time periods
are indicated by the different zones. The points are average ages and
proportion of forest experiencing canopy consumption for each of six age
classes, and for each of the four zones (Figure 2), and for all four zones
combined. Only data with 15 or more data points are shown to reduce
noise arising from small sample sizes. The solid line is the mean of the
posterior prediction of the probit regression model fitted to a stratified
sample of the data, and the dashed lines are 95% credible intervals.
Rather, it was spatially consistent under extreme weatherconditions and the time periods immediately followingthe wind change.
We found that canopy consumption or canopy scorchrarely occurred in stands of Mountain Ash younger than7 years. Similarly to canopy consumption alone, theproportion of forest sustaining canopy consumption orcanopy scorch increased to a maximum at around 15years of age, where most of the stand sustained a high-severity fire (mi > 0.9). The probability declined after15 years, but less quickly with increasing age than forthe probability of canopy consumption alone (Figure 8).At 300 years, the probability of stands sustaining firesof canopy consumption or canopy scorch declined toaround 0.7 (mi � 0.7). This pattern was again consistentfor all four time periods. However, the range of probabil-ity across the age class of 300 years shows a significantshift relative the range in younger age classes back to 15years (Figure 8).
The probit regression models supported the analysis ofthe raw data, although the probit regression may haveunderestimated the degree to which the probability ofsevere fire changes with forest age (Figures 7 and 8).The regression coefficient β1 is clearly greater than zeroand β2 is clearly less than zero (i.e., the 95% CIs do notencompass zero), meaning that the data strongly indi-cate that the probability of fire was maximized at inter-mediate ages (Table 1). These maxima were achieved at
Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc. 363
NND.001.00076.01_0032
Nonlinear stand age effects on fire severity C. Taylor et al.
Figure 8 Probability of canopy consumption or canopy scorch versus
stand age based on data from the Kilmore East and Murrindindi Fires
during the periods before and after the wind change on 7 February 2009.
These time periods are indicated by the different zones. The points are
average ages and proportion of forest experiencing canopy consumption
for each of six age classes, and for each of the four zones (Figure 2), and
for all four zones combined. Only data with 15 or more data points are
shown to reduce noise arising from small sample sizes. The solid line is
the mean of the posterior prediction of the probit regression model fitted
to a stratified sample of the data, and the dashed lines are 95% credible
intervals.
approximately 15 years for both classifications of severefires (Figures 7 and 8). The occurrence of high-severityfire was, as expected, clearly spatially correlated, withstrong positive correlation at short distances that declinedto near zero at distances greater than 0.5–1 km (Figure 9).
Discussion
The potential relationships between anthropogenic dis-turbance and fire severity of natural forests have beencontroversial. The review by Lindenmayer et al. (2009)precised evidence from a range of forest types aroundthe world and suggested that while strategic forms of log-ging in dry forests can reduce fire severity, the reverse istrue in moist forests, including boreal forests (Krawchuk& Cumming 2009), coniferous forest of north-westernNorth America (Thompson et al. 2007) and tropical for-
Figure 9 Estimated correlation in the occurrence of canopy consumption
(solid line) and canopy consumption or scorch (dashed line) for pairs of
points as a function of distance between them, basedon theprobit regres-
sion model (Chib & Greenberg 1998) and assuming a correlation function
given by Equation (3).
est worldwide (Cochrane & Barber 2009). In the analysesthat we report here, we identified a strong relationshipbetween the age of Mountain Ash stands and the occur-rence of canopy consumption and canopy scorch underextreme weather conditions in the February 2009 fires.We further discuss our findings in the remainder of thisarticle. We conclude with a discussion of the implicationsof our work for forest management and forest biodiver-sity conservation.
Why might stand age influence fire severity inMountain Ash forests?
Our analyses revealed that stand age was a dominantvariable influencing fire severity in stands of MountainAsh forest for our study areas. The lowest fire severity im-pacts were sustained in stands less than 7 years old. Weobserved the inverse of this trend in age classes between 7and 36 years, where the greatest proportion of stands sus-tained either canopy consumption or canopy scorch, withthe maximum around the age of 15 years. These youngeraged stands were established following clearfell logging.
Table 1 Parameter estimates of the multivariate probit regression for canopy consumption and for fires that burn or scorch the canopy. The values are
the mean of the posterior estimates and 95% credible intervals
Parameter Canopy consumption Canopy consumption or Scorch
β0 (intercept) –1.82 [–2.21, –1.38] 0.63 [0.34, 1.05]
β1 (1/t) 58.2 [41.8, 73.7] 31.2 [12.6, 44.5]
β2 (1/t2) –358.3 [–470.9, –255.6] –223.8 [–314.2, –107.4]
ϕ (decay of correlation with distance) 4.7 [2.4,8.3] 8.34 [1.81, 18.8]
364 Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc.
NND.001.00076.01_0033
C. Taylor et al. Nonlinear stand age effects on fire severity
Interestingly, maximum fire severity around this age hasbeen observed in other forest types, such as conifer plan-tations in the Klamath Mountains in southwest Oregonand northwest California in the United States (Thompsonet al. 2011). Older Mountain Ash stands of 70 years and300 years displayed low incidences of canopy consump-tion and the 300 year age class showed relatively lowerlevels of scorch, but much less than canopy consumptionalone. These stands were previously established followingmajor fire events in 1939 (70 years) (and possibly post firelogging) and after fires in the 1700s (300 years).
Several key factors may lead to different fire severityimpacts among stands of different age with stands aged7–36 years being particularly susceptible to canopy con-sumption or canopy scorching high-severity fire. First, foryoung stands established following clearfell logging, thelogging practice itself alters the architecture of the forest.Large old trees are removed or lost from these sites. Ini-tially, the slash burn following logging removes a largeproportion of the available fine fuel that would sustain afire. The newly established trees are characterized by highstem densities within the first 7 years of growth, withover 200,000 stems per hectare being recorded (Ashton1976). However, the rate of self-thinning in regeneratingstands of Mountain Ash trees increases rapidly from thisearly stage of growth until 40–80 years old, where 126–227 stems per hectare have been recorded, respectively(Ashton 1976). Similarly to young Australian AlpineAsh forests (which are ecologically similar to the standstargeted in our investigation), the vegetation of youngMountain Ash is well-aerated and comprises a horizon-tally and vertically continuous fuel layer that is highlyflammable (Bowman et al. 2014). Natural self-thinningcreates large amounts of fine fuels from suppressed plantsin the early stages of regrowth, which, upon dying anddrying out, become more susceptible to fires. In this con-text, Florence (1994, p. 24) explained that: “Where re-growth develops following a severe perturbation, the forest floorbiomass builds up rapidly to a point of peak fuel energy during
the forest’s rapid early growth stage. This point may be as soonas 35 years in stands of fast growing species.” This rapid accu-mulation of fine fuels during the early stages of growth inMountain Ash forest, and its greater susceptibility to ei-ther canopy-consuming or canopy-scorching fire, placesyounger stands at greater risk of being eliminated from asite if a high-severity fire occurs. This is because the treeswould be killed prior to producing seed (i.e., <20 yearsof age) (Mackey et al. 2002). This assumption is similar tothe finding of Bowman et al. (2014), who observed a 97%attrition rate among juvenile trees in stands of AustralianAlpine Ash forests that were recovering from a previousrecent fire. However, the reasons why the crowns of veryyoung stands (<7 years) are less likely to be damaged re-
main unclear, but they may be associated with the ab-sence of large amounts of dead fine fuels at this growthstage, possibly because the process of rapid self-thinninghas not yet commenced. The mortality of regeneration inthese stands is presently unknown and is a topic of fur-ther research.
A second reason that stands aged 7–36 years old areat risk of canopy consumption or canopy scorch may berelated to their limited height. A fire consuming ground-based fuel in such stands has the potential to burn the treecanopy at a lower severity than the canopy of an olderforest (Mackey et al. 2002; Price & Bradstock 2012). In asimilar study of conifer plantations in the United States,Thompson et al. (2011) explained that a high canopy-base-height is a likely explanation for decreasing canopydamage in older stands. A low canopy height is also com-pounded by increased bulk density, which is more abun-dant in fine fuels.
Third, young stands regenerating after clearfell logginghave an altered floristic composition which may influ-ence fire severity. Species characteristic of wet forests,such as the soft tree fern (Dicksonia antarctica), are signifi-cantly less abundant on logged sites (Ough 2001). Thesespecies cast dense shade, which influences rates of dry-ing on the forest floor and hence moisture content of finefuels (Wood et al. 2014).
Other factors may have an important effect on rela-tionships between stand age and fire severity. For ex-ample, older aged stands of Mountain Ash forest accu-mulate large quantities of coarse woody debris and theseare typically colonized by dense and luxuriant moss mats(Lindenmayer et al. 1999) that have very high water hold-ing capacity. In addition, older aged stands of Moun-tain Ash forest are more likely to develop a luxuriantunderstorey layer that can be dominated by Gondwanicrainforest elements, such as Myrtle Beech (Nothofagus
cunninghamii) (Lindenmayer et al. 2000a,b). These under-storey layers limit light penetration to the forest floor,helping development of cool and moist microclimaticconditions that are unfavorable to fire propagation. Re-duced flammability of rainforest was recognized by Woodet al. (2014).
Limitations of analysis
The key result of this empirical study was a nonlinear re-lationship between stand age and fire severity, in whichtrees in young forest are more likely to be impacted andkilled by fire than those in older forests or very youngforest (<7 years). The evidence for this relationship wascompelling and it was broadly consistent with that fromother studies, both in Australian wet eucalypt forests(e.g., Price & Bradstock 2012) and moist forests elsewhere
Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc. 365
NND.001.00076.01_0034
Nonlinear stand age effects on fire severity C. Taylor et al.
in the world (e.g., Thompson et al. 2007). Nevertheless,we recognize there were some limitations to our study,although these were unlikely to have major impacts onour general findings. First, fire severity information inour study was based on remotely sensed data calibratedagainst field observations and assigned to five fire sever-ity classes. This dataset and others similar to it used inother studies (e.g., Collins et al. 2007; Thompson et al.
2007; Bradstock et al. 2010) provide a useful general in-dication of the immediate postfire impact on vegetation,but do not distinguish between age classes nor does it in-clude measures of the long-term survival of fire-impactedstands.
A second limitation was that there has been poor his-torical recording keeping of data on the relatively largeareas of forest subject to postfire logging, particularly afterthe 1939 fires. Salvage logging removes biological legaciesremaining after a fire and which might otherwise havepersisted in the regenerating stand (Lindenmayer et al.
2008). This removes the possibility of monitoring long-term impacts resulting from fire across specific age classesthat have been targeted for such treatment.
Comparisons with other studies on theFebruary 2009 fires
The results of this investigation build upon those of pre-vious studies conducted on the 2009 fire (e.g., Price& Bradstock 2012), as well as those on other fires inAustralia (e.g., Collins et al. 2007; Murphy & Russell-Smith 2010; Bowman et al. 2014). The study by Price &Bradstock (2012) used the same fire severity dataset asin our study, but they applied it across broader land-scapes, multiple fire events, and longer time frames(extending several days following ignition of the fire). Inaddition to determining that weather was the primary in-fluence on fire severity, they found that the probabilityof canopy consuming fires in Ash forest was higher inrecently logged areas than in areas less recently loggedand in more recently burnt areas than those less recentlyburnt. Price & Bradstock (2012) used a logistic (binomial)regression to develop predictive models of the probabil-ity of occurrence of crown fire and understorey fire. Theyused a sample of grid points with 500-m separation acrossthe multiple fire impacted areas to match the ridge valleydistance across the fire impacted areas. They included dry,damp and wet forest in their analysis. Our study exam-ined stand level responses to the 2009 fire. At this scale,we were able to detect a nonlinear response, with standsyounger than 7 years and older than 40 years experienc-ing very little canopy consumption, but the highest im-pact being experienced at approximately 15 years old.
The study by Attiwill et al. (2013), which also analyzedthe fire severity impacts of the February 2009 fires, differsfrom our study and those of Price & Bradstock (2012).Attiwill et al. (2013) argue that there is no relationshipbetween fire severity and stand age across wet Eucalyp-
tus forest (including Mountain Ash). Differences betweenthe results of our investigation and those by Attiwill et al.(2013) likely occurred for several key reasons. First,Attiwill et al. (2013) did not differentiate between areasburnt under extreme fire weather conditions from thoseburnt under low to moderate weather conditions. Sec-ond, they did not differentiate areas that were burnt bythe fire itself from those areas treated with human in-duced back-burning. Third, they downplayed the signifi-cance of the extensive areas of young stands (>7 years)that had been established following clearfell logging andwhich burned at a high severity in the 2009 wildfires. Fi-nally, Attiwill et al. (2013) did not quantitatively analyzepatterns of fire severity across the landscape through for-mal analysis (see Bradstock & Price 2014).
Implications for forest management and forestbiodiversity conservation
The results of this investigation have some importantand often intimately interrelated implications for man-agement and conservation, particularly in those moistforest environments that are subject to intensive and ex-tensive logging operations and where high-severity firescan occur.
The elevated risks of high-severity fire in young standsregenerating after logging, uncovered in this study andin previous work (e.g., Thompson et al. 2007; Price &Bradstock 2012), suggests there is a need for forest man-agers to carefully consider the amount and spatial dis-tribution of forest age classes in wood production land-scapes. For example, there may be a need to scheduleharvesting operations to limit the total amount of youngforest in a given landscape. Forest planning may be re-quired to avoid wood production landscapes from becom-ing heavily dominated by extensive and contiguous ar-eas of, for example, 10–40 years forest that is prone tocanopy-consuming fire.
There may be other, broader implications of stand age-fire severity relationships in moist forests such as thoseassociated with the overall rotation time across wood pro-duction landscapes. For example, in the Mountain Ashforests of Victoria, many stands have been cut on a ro-tation as frequent as 48 years to maintain wood flows(Squire et al. 1991). Given the maintenance of thesestands as young forests, which if burned, are at riskof high-severity fire, there may be a need to reassess
366 Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc.
NND.001.00076.01_0035
C. Taylor et al. Nonlinear stand age effects on fire severity
Figure 10 Aerial image of subregion (R2), showing stands of fire impacted Mountain Ash forest across Deep Creek and Paradise Plains in 2013, 4 years
following the February 2009 fires. The areas where live trees are present fall within the modeled old growth sites. These forests were aged around 300
years at the time of the 2009 fire. The forests in the foreground consist of amixture of very young stands (<7 years), young stands (8–40 years) and stands
originating from the 1939 fires (70 years). These areas (8–40 and 70 years) are largely absent of live trees. (Note: image was taken in April 2013—4 years
after the fire).
rotation times to increase the interval between repeatedlogging events in the same stand.
A further key implication of our findings relates to theimpacts of young flammable stands on the ecological in-tegrity of adjacent older forests, which can be critical forthe persistence of disturbance-sensitive taxa in otherwiselogged landscapes (Gustafsson et al. 2012). Several studiesin a range of moist forest ecosystems have indicated therecan be strong spatial autocorrelation in patterns on firespread (Whelan 1995; McCarthy & Lindenmayer 1998).This makes it important to determine the influence of el-evated fire severity in logged and regenerated stands onadjacent undisturbed areas. In this context, a key researchtask will be to determine both the risks posed by younglogged forests for neighboring older stands and what sizeof old growth patches might be needed to reduce spatialcontagion in high-severity fire across otherwise loggedlandscapes.
The results of our work underscore the importance ofstands of old forest in ecosystems that are subject to high-severity fire. Indeed, although we observed widespreadmortality of Mountain Ash trees across all age classesover 7 years of age, there was a greater probability of treesurvival in older (300 years) stands than in the youngerage classes (Figure 10). Surviving large old trees are crit-ical postdisturbance biological legacies that play manykey ecological roles like storing large amounts of carbonand providing habitat for biodiversity (Lindenmayer et al.
2014b). They also can be critical for promoting the regen-eration of stands after disturbance (Bowman et al. 2014).
The conservation and management implications of ourfindings highlight the critical importance of quantify-ing relationships between fire and anthropogenic dis-turbance and, in turn, taking steps to ensure that risksof elevated increased fire severity are mitigated wher-ever possible. We argue that human disturbances, like
Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc. 367
NND.001.00076.01_0036
Nonlinear stand age effects on fire severity C. Taylor et al.
clearfell logging, can add further to the risk and should beminimized where possible. Such additional understand-ing will be critical in the wake of rapid climate change,where fires might become more widespread, frequent,and intense (Cary et al. 2012).
Acknowledgments
CT was supported by the Melbourne Sustainable SocietyInstitute at the University of Melbourne. MM was sup-ported by the Australian Research Council (ARC) Cen-tre of Excellence for Environmental Decisions, the Na-tional Environmental Research Program EnvironmentalDecisions Hub, and an ARC Future Fellowship. DBL wassupported by the ARC Centre of Excellence for Environ-mental Decisions, the National Environmental ResearchProgram Environmental Decisions Hub, and an ARC Lau-reate Fellowship. Comments by M. Pinard and severalanonymous referees greatly improved earlier versions ofthis manuscript.
Supporting Information
Additional Supporting Information may be found in theonline version of this article at the publisher’s web site:
Methods: Deriving the Forest Fire Danger Index, andvariables used in the analysis
Table S1: Fire severity classes and descriptionsTable S2: Input variables attributed to the 633
randomly selected sites that were used for theanalysis
Figure S1: Comparative Weather Conditions and For-est Fire Danger Index for the Kilmore Gap, Coldstream,and Eildon Fire Tower weather stations.
Figure S2: Subregion R1 (Mount Disappointment)showing fire severity classes within stands of MountainAsh forest.
Figure S3: Subregion R2 (Paradise Plains and DeepCreek) showing fire severity classes within stands ofMountain Ash forest.
References
Ashton, D.H. (1976). The development of even-aged stands of
Eucalyptus regnans F. Muell. in central Victoria. Aust. J. Bot.
24, 397-414.
Attiwill, P.M., Ryan, M.F., Burrows, N. et al. (2013). Timber
harvesting does not increase fire risk and severity in wet
eucalypt forests of southern Australia. Conserv. Lett. 7,
341-354.
Bachelet, D., Lenihan, J., Neilson, R., Drapek, R. & Kittel, T.
(2005). Simulating the response of natural ecosystems and
their fire regimes to climatic variability in Alaska. Can. J.
Forest Res. 35, 2244-2257.
Bowman, D.M.J.S., Balch, J.K., Artaxo, P. et al. (2009). Fire in
the earth system. Science 324, 483-486.
Bowman, D.M.J.S., Balch, J.K., Artaxo, P. et al. (2011) The
human dimension of fire regimes on Earth. J. Biogeogr. 38,
2223-2236.
Bowman, D.M., Murphy, B.P., Neyland, D.L.J., Williamson,
G.J. & Prior, L.D. (2014). Abrupt fire regime change may
cause landscape-wide loss of mature obligate seeder forests.
Global Change Biol. 20, 1008-1015.
Bradstock, R.A. & Price, O.F. (2014). Logging and fire in
Australian forests: errors by Attiwill et al. Conserv. Lett. 7,
419-420.
Bradstock, R.A., Gill, A.M. & Williams, R.J. (2012). Flammable
Australia: fire regimes, biodiversity and ecosystems in a changing
world. CSIRO Publishing, Melbourne.
Bradstock, R.A., Hammill, K.A., Collins L. & Price O. (2010).
Effects of weather, fuel and terrain on fire severity in
topographically diverse landscapes of south-eastern
Australia. Landscape Ecol. 25, 607-619.
Cary, G.J., Bradstock, R.A., Gill, A.M. & Williams, R.J. (2012).
Global change and fire regimes in Australia. Pages 149-170
in R.A. Bradstock, A.M. Gill & R.J. Williams, editors.
Flammable Australia: fire regimes, biodiversity and ecosystems in
a changing world. CSIRO Publishing, Melbourne.
Chib, S. & Greenberg, E. (1998). Analysis of multivariate
probit models. Biometrika 85, 347-361.
Cochrane, M.A. & Barber, C.P. (2009). Climate change,
human land use and future fires in the Amazon. Global
Change Biol., 15, 601-612.
Collins, B.M., Kelly, M., van Wagtendonk, J.W. & Stephens,
S.L. (2007). Spatial patterns of large natural fires in Sierra
Nevada wilderness areas. Landscape Ecol. 22, 545-557
Cruz, M.G., Sullivan, A.L., Gould, J.S. et al. (2012). Anatomy
of a catastrophic wildfire: the Black Saturday Kilmore East
fire in Victoria, Australia. Forest Ecol. Manag. 284, 269-285.
Dovers, S. (2003). Uncertainty, environment and fire:
research and policy issues. In G. Cary, D.B. Lindenmayer &
S. Dovers, editors. Australia burning: fire ecology, policy and
management issues. CSIRO Publishing, Melbourne.
DSE. (2005). Native Vegetation—Modelled 2005 Ecological
Vegetation Classes (with Bioregional Conservation Status)
(Shapefile). Department of Sustainability and
Environment, Melbourne.
DSE. (2007a). Victoria’s Statewide Forest Resource Inventory:
Central, Dandenong and Central Gippsland forest
management areas. Department of Sustainability and
Environment, Melbourne.
DSE. (2007b). OG100-Modelled old-growth forest boundaries
(Shapefile). Department of Sustainability and
Environment, Melbourne.
DSE. (2009a). Annual Report 2009. Department of
Sustainability and Environment, Melbourne.
368 Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc.
NND.001.00076.01_0037
C. Taylor et al. Nonlinear stand age effects on fire severity
DSE. (2009b). Remote Sensing Guideline for Assessing
Landscape Scale Fire Severity in Victoria’s Forest Estate,
Version 1.4. Department of Sustainability and
Environment, Melbourne (unpublished).
DSE. (2009c). Victorian Bushfires Severity Map 2009
(RASTER GRID). Department of Sustainability and
Environment, Melbourne.
DSE. (2011). Forest Explorer 5. http://nremap-sc.nre.vic.gov.
au/MapShare.v2/imf.jsp?site=forestexplorer (visited July
18, 2011). Department of Sustainability and Environment,
Melbourne.
ESRI. (2011). ArcGIS desktop: release 10. Environmental
Systems Research Institute, Redlands, CA.
Ferguson, I. & Cheney, P. (2011). Wildfires, not logging,
cause landscape traps. Aust. Forest. 74, 362-365.
Florence, R.G. (1994). The ecological basis of forest fire
management in New South Wales. In P.M. Attiwill, R.G.
Florence, W.E. Hurditch & W.J. Hurditch, editors, The
burning continent: forest ecosystems and fire management in
Australia. Institute of Public Affairs, Jolimont, Victoria.
Franklin, J.F., Spies, T.A., Pelt, R.V. et al. (2002). Disturbances
and structural development of natural forest ecosystems
with silvicultural implications, using Douglas-fir forests as
an example. Forest Ecol. Manag. 155, 399-423.
Gellie, N., Mattingley, G., Gibos, K., Wells, T. & Salkin, O.
(2013). Reconstructed Fire Spread and Dynamics of the
Black Saturday Bushfires. Unpublished Draft Report.
Department of Sustainability and Environment,
Melbourne.
Gibbons, P., van Bommel, L., Gill, A.M. et al. (2012). Land
management practices associated with house loss in
wildfires. PLOS One, e29212.
Gill, A.M. (2012). Bushfires and biodiversity in southern
Australian forests. Pages 235-252 in R.A. Bradstock, A.M.
Gill & R.J. Williams, editors. Flammable Australia: fire
regimes, biodiversity and ecosystems in a changing world. CSIRO
Publishing, Melbourne.
Griffiths, T. (2001). Forests of ash: an environmental history.
Cambridge University Press, Cambridge.
Gustafsson, L., Baker, S., Bauhus, J. et al. (2012). Retention
forestry to maintain multifunctional forests: a world
perspective. BioScience, 62, 633-645.
Hollis, G. (2004). Ecology and conservation biology of the
Baw Baw Frog Philoria frosti (Anura: Myobatrachidae).
Department of Zoology, University of Melbourne,
Melbourne.
Keeley, J.E. (2009). Fire intensity, fire severity and burn
severity: a brief review and suggested usage. Int. J. Wildland
Fire 18, 116-126.
Keith, H., Mackey, B.G. & Lindenmayer, D.B. (2009).
Re-evaluation of forest biomass carbon stocks and lessons
from the world’s most carbon-dense forests. Proc. Natl.
Acad. Sci. U.S.A. 106, 11635-11640.
Krawchuk, M.A. & Cumming, S.G. (2009). Disturbance
history affects lightning fire initiation in the mixedwood
boreal forest: observations and simulations. Forest Ecol.
Manag. 257, 1613-1622.
Leonard, S.W., Bennett, A.F. & Clarke, M.F. (2014).
Determinants of the occurrence of unburnt forest patches:
potential biotic refuges within a large, intense wildfire in
south-eastern Australia. Forest Ecol. Manag. 314, 85-93.
Lindenmayer, D.B., Banks, S.C., Laurance, W.F., Franklin,
J.F. & Likens, G.E. (2014b). Broad decline of populations of
large old trees. Conserv. Lett. 7, 72-73.
Lindenmayer, D.B., Blanchard, W., McBurney, L. et al.
(2013). Fire severity and landscape context effects on
arboreal marsupials. Biol. Conserv. 167 137-148.
Lindenmayer, D.B., Blanchard, W., McBurney, L. et al.
(2014a). Complex responses of birds to landscape-level fire
extent, fire severity and environmental drivers. Divers.
Distrib. 20, 467-477.
Lindenmayer, D.B., Burton, P.J. & Franklin, J.F. (2008).
Salvage logging and its ecological consequences. Island Press,
Washington DC.
Lindenmayer, D.B., Cunningham, R.B., Donnelly, C.F. &
Franklin, J.F. (2000a). Structural features of old-growth
Australian montane ash forests. Forest Ecol. Manag. 134,
189-204.
Lindenmayer, D.B., Hobbs, R.J., Likens, G.E., Krebs, C. &
Banks, S.C. (2011). Newly discovered landscape traps
produce regime shifts in wet forests. Proc. Natl. Acad. Sci.
U.S.A. 108, 15887-15891.
Lindenmayer, D.B., Hunter, M.L., Burton, P.J. & Gibbons, P.
(2009). Effects of logging on fire regimes in moist forests.
Conserv. Lett. 2 271-277.
Lindenmayer, D.B., Incoll, R.D., Cunningham, R.B. &
Donnelly, C.F. (1999). Attributes of logs on the floor of
Australian Mountain Ash (Eucalyptus regnans) forests of
different ages. Forest Ecol. Manag. 123, 195-203.
Lindenmayer, D.B., Mackey, B.G., Cunningham, R.B. et al.
(2000b). Factors affecting the presence of the cool
temperate rain forest tree myrtle beech (Nothofagus
cunninghamii) in southern Australia: integrating climatic,
terrain and disturbance predictors of distribution patterns.
J. Biogeogr. 27, 1001-1009.
Lutze, M.T., Campbell, R.G. & Fagg, P.C. (1999).
Development of silviculture in the native State forests of
Victoria. Aust. Forest. 62, 236-244.
McCarthy, M.A. & Lindenmayer D.B. (1998). Multi-aged
mountain ash forest, wildlife conservation and timber
harvesting. Forest Ecol. Manag. 104, 43-56.
McCarthy, M.A., Gill, A.M. & Bradstock, R.A. (2001).
Theoretical fire interval distributions. Int. J. Wildland Fire
10, 73-77.
McCarthy, M.A., Gill, A.M. & Lindenmayer, D.B. (1999). Fire
regimes in mountain ash forest: evidence from forest age
structure, extinction models and wildlife habitat. Forest
Ecol. Manag. 124, 193-203.
McCullagh, P. & Nelder, J.A. (1989). Generalised linear models,
2nd ed. Chapman and Hall, New York.
Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc. 369
NND.001.00076.01_0038
Nonlinear stand age effects on fire severity C. Taylor et al.
Mackey, B., Lindenmayer, D.B., Gill, A.M., McCarthy M.A. &
Lindesay J.A. (2002). Wildlife, fire and future climate: a forest
ecosystem analysis. CSIRO Publishing, Melbourne.
Murphy, D. P. & Russell-Smith, J. (2010). Fire severity in a
northern Australian savanna landscape: the importance of
time since previous fire. Int. J. Wildland Fire 19, 46-51.
Noble, I.R., Bary, G.A.V. & Gill, A.M. (1980). McArthur’s
fire-danger meters expressed as equations. Aust. J. Ecol. 5,
201-203.
Ough, K. (2001). Regeneration of wet forest flora a decade
after clearfelling or wildfire—is there a difference? Aust. J.
Bot. 49, 645-664.
Price, O.F. & Bradstock, R.A. (2010). The effect of fuel age on
the spread of fire in sclerophyll forest in the Sydney region
of Australia. Int. J. Wildland Fire 19, 35-45.
Price, O.F. & Bradstock, R.A. (2012). The efficacy of fuel
treatment in mitigating property loss during wildfires:
insights from analysis of the severity of the catastrophic
fires in 2009 in Victoria, Australia. J. Environ. Manag. 113,
146-157.
Raison, R.J. (1980). A review of the role of fire in nutrient
cycling in Australian native forests, and of methodology for
studying the fire-nutrient interaction. Aust. J. Ecol. 5, 15-21.
Smith, R.B. & Woodgate, P.W. (1985). Appraisal of fire
damage and inventory for timber salvage by remote
sensing in mountain ash forests in Victoria. Aust. Forest. 48,
252-263.
Spiegelhalter, D., Thomas, A., Best, N. & Lunn, D. (2012).
OpenBUGS User Manual. Version 3.2.2.
Squire, R.O., Campbell, G.G., Wareing, K.J., & Featherston,
G.R. (1991). The mountain ash forests of Victoria: Ecology,
silviculture and management for wood production. Pages
38-57 in F.H McKinnell, E.R. Hopkins & J.E.D. Fox,
editors. Forest management in Australia. Surrey Beatty &
Sons, Chipping Norton.
Sullivan, A.L., McCaw, W.L., Cruz, M.G., Matthews, S. &
Ellis, P.F. (2012). Fuel, fire weather and fire behaviour.
Pages 51-79 in R.S. Bradstock, M.A. Gill & R.J. Williams,
editors. Flammable Australia: fire regimes, biodiversity and
ecosystems in a changing world. CSIRO Publishing,
Melbourne.
Thompson, J.R., Spies, T.A. & Ganio, L.M. (2007). Reburn
severity in managed and unmanaged vegetation in a large
wildfire. Proc. Natl. Acad. Sci. U.S.A. 104, 10743-10748.
Thompson, J.R., Spies, T.A. & Olsen, K.A. (2011). Canopy
damage to conifer plantations within a large
mixed-severity wildfire varies with stand age. Forest Ecol.
Manage. 262, 355-360.
Viggers, J.I., Weaver, H.J. & Lindenmayer, D.B. (2013).
Melbourne’s water catchments. Perspectives on a world class water
supply. CSIRO Publishing, Melbourne.
Vivian, L.M., Cary, G.J., Bradstock, R.A. & Gill, A. (2008).
Influence of fire severity on the regeneration recruitment
and distribution of eucalypts in the Cotter River
Catchment, Australian Capital Territory. Austral Ecol. 33,
55-67.
Whelan, R.J. (1995). The ecology of fire. Cambridge University
Press, Cambridge, England.
Wood, S., Bowman, D., Prior, L., Lindenmayer, D.B.,
Wardlaw, T. & Robinson, R. (2014). Tall eucalypt forests.
pages 519-569 in D.B. Lindenmayer, E. Burns, N.Y.
Thurgate & A. Lowe, editors. Biodiversity and environmental
change: monitoring, challenges and direction. CSIRO
Publishing, Melbourne.
370 Conservation Letters, July/August 2014, 7(4), 355–370 Copyright and Photocopying: C©2014 Wiley Periodicals, Inc.
NND.001.00076.01_0039
commentPUBLISHED: 24 JANUARY 2017 | VOLUME: 1 | ARTICLE NUMBER: 0031
Please do not disturb ecosystems furtherDavid Lindenmayer, Simon Thorn and Sam Banks
Clearing up after natural disturbances may not always be beneficial for the environment. We argue that a radical change is needed in the way ecosystems are managed; one that acknowledges the important role of disturbance dynamics.
NND.001.00076.01_0040
© 2017
Macmillan
Publishers
Limited,
part
of
Springer
Nature.
All
rights
reserved.
NND.001.00076.01_0041
© 2017
Macmillan
Publishers
Limited,
part
of
Springer
Nature.
All
rights
reserved.
NND.001.00076.01_0042
Recommended