Decision Point #22 - 1

Decision PointSmart science for wise decisionsIssue 22 / September 2008

Connecting conservation policy makers, researchers and practioners

Other stories Maths, modelling and decision making 2Managing emergent wildlife disease 4Ecofarm and Ecoforest - 2 decision games 7Modelling in the face of severe uncertainty 8Australia’s land surface & climate change 10

The passing of the paddock treeThe future of native tree cover in agriculture landscapesPage 11

Saving the travelling stock routes The many values of the Long Paddock - p3

Decision PointDecision Point is a monthly magazine presenting news, views and ideas on environmental decision making, biodiversity, conservation planning and monitoring. It is produced by AEDA – the Applied Environmental Decision Analysis CERF Hub. For more info on AEDA, visit our website at www.aeda.edu.au or see the back cover.

Green carbon and nature forests A safer carbon bank - p6

When to translocate climate victims A new decision framework - p12

Decision Point #22 - 2

Models, problems and algorithms What are they and why should I care?

The Dpoint editorial

By Hugh Possingham (Director, AEDA)

One of the most frequent questions I am asked after seminars about using decision theory for solving conservation problems is: “But this is all modeling, and we know that ecological models are invariably poor, so isn’t all this decision theory rubbish!”. The short answer is that much of the maths in decision theory is not modeling, but problem formulation and problem solving algorithms. Is this just Hugh’s semantic trick intended to baffle the less numerate? No it’s not, but let me elaborate with an example.

Conservation planning is probably the oldest and most widely used application of decision theory to biodiversity conservation. A typical conservation planning problem can be stated in words – conserve 30% of every kind of habitat in a region for the minimum possible ‘cost’ (which could simply be management and acquisition cost, or many other things). I think two CSIRO researchers, Cocks and Baird (1989), were the first people to turn this problem into mathematics. This is NOT modelling, it’s problem formulation.

It is more akin to translation than modeling. The problem, stated initially in English, could have been translated into Spanish, but instead it was translated into mathematics, which after all, is just another language. The benefit of translating the problem into mathematics is that we have a huge raft of algorithms for finding optimal or near optimal solutions to mathematical problems. Translating the problem into Spanish is useful to people fluent in Spanish, but doesn’t get us any closer to solving the problem.

The process of problem formulation is potentially 100% accurate. For example the problem above can be translated into mathematics exactly, with no error. It does rely on a few things where there could certainly be a discussion though. For example, what do we mean by ‘every kind of habitat’ and who defines them? There may be some statistical modeling, like cluster analysis, behind the definition of habitat. Two vegetation ecologists are unlikely to define exactly the same suit of habitats in a large area. The problem needs some data on how much habitat is in any planning unit and the cost of those planning units – this may also involve some modeling and errors will appear.

Problem formulation is the translation of our aspirations into mathematics. The translation can be perfect, although being sure we have been clear about our aspirations is critical. There is a great deal to be argued about in the field of problem formulation. For example, is it the right problem? But this is an ecological problem, not a mathematical problem. The discussion later in Decision Point about ‘info-gap’ theory (see page 8) is a lot about problem formulation, not modeling.

Once we have a well-posed mathematical problem, the next task is to solve it. This involves a lot of mathematics but this is generally not very interesting or contentious. Advanced mathematical algorithms for solving complex problems have been worked on by people a lot smarter than me for a long time.

Mathematicians have found many excellent algorithms, and many will deliver exact solutions to problems. Most will deliver solutions so close to the best solution we shouldn’t worry about them (compared to the substantial potential errors in models that underpin data and disputes about the problem). People are successfully using algorithms every day, for example every time they use a hand calculator or computer to multiply two numbers together. Marxan is the most widely used tool for finding good solutions to some versions of the conservation planning problem – unless the user makes an error, the solutions will all be pretty good.

So what should we be worried about most in solving nature conservation problems? The biggest problem for government and non-government organisations trying to solve nature conservation problems is that they don’t formulate the problem properly, if at all!

Consider the hypothetical problem: how should we allocate stewardship and restoration funds in the Burdekin catchment to maximize biodiversity benefit. This is not yet a well-posed problem. There is no clear definition or quantification of the objective, ‘biodiversity benefit’; are we trying to conserve corals, or water quality in the river, or birds, or habitats? Because of our inability to clearly state the problem, agencies skip the step of mathematically formulating the problem and go straight to an algorithm for solving the problem. This algorithm is invariably an ad hoc scoring system, and ultimately we obtain a poor solution to the problem because it hasn’t been properly formulated.

There are many advantages to mathematical problem formulation – transparency, repeatability, clear quantifiable objectives that can be audited, and the ability to tap into 2,500 years of mathematics. Problem formulation is not modelling, but models may be hidden inside the problem. These may be process-based or statistical models – models that tell us how water quality responds to reduced grazing, or how birds species respond to changes in habitat. There will be uncertainty and risk in all these models, but even that can be happily handled using decision theory tools, as long as we get the formulation right and choose our algorithm wisely. Knowing whether we are talking about problem formulation, modeling or algorithms is a critical conceptual leap for making smart conservation decisions.

AEDA researchers are formulating, and have formulated and solved, a huge variety of natural resource management problems over recent years – a tiny fraction include: how can we maximise the number of threatened species recovered for a fixed budget, how can we maximise the chance a weed is eradicated, how long should we unsuccessfully search for a threatened species before we stop and give up, what are optimal environmental flows, what actions will best maximise ecosystem services, what kind of tree planting maximises the area revegetated, what fire regimes minimise the chance a species goes extinct, or a habitat type is lost, where should revegetation occur to maximise the number of species save in a production landscape?

Every natural resource management conundrum has a potential mathematical formulation and hence a good solution. The challenge is translating policy and management aspirations into maths.

Reference

Cocks KD and Baird IA (1989), Using mathematical programming to address the multiple reserve selection problem – an example from the Eyre Peninsula, South Australia. Biological Conservation, 49(2): pp113-130

“The biggest problem for government and non-government organisations trying to solve nature conservation

problems is that they don’t formulate the problem properly, if at all!”

Decision Point #22 - 3

continued on page 4

Saving a biodiversity jewel Understanding the value of Long Paddock

At the end of August more than 450 ecologists and wildlife scientists called on the premiers of NSW and Queensland to protect the 3.2 million hectare travelling stock route (TSR) network. Why all the fuss? (After all, it’s not often you get a roll call of Australia’s best and brightest environmental minds standing up as one and calling for the urgent protection of a paddock.)

The stock routes, also known as the Long Paddock, are an irreplaceable biodiversity treasure. They’re a legacy of our grazing history, and one of the few land assets we have that enhance the landscape’s capacity to cope with climate change. They provide refuge for endangered species and in many cases are the best remaining examples of native vegetation in a highly cleared landscape.

“Administrative changes imminent in both NSW and Queensland threaten the future health, functionality and integrity of this vast, public network of routes and reserves that stretches across eastern Australia,” says AEDA’s Director, Professor Hugh Possingham. Hugh, along with Professor Henry Nix from ANU, headed up the scientists demanding better treatment for the TSRs.

“In Queensland it is likely that large areas will be subject to long-term leases allowing permanent grazing,” says Hugh. “This is incompatible with the maintenance of biodiversity values on these precious lands.

“In NSW the threat is even greater as changes to the administering act, likely to occur in September, will make it probable that much of the network may be handed back to the Department of Lands and then leased for the long term or sold.

“Research shows the network supports some of the last strongholds of Australia’s most threatened native animals and plants on public land, and it provides some of the only connections of nature in our extensively cleared and modified landscapes, thus facilitating the movement of animals and plants across the landscape.

“Being fairly straight lines, the stock routes tend to sample the vegetation types fairly evenly, often better than our reserve system. Furthermore, the network of routes and reserves provides important connectivity over a vast area.

Imagine if you tried to design this“Few people recognise the potential of road and rail reserves and stock routes as wildlife corridors,” says Dr Bob Sutherst, an ecologist at the Spatial Ecology Lab at the University of Queensland who has devoted himself to protecting the Long Paddock. “However, considered together they offer us a precious biodiversity resource.”

He points out that in NSW the TSRs and road reserves cover about 5% of the state, nearly as much as the 7% dedicated to National Parks. In Queensland stock routes cover 71,650 km or 2,601,510 ha, plus 762 stock reserves of 395,879 ha and 742 water facilities.

“When you look at the maps of the stock routes in NSW and Queensland you quickly appreciate that they offer a magical network in terms of connectivity,” says Bob. “Imagine if you got an area where you’re trying to generate some mechanism to help species move through the landscape and you suddenly discovered the extent of the stock routes.”

Whereas reserves are discontinuous and tend to be targeted to a specific landform or vegetation, stock routes and roads are continuous and incorporate a variety of local landform and vegetation types, and watering points. They also often contain fertile soils, remnant vegetation and much greater biodiversity than adjoining private, grazed land.

In the context of climate change, road and rail reserves and stock routes form an extraordinarily fortuitous, extensive network of corridors, which, with a contribution from the road network itself, could facilitate the movement of species in response to shifting climatic zones. This network is in public hands and its potential in protecting biodiversity under climate change makes it imperative that local, state and federal governments recognise and manage them for their biodiversity values.

Further reading:

Sutherst, B, Szabo, J, and Cleland, E (2007) The Stock Routes and road network—strengthening the biodiversity links, The State of Australia’s Birds, 2007.

Available free at:

http://www.birdsaustralia.com.au/soab/state-of-australias-birds.html

“Don’t it always seem to go that you don’t know what you’ve got till it’s gone” Excerpt from Big Yellow Taxi, (by Joni Mitchell)

Decision Point #22 - 4

Managing an emergent disease The dilemma of short-term conservation vs longer term learningBy Eve McDonald-Madden and William Probert (UQ, Brisbane Node, AEDA)

Devil Facial Tumour Disease (DFTD) poses a new threat to the Tasmanian devil (Sarcophilus harrisii). Since its appearance there has been a rapid decline in Devil numbers with many populations experiencing declines of up to 95%. This is a massive challenge with the disease significantly increasing the 25 year probability of extinction of the Tasmanian devil in the wild.

In light of its dramatic impact and the uniqueness of the devil, DFTD demands an urgent response from managers. The novelty of DFTD has led to multiple hypotheses regarding how the disease behaves; for example, how it is transmitted within the population, how soon do symptoms show in an infected individual, and how much impact it has on reproductive capability of a population. Because of this uncertainty there is not only a long-term objective to maintain devil populations but also a pressing need to understand how the disease behaves so an appropriate course of management can be implemented. How might this be best achieved?

One way forward is to deliberately implement actions that enable us to distinguish between our different theories (ie, set up a framework for learning). For example, if we did not implement any direct management action and the population increased we may discern that the disease actually had no population level impact on the Tasmanian devil. Or, if there was a negative response, we could say that the disease does have an impact. Of course, this is a very simple example, and one to which we already have an answer (not managing directly has a large negative response). But how can we learn, for example, the length of time it takes for a tumour to appear on a Tasmanian devil infected with DFTD (the latency of the disease)?

Conservation actions that can be implemented within a diseased population (in situ actions) are currently limited to the removal of individuals from a population. Removal of individuals is expected to reduce the prevalence of the disease in the population. By suppressing the prevalence of the disease, management aims to give the devil population a better chance at recovery.

Each removal strategy can affect the growth of the devil population in a different way and it is the difference in the response of the population that can enable us to learn which model of disease symptom latency is most likely to be true. But which action should we implement and when should we implement them is the question we must answer. Should we invest in learning early in the management time-frame by implementing several different

“Should we invest in learning early in the management time-frame by implementing several different

management actions and thus hone our belief in how the system functions? Or, should we instead learn nothing and implement one management action?”

“Since the 19th century the routes have provided an important resource to graziers. More recently it has been recognised that in extensively cleared landscapes, the network contains some of the best remaining examples of threatened vegetation, such as the grassy box woodlands.”

“The network provides critical resources for migrating birds across eastern Australia, and the value of the network for movement of wildlife and genetic material of native animals and plants is even more important now that our natural systems are adapting to climate change,” says Henry Nix.

“The extent of support from scientists around Australia indicates strong scientific consensus about the importance of protecting and maintaining the outstanding ecological and cultural values of the Travelling Stock Route network,” says Hugh Possingham.

“Retention of original vegetation such as that found in the stock routes and reserves is by far the most efficient way of enabling landscape connectivity and restoration. These areas provide a backbone for revegetation and a major source of native seed resources.

“The best available science shows protecting and actively managing the entire network of travelling stock routes and reserves is one of the most efficient steps to ensure our Australian landscapes, our unique wildlife, our cultural heritage, and our travelling stock tradition survive.”

More info: h.possingham@uq.edu.au

Continued from page 3.

“They’re a legacy of our grazing history, and one of the few land assets we have that enhance the

landscape’s capacity to cope with climate change. ”

Stock routes like this one near Molong in NSW, sample all vegetation and landscape types. There’s nothing quite like them and now they’re under threat.

Imag

e by

Cat

her

ine

Sta

ple

ton

Decision Point #22 - 5

management actions and thus hone our belief in how the system functions? Or, should we instead learn nothing, implement one management action without increasing our knowledge and thus take a chance on our belief? Or should we actively learn in some populations and take a chance on one action in the other populations?

We have solved this problem for a simplified model of learning about the latency of the DFTD whilst managing. We show that a management strategy that explicitly incorporates learning does not mean that we implement an action a manager might implement without considering learning. Indeed when we need to learn, we may have to take risks by implementing actions that may not maximise the recovery of the devils in the short term. The key here is that these riskier actions can significantly increase our ability to distinguish between our notions of disease latency - they enable us to learn.

The devil’s dilemmaDevil Facial Tumour Disease appears to be a new disease that is restricted to devils. What were probably some of the first cases were photographed in north-east Tasmania in 1996. As at May 2008, DFTD had been confirmed at 63 separate sites covering more than 60% of the State.

DFTD is a contagious tumour that is spread between individuals, most probably through biting. The foreign cells of the tumour aren’t rejected by the animal’s immune system because of a lack of genetic diversity among Tasmanian devils.

Once the cancer becomes visible, animals usually die within a few months. Devils appear to succumb between two and three years of age, although some juveniles as young as one have also become infected. This is resulting in very young age-structured populations in which females have only one breeding event (usually they have three).

Populations in which the Tasmanian devil disease has been observed for several years have declined by up to 95%, with no evidence to date of either a cessation of decline or a diminution in the prevalence of the disease.

More info: http://www.dpiw.tas.gov.au/inter.nsf/WebPages/JCOK-65X2Y6?open

The safest option is to remove all diseased animal (which reassuringly is the current strategy employed in Tasmania to manage the disease). Which actions enable learning and when and where they should be implemented depends on a number of factors; the response of the population given an action is implemented, the time remaining for learning to occur and our confidence in each model of disease latency.

The most significant result arising from our analysis has been that implementing no direct management action, or doing nothing to manage a threatened species, can be the best action to enable learning. That is, we can discover the latency of the disease most efficiently by not removing any animals, monitoring the population response and then implement the management action that is appropriate for the Tasmanian devil.

Of course this is a controversial result and one that needs to be taken in light of the assumptions and simplifications employed in the analysis of the system (such as the ability to detect latency and population growth when monitoring, the reality of only three discrete theories regarding latency in this disease, and the possibility of losing the species during the learning process when we consider population growth not species persistence as our objective).

Our conclusion (that doing nothing is not really doing nothing when we want to learn), is one that may make us uneasy as conservationists. However, it’s a conclusion that should be considered when contemplating an ‘active’ adaptive approach to managing biodiversity.

It is possible to build more complex models for specific management cases that deal with more intricacies of the system, a direction we plan to go in to better aid management of the Tasmanian devil, but what we have strived for in this work in the first instance is to highlight the role of learning in managing threatened systems in which we have limited knowledge. A fact true in almost all threatened species management questions worldwide.

More info: Eve McDonald-Madden e.mcdonaldmadden@uq.edu.au

(For a good background article on active and passive adaptive management see AEDA information sheets 3.2 and 3.3 at http://www.aeda.edu.au/aeda-research-themes).

Note: this work was initiated at an AEDA workshop on optimal monitoring held in Tasmania in August last year.

Distribution of DFTD in Tasmania in February 2008 (from the Save the Tasmanian Devil site http://www.tassiedevil.com.au/index.html

DFTD is extremely unusual as it is only one of three recorded cancers that can spread like a contagious disease. The cancer is passed from devil to devil through biting.

Decision Point #22 - 6

What’s the role played by natural forests in storing carbon? It seems like a big and important question in greenhouse challenged world but it was only posed last year at an international level at the 2007 UN Climate Change Conference in Bali. In response, scientists from ANU (including AEDA’s David Lindenmayer) conducted a series of investigations into the carbon stocks of intact natural forests over large geographical areas, inclusive of environmental factors operating at landscape and regional scales. Their analysis (a report of which is now available free from ANU E Press*) has revealed that Australia’s natural forests have far larger carbon stocks than has been previously recognised. What’s more this carbon bank is self running and safer than industrial forests.

The report, which draws on decades of research into soil and wood samples as well as new field work, examined carbon storage at 240 sites across southern NSW, Tasmania and Victoria. It found that Australia’s 14.5 million hectares of undisturbed eucalypt forest holds 9.3 billion tonnes of carbon in its wood and soil, offsetting about 460 million tonnes of carbon emissions each year for the next century.

Figures from the Intergovernmental Panel on Climate Change showed the same forests as capable of only storing 3.1 billion tonnes. Even the Federal Government’s accounting system has underestimated the carbon storage, because it’s designed to measure biomass growth in reafforestation and plantation forests, rather than dense bushland that has never been disturbed.

The researchers acknowledged that young plantations do absorb carbon quickly as they grow, but point out that this does not compensate for the big carbon losses when established forests are cut down for the first time. Not only do natural forests store more carbon, but because they remained untouched they stored the carbon for longer than plantation forests which were cut down on a rotation basis.

Natural forests are more resilient to climate change and disturbances than plantations because of their genetic, taxonomic and functional biodiversity. This resilience includes regeneration after fire, resistance to and recovery from pests and diseases, and adaptation to changes in radiation, temperature and water availability (including those resulting from global climate change). While the genetic and taxonomic composition of forest ecosystems changes over time, natural forests will continue to take up and store carbon as long as there is adequate water and solar radiation for photosynthesis.

The green carbon in natural forests is stored in a more reliable stock than that in industrialised forests, especially over ecological time scales. Carbon stored in industrialised forests has a greater susceptibility to loss than that stored in natural forests. Industrialised forests, particularly plantations, have reduced genetic diversity and structural complexity, and therefore reduced resilience to pests, diseases and changing climatic conditions.

The carbon stock of forests subject to commercial logging, and of monoculture plantations in particular, will always be significantly less on average (~40 to 60 per cent depending on the intensity of land use and forest type) than the carbon stock of natural, undisturbed forests. The

rate of carbon fixation by young regenerating stands is high, but this does not compensate for the smaller carbon pools in the younger-aged stands of industrialised forests compared with those of natural forests. Carbon accounts for industrialised forests must include the carbon emissions associated with land use and associated management, transportation and processing activities.

Reference

Mackey BG, Keith H, Berry SL and Lindenmayer DB (2008), Green Carbon, The role of natural forests in carbon storage, Part 1. A green carbon account of Australia’s south-eastern Eucalypt forests, and policy implications, ANU E Press.

*The report on green carbon can be downloaded from ANU ePress at http://epress.anu.edu.au/green_carbon_citation.html

A bigger and safer carbon bank Green carbon and natural forests

Show us the colour of your carbonThe report says it’s important to consider the different sources of carbon when considering the role of natural forests in the global carbon cycle. They elected to colour code carbon along the following lines

Grey carbon: is the carbon stored in fossil fuel (coal, oil and gas deposits in the lithosphere).

Green carbon: is the carbon stored in the biosphere. They called it ‘green’ because carbon is taken up from the atmosphere by plants through the process of photosynthesis (involving the green pigment chlorophyll).

Brown carbon: is the carbon stored in industrialised forests. These are forests that are logged commercially for their wood, which is used as a source of raw material for industrial manufacturing processes. Industrialised forests constitute a stock of organic carbon and are therefore part of the biosphere; however, the report considered this carbon to be ‘brown’ in colour rather than ‘green’ in order to stress the fact that industrialised forests are a ‘mix’ of green and grey carbon. Fossil fuel is expended and therefore grey carbon emitted in managing these forestry operations and from the associated industrial processes.

Blue carbon: refers to the inorganic carbon stored in the atmosphere (carbon dioxide, CO2) and oceans (carbonate, CO32-).

Currently, international rules are blind to the colour of carbon so that the green carbon in natural forests is not recognised, resulting in perverse outcomes including ongoing deforestation and forest degradation, and the conversion of extensive areas of land to industrial plantations.

“The green carbon in natural forests is stored in a more reliable stock than that in industrialised forests,

especially over ecological time scales.”

Decision Point #22 - 7

“On the EcoFarm you’re the manager deciding how you’ll manage crop and grazing paddocks,

woodland remnants, dams and creeks. Each year your property is reviewed in terms of biodiversity and profitability,

and you track your performance over five years.”

Last year AEDA’s Canberra Node leader David Lindenmayer launched his most personal book on the state of Australia’s environment. Titled On borrowed time: Australia’s environmental crisis and what we must do about it, David described the publication as his ‘grumpy old man’s book’ written as a call to arms to his fellow Australians that the time to act on the declining state of Australia’s natural heritage is now. The book was well received and has led to the development of a complementary educational website aimed at engaging young Australians in the critical issue of caring for Australia’s precious biodiversity.

The website was created by CSIRO Publishing with the generous support of the Purves Environmental Fund. It consists of four inquiry-based teaching and learning units and two decision-making interactives which allow students to reflect, consider and make decisions relating to Australia’s biodiversity.

The four learning units focus on

Adaptation• – How does a species evolve and adapt to its environment?

Fire• – What is the role of fire in Australia’s ecosystems – Is it a friend or a foe?

Forest• – Is the biodiversity greater in an old growth forest compared with a new forest?

Farming• – Can farms protect and enhance biodiversity while increasing their profitability?

Each learning unit presents a wide range of information and exercises in the areas of science, maths and English, as well as providing a rich list of learning resources and some stunning imagery. Students even get to work with real ecological data to explore

On borrowed time – the website Make decisions while you manage a virtual farm or forest

concepts and themes. For example, under the theme of ‘fire’ students are provided with results of studies undertaken at Jervis Bay that examine how native animals like the ringtail possum and brown antechinus coped with a wildfire that struck the area in 2003. Or under the ‘adaptation’ theme students are asked to do a statistical analysis on different populations of black snakes and work out possible impacts of climate change on population sizes.

To consolidate some of this learning, the website also challenges visitors to play two decision-making interactives that explore farming and forestry and their interaction with biodiversity. On the EcoFarm you’re the manager deciding how you’ll manage crop and grazing paddocks, woodland remnants, dams and creeks. Each year your property is reviewed in terms of biodiversity and profitability, and you track your performance over five years.

Then there’s the EcoForest. You take on the role of the forest manager and see if you can protect the habitats of five important species while sustainably logging the forest to keep jobs in the local town. You have to choose between four different harvesting options and track what happens to the biodiversity and timber productivity of the forest over three working cycles (each cycle being 50 years old).

And if all that wasn’t enough, the website also offers schools a free CD-ROM of all the material on the website.

“The hope is that this material will help students take action towards the conservation of biodiversity on a personal, school and local community level,” says David Lindenmayer. “This teaching resource will teach you something about how Australia’s environments work, why biodiversity is important, and how we need to manage forests, national parks and farmlands in an ecologically sustainable way for the future of this continent and the people who live here.”

Check out the site for yourself at: http://www.publish.csiro.au/onborrowedtime/sections/The website also provides full details on how you can obtain the book – On borrowed time.

Fire is one of four learning themes explored in many different ways on the On borrowed time website.

Decision Point #22 - 8

From Shakespeare to Wald: Modeling worst-case analysis in the face of severe uncertainty by Moshe Sniedovich

Severe uncertainty is an ever-present phenomenon, ever lurking in the background. Yet, for all its dreaded implications, we seem to go about our daily affairs without giving it much thought. The experts, on the other hand, devote a great deal of thought to this topic, so much so that severe uncertainty has long been a central subject of study in the area of classical decision theory and its modern offspring.

Classical decision theory offers two basic conceptual paradigms – and many variations thereof – to tackle severe uncertainty. Needless to say, their mathematical formulations yield austerely simple models. The simplicity is due to the fact that conditions of severe uncertainty entail a dearth of information, data, knowledge and so on, which in turn entail meagre material to work with in the formulation of the respective mathematical models.

These two basic modelling paradigms, often referred to as principles, axioms, rules, are known as: • Laplace’s (1749-1827) Principle of Insufficient Reason • Wald’s (1902 - 1950) Maximin Principle

If you follow Laplace’s counsel, you would quantify severe uncertainty probabilistically, pretending that the uncertain events under consideration are equally likely. Once you adopt this stance you will find yourself in the realm of decision-making under risk, where statistics and probability theory reign supreme.

On the other hand, if you follow Wald’s counsel, you would pretend that omnipresent Nature is your adversary and that you should therefore expect the worst. In other words, Wald’s prescription for severe uncertainty dictates solving a problem subject to these conditions via a worst-case analysis.

Wald’s formulation of this idea was inspired by von Neumann’s (1903-1957) work on Maximin problems in the context of two-person games. With great insight Wald cast Nature as the player pit against the decision maker. In this casting, Nature, as the second antagonistic player, represents uncertainty, or more accurately the analyst’s attitude towards uncertainty.

But the origins of this basic idea actually go further back in time:

The gods to-day stand friendly, that we may, Lovers of peace, lead on our days to age! But, since the affairs of men rests still incertain, Let’s reason with the worst that may befall. William Shakespeare (1564-1616) Julius Caesar, Act 5, Scene 1

In a language less poetic, Wald’s Maximin Principle states that you should: Rank alternatives by their worst possible outcomes: adopt the alternative the worst outcome of which is at least as good as the worst outcome of the others.

The austere simplicity characterising the mathematical formulation of this recipe is a mixed blessing. On the one hand, it has the merit of doing away with the uncertainty altogether. This of course is due to Nature’s assumed consistent antagonism vis-a-vis the decision-maker, which makes it completely predictable.

On the other, the price tag attached to this convenience is significant: this pessimistic (conservative) stance with regard to uncertainty can result in huge over-protection costs.

So, over the years several adaptations of Wald’s original model have been proposed. Their objective has always been to mitigate Maximin’s conservatism and to bring it more in line with ‘real’ decision-making in practice.

It is also important to note that these principles do not necessarily ensure a smooth sailing in the management and treatment of severe uncertainty. Indeed, their application is often fraught with difficulties so that decision-making under severe uncertainty is as ever more art than science (Sniedovich 2007).

The inevitable question is then: how is it that with all its obvious, and no so obvious, shortcomings – some would say faults – Wald’s Maximin Principle remains one of the modeling pillars of decision-making under severe uncertainty? The answer to this intriguing question is simple: it is the familiar ten-letter word: Robustness.

Wald’s Maximin Principle outlines a recipe for obtaining the ultimate robustness in the face of severe uncertainty. For this reason it is the principal paradigm used in robust decision-making in general and robust optimisation in particular (Sniedovich 2007).

The recent resurgence of these fields is due to technological (computing power) and methodological (models and algorithms) advances that enable solving large-scale mathematical models employing this principle. That said, the question is then whether robust decision-making methods á la Maximin do/should/can play a role in environmental decision analysis in general and in the analysis of environmental decision problems relevant to Australia in particular?

This is a very relevant and timely question that is an ongoing topic of research at the Department of Mathematics and Statistics, University of Melbourne. And, as it turns out, it is also very relevant to a decision theory that is currently popular in environmental decision analysis in Australia, namely Info-Gap decision theory (Ben-Haim 2006). In brief, the centre piece of Info-Gap decision theory is its robustness model. This model specifies, for each decision, the largest region of uncertainty over which a prescribed performance requirement is satisfied at each point in this region. Decisions are then ranked according to their robustness.

Info-Gap decision theory is proclaimed in the literature as novel, revolutionary, and radically different from all current theories of decision under uncertainty.

“the conservation biology and applied ecology literatures are spotted with assertions that Info-Gap’s robustness model addresses the following question: how wrong

can I be, yet get an acceptable level of performance? As indicated, this

question is neither addressed by Info-Gap decision theory nor can it be

answered by it.”

Decision Point #22 - 9

Specifically, it is claimed that Info-Gap’s robustness model is not a Maximin model. Yet, with the aid of standard mathematical modeling tools it can be easily shown that Info-Gap’s robustness model is in fact a simple Maximin model (Sniedovich 2007).

However, the trouble is that Info-Gap’s robustness model is, by definition, inherently local in nature. That is, the robustness analysis that it prescribes consists of a worst-case (Maximin) analysis in the immediate neighborhood of an estimate of the parameter of interest. For this reason, Info-Gap’s robustness, by definition, does not represent robustness over the complete region of uncertainty under consideration.

More than that, under conditions of severe uncertainty, the estimate in question is a wild guess, a poor indication of the true value of the parameter of interest, and is likely to be substantially wrong. Thus, since the results based on this estimate can be only as good as the estimate, they are also wild guesses . . . and are likely to be substantially wrong.

So, not with standing Info-Gap being hailed as a decision theory that is designed specifically for severe uncertainty, one can easily prove that Info-Gap’s robustness model is in fact invariant to the severity of the uncertainty that it is supposed to tackle (Sniedovich, 2007).

In particular, under ‘normal’ conditions, the results generated by Info-Gap’s robustness model remain unaffected by the severity of the uncertainty, as measured by the ‘size’ of the region of uncertainty. That is, one can double, triple, quadruple, or for that matter increase indefinitely, the size of the complete region of uncertainty without this having the slightest impact whatsoever on the results generated by the model.

To see what kind of trouble one might fall into by ignoring the local nature of Info-Gap’s robustness model, consider this: the conservation biology and applied ecology literatures are spotted with assertions that Info-Gap’s robustness model addresses the following question: how wrong can I be, yet get an acceptable level of performance?

As indicated above, this question is neither addressed by Info-Gap decision theory nor can it be answered by it. And this for the simple reason that conditions of severe uncertainty imply that it is impossible to know how ‘wrong’ we are or can be because we do not know the true value of the parameter of interest, and in any case Info-Gap’s robustness analysis is local.

Rather, Info-Gap’s robustness model addresses the following, completely different, much easier, and – under severe uncertainty – not so interesting question: How much can I deviate from a given estimate so that the performance requirement is satisfied throughout the region of uncertainty (around the estimate) which is contained within this deviation?

In other words, Info-Gap’s robustness model simply guarantees that the performance constraint will be satisfied if the true value is in the ‘safe’ sub-region of uncertainty determined by the robustness analysis. But the point is that there is no guarantee that the true value is in this “safe” sub-region. In fact, under severe uncertainty this is very unlikely, namely this ‘if’ is a big, indeed very big, ‘if’.

In a nutshell, robustness á la Info-Gap’s Maximin model is defined in total disregard for the severity of the uncertainty under consideration.

So what should environmental analysts do to avoid

Hunting for treasure

• The island represents the region of uncertainty under consideration (the region where the treasure is located).

• The tiny black dot represents a wild guess of the parameter of interest (location of the treasure).

• The large white circle represents the region of uncertainty affecting Info-Gap’s robustness analysis.

• The small white square represents the true (unknown) value of the parameter of interest.

Clearly then, under severe uncertainty Info-Gap may conduct its robustness analysis in the vicinity of Brisbane, whereas for all we know the true location of the treasure may be somewhere in the middle of the Simpson desert or perhaps in down town Melbourne. Perhaps.

This is a vivid illustration of the importance of mathematical modeling in decision-making and the validity of the dictum: modeling is more art than science (Sniedovich 2007).

such methodological/conceptual pitfalls when modeling decision-making problems that are subject to severe uncertainty? The short answer is: either one develops/acquires appropriate mathematical modeling skills, and/or one seeks advice.

Whatever the case, it is important to remember that one must never fall in love with one’s favorite (mathematical) model! A model is just . . . a model.

More info: www.moshe-online.com

Dr Moshe Sniedovich is a Principal Fellow in the Department of Mathematics and Statistics at the University of Melbourne.

References

Ben-Haim, Y. (2006) Info-Gap Decision Theory: Decisions Under Severe Uncertainty, Elsevier, Amsterdam.

Sniedovich, M. (2007) The art and science of modeling decision-making under severe uncertainty, Decision Making in Manufacturing and Services, 1(1-2):111-136.

Decision Point #22 - 10

Native veg & climateNative vegetation moderates climate fluctuations by recycling moisture back into the atmosphere as well as reflecting less solar radiation, cooling the surface temperature and aiding cloud formation. According to McAlpine et al. (2007) it is too simplistic to attribute climate change solely to greenhouse gases. This research found that vegetation clearing increased average summer temperatures in eastern Australia by between 0.4 and 2 degrees and decreased summer rainfall by 4-12%, contributing to hotter and longer droughts. This represents a major economic and environmental risk, by further accentuating the climate impacts of greenhouse gases.

By Clive McAlpine (UQ, Brisbane Node, AEDA)

Australia’s natural resources are particularly vulnerable to global climate change, especially from droughts. And these droughts are getting hotter and persisting longer, reducing scarce run-off into our river systems. Current global climate trends are exceeding worst case scenarios predicted by the UN’s Intergovernmental Panel on Climate Change.

To date, Australia’s policy response to climate change has focused on the mitigation of our greenhouse gas emissions and adaptation of forecasted land use. However, these responses don’t go far enough because they don’t acknowledge that increasing greenhouse gases and land clearing are part of a complex coupling of the land surface and atmosphere.

The work of McAlpine et al. (2007) highlights the need to include the management of Australian land surface as an additional mitigation strategy to climate change. This was the message recently presented to policy makers in Canberra by Dr Clive McAlpine, Jozef Syktus, climate modeller from the Queensland Climate Change Centre of Excellence, supported by Dr Stuart Pearson from Land &Water Australia.

Recognising the problem is the first step towards a solution. Recognition involves adopting a broader perspective on climate change which includes multiple processes and feedbacks (besides CO2 levels) impacting on Australia’s regional climate. This requires a new paradigm which acknowledges the tight coupling of the land surface and Australia’s regional climate combined with an anticipatory policy and management framework. This framework aims to maintain the beneficial feedbacks of deep-rooted native vegetation on the climate system.

Long-term anticipatory (rather than reactive) policies will be crucial to mitigating and successfully adaptating to dangerous climate change. Anticipation involves recognising the uncertainties of climate change, and developing options under uncertainty as a foundation for strategic policy making. The ground rules for managing Australia’s natural resources and landscapes are changing rapidly. Scientists need to actively inform this process so as to ensure win-win sustainable solutions for land and water resources, biodiversity conservation, rural communities and the agricultural sector.

Despite the uncertainties regarding the feedbacks on climate stemming from land clearing, McAlpine and colleagues advocate anticipatory policies which aim to reduce pressures on Australia’s landscapes and restore native vegetation cover before irreversible change occurs.

Managing for climate change is more than cutting CO2 Policy needs to incorporate how we manage Australia’s land surface

Reduction in greenhouse gas emissions is essential but not sufficient to avoid both the risks of positive feedbacks of the land surface on climate and associated risks of accelerated environmental degradation.

Potential mitigation and adaptation measures include: 1) tighter legislative controls on the clearing of remnant native vegetation; 2) strategic protection of regrowth native vegetation in previously cleared sub-tropical landscapes; 3) expanded investment in restoration of native vegetation in the highly modified landscapes of southern Australia; 4) an evaluation of the long-term viability of marginal cropping and grazing lands and their vulnerability to soil and vegetation degradation; 5) maintenance of cover on fallow cropping lands; and 6) adaptive management of stocking rates according to climate conditions.

The time scale of policy also needs to expand. Historically, policy and land management cycles have been a short-term reaction to drought and economic hardship. Policy makers are responding to global warming through emissions reductions and adaptation strategies, but are failing to see the whole system. Carbon sinks (eg, carbon credits, plantations, green fleet schemes) are used to justify business as usual, while the biophysical and ecohydrological functions of whole landscapes are being ignored.

The increased frequency and extent of severe droughts and bushfires associated with climate change is likely to be the next major driver of vegetation loss. A move to long-term anticipatory policy and planning strategies are urgently needed to restore and maintain the favourable feedbacks between native vegetation cover and Australia’s local-regional climate. These feedback can then buffer the more severe affects of greenhouse-driven climate change.

More info: c.mcalpine@uq.edu.au

Further reading

McAlpine, CA, Syktus, J, Deo, RC, Lawrence, PJ, McGowan, HA, Watterson, IG, and Phinn, SR. (2007) Modeling the Impact of Historical Land Cover Change on Australia’s Regional Climate. Geophysical Research Letters 34, L22711, doi:10.1029/2007GL031524.

“The increased frequency and extent of severe droughts and bushfires

associated with climate change is likely to be the next major

driver of vegetation loss”

Imag

e by

Phil

Gib

bons

Decision Point #22 - 11

By Philip Gibbons (ANU, Canberra Node, AEDA)

When the average person drives down the Hume, Newell or any of our other inland highways they see a park-like landscape of pasture or crops with a scattering of mature trees and small clumps of trees – or ‘paddock trees’.

These landscapes are repeated around the globe: in oak woodlands of Europe, Central Asia and North America; in the savannahs of Africa; in farmland of Central America; and in semi-arid rangelands of South America.

The trees that are a dominant feature of these landscapes are described by some ecologists as ‘keystone structures’ because of their ecological importance relative to their low abundance.

Paddock trees provide habitat to many species, they help maintain the viability of some native fauna by acting as stepping stones across highly cleared landscapes, they improve soil properties, help keep salinity at bay, mitigate soil loss, provide shade for stock and are a cost-effective source of seed for regeneration. Many of these functions are performed only by mature trees.

Paddock trees represent most of what is left of some ecological communities. For example, did you know that remnants less than one hectare represented around half of remaining box gum grassy woodland in SE Australia.

In a recently published study undertaken as part of my activities with AEDA, colleagues and I predicted that under a business-as-usual scenario, most scattered trees will be lost from the world’s agricultural landscapes by the end of the century.

This prediction equates to the loss of around 60% of tree cover on freehold land in temperate agricultural landscapes of SE Australia.

The passing of the paddock tree The future of native tree cover in agricultural landscapes

What can we do to reverse this trend?

The most obvious reason for the decline of paddock trees is the absence of sufficient recruitment of new trees and most management recommendations reflect this observation.

However, we found that increasing recruitment AND reducing mortality among existing trees was a much more effective management strategy than addressing recruitment alone. This recommendation is similar to that for other long-lived species that are under some form of harvesting or population pressure such as cold water fish and marine turtles.

Strategies such as reducing land clearing (paddock trees continue to be cleared in most Australian states), controlling spray drift and preventing stock congregating around the root zone of trees are therefore important management actions that should be supported in any program that aims to address this issue.

Using a simulation approach we also found that, provided mortality could be kept at low levels, new trees would only have to be recruited at a frequency equivalent to around 15% of the life expectancy of the tree species (30+ years). This is an important result because there is a cost associated with tree recruitment in agriculturally productive landscapes – a cost that the public often bears through agri-environmental, incentive or stewardship schemes.

One other alarm bell sounded from our work: the longer we wait to act, the less effective any management interventions will be.

Further reading

Gibbons P, Lindenmayer DB, Fischer J, Manning AD, Weinberg A, Seddon J, Ryan P and Barrett G. (2008). The future of scattered trees in agricultural landscapes. Conservation Biology. DOI: 10.1111/j.1523-1739.2008.00997.x.

“Under a business-as-usual scenario, most scattered trees will be lost from the world’s agricultural landscapes by the

end of the century. This prediction equates to the loss of around 60% of tree cover on freehold land in temperate agricultural

landscapes of SE Australia.”

Paddock trees are a dominant landscape feature in agricultural landscapes in SE Australia.

Imag

e by

Bre

ndan

Win

tle

Decision Point #22 - 12

Scary by half

The Introduced flora of Australia and its weed status was launched last month and it contains some scary numbers. Some 28,000 plant species have been introduced to Australia since European settlement meaning around half of Australia’s plants originated from overseas.

The Introduced flora lists 2739 foreign species that have become weedy (like the Pattersons curse in the picture), and a further 5907 that have a history of becoming weeds overseas (though they’re not weeds yet).

“Australia has such a diversity of climates we can be sure than many of these ‘weeds in waiting’ will eventually find their way to a site that suits them – and then they will simply explode in numbers,” say Rod Randal, the scientist who compiled the flora. Mr Randal is with the WA Department of Agriculture and Food and the Weeds CRC. “We’re pretty adept at moving plants and seeds around, on purpose or by accident. This gives weeds the chance they need to spread and try their luck in new locations.”

The flora is available free from the Weeds CRC at http://www.weedscrc.org.au/weed_management/intro_flora.html

Applied Environmental Decision Analysis A Commonwealth Environment Research Facility

aedaSmart science for wise decisions

AEDA stands for Applied Environmental Decision Analysis, a research hub of the Commonwealth Environment Research Facility program. The CERF program is funded by the Australian Government’s Department of the Environment, Water, Heritage and the Arts.

AEDA’s members are primarily based at the University of Queensland, the Australian National University, the University of Melbourne and RMIT.

Decision Point is the monthly magazine of AEDA. It is available free from the AEDA website . If you would like to receive an email alerting you to new issues as they are released, please visit http://www.aeda.edu.au/news

Decision Point is written and produced by David Salt. If you have news or views relating to AEDA or of interest to AEDA members, please send it to David at David.Salt@.anu.edu.au

Decision Point is printed on recycled stock.

THE AUSTRALIAN NATIONAL UNIVERSITY

Moving species to sites where they do not currently occur or have not been known to occur in recent history is anathema to traditional conservation biology (think cane toads, rabbits and foxes). But in an age of climate change, disappearing cloud forests and stressed out coral reefs, assisted colonisation is a notion that can’t be ignored.

In July a panel of eminent ecologists (including AEDA core researchers Hugh Possingham and David Lindenmayer) released a discussion paper in the journal Science on assisted colonisation that advanced the debate by proposing a decision framework to facilitate the process. Their decision framework can be used to outline potential actions under a suite of possible future climate scenarios.

Determining whether a species faces significant risk of decline or extinction under climate change requires an in-depth knowledge of the underlying species’ biology as well as the biological, physical, and chemical changes occurring within its environment. The risk of extinction for many widespread, generalist species found across a range of habitats may be low. In this case, the option of moving such species outside their present ranges would be dismissed.

Some species will also disperse sufficiently to maintain large populations and range sizes (for example, highly dispersive insects or birds with generalist life histories) and others may adapt in situ. Where species are perceived as being at moderate risk from climate change, improvements in connectivity to actual or potential habitat at higher latitudes and altitudes may be sufficient.

But where there is a high risk of extinction with climate change, for example species that are confined to disappearing habitats, then translocation should definitely be explored as an option. And exploring this option involves assessing whether the translocation is technically possible and whether the benefits of doing so outweigh the biological and socioeconomic costs.

By applying a decision framework to the translocation debate its possible to identify low-risk situations where the benefits of assisted colonisation might be realised with adverse outcomes minimised.

Reference

Hoegh-Guldberg O, Hughes L, McIntyre S, Lindenmayer DB, Parmesan C, Possingham HP and Thomas CD (2008), Assisted colonization and rapid climate change, Science, 321: pp345-346

For a lively Possingham discussion on assisted colonisation, check out his editorial in DPoint #17 (downloadable free from the AEDA news page at http://www.aeda.edu.au/news)

Assisted colonisation Moving with the times

the funny end bit

Populations of staghorn corals have already been lost from some high-latitude locations because of increasing thermal stress. Introducing lower-latitude, heat adapted genotypes to these degraded sites may hold little risk.