Michael Drielsma and Simon Ferrier are part of the GIS Research and Development Unit within the New South Wales Department of Environment and Conservation (PO Box 402, Armidale, NSW 2350, Australia. Tel. +61-267760033. Email: email@example.com; firstname.lastname@example.org). The work presented here has evolved through the author's involvement in collaborative projects between modelling specialists, biologists, planners and stakeholders, aimed at finding better ways to inform decision–making at the scale of the region or landscape, where information on individual species, processes and specific locations is incomplete.
Summary Landscape scenario modelling is a useful aid to planning for biodiversity conservation. Vegetation condition modelling is increasingly being integrated into such analysis. Model complexity and model uncertainty are critical factors that must be addressed when tailoring vegetation condition modelling to individual applications. We describe three approaches that we have used to compare the effects of different landscape scenarios on vegetation condition. The first is a simple land-use–condition approach where vegetation condition is determined solely by land use. The second is a land-use–regeneration approach that introduces transition functions to model vegetation condition dynamics associated with land use change. The third is a threat–regeneration approach, which models vegetation condition dynamics based on the interaction between regeneration and a range of mapped threats. The three approaches represent a progression towards increased refinement and realism, but also increased complexity and data requirements. We examine the relative usefulness of the three approaches and conclude that there is no single ‘silver bullet’ solution but recommend judicious matching of approaches to applications within a collaborative and adaptive setting.
The vegetation condition of a site is often based on its deviation from an undisturbed state for the vegetation type concerned (Parkes et al. 2003). We extend this definition by viewing vegetation condition as an estimate of the proportion of the species of plants and animals originally occurring at the site (if it were in an undisturbed state) that are expected to be present at the site in its current state. Although higher accuracy can be achieved by modelling the suitability of habitat for individual species, vegetation condition is a useful abstraction because it is able to encapsulate a significant component of habitat quality for a broad range of species (Parkes et al. 2003).
In this paper, we explain the role of landscape scenario modelling of vegetation condition in the context of biodiversity conservation. We explore some of the issues of complexity and uncertainty that must be addressed within model design. We present and then compare three approaches to scenario modelling of vegetation condition from regional scale projects carried out in New South Wales, Australia. These examples represent a progression from a simple data-lean approach to more complex, process-based, realistic and data-hungry approaches.
Scenario modelling of vegetation condition: Why do it?
Although we use the term prediction throughout this paper, the modelling approaches we present here are not intended to accurately predict in the same way as, for example, weather models. Scenario modelling assists planning by anticipating plausible futures (Clark et al. 2001). It can also help to shape the future (Haag & Kaupenjohann 2001; Peterson et al. 2003; Schreiber et al. 2004) by facilitating the integration of information and knowledge into the adaptive management process as it emerges (Carpenter 2002). Unlike weather forecasting, the objective of scenario modelling is not to foresee the future, but rather it is to describe what is possible (Peterson et al. 2003). Scenario modelling of vegetation condition can take on a range of roles including the role of a learning tool where stakeholders can configure, interact with and assess scenarios (as well as the ecological models themselves) based on their evolving mental models (Peterson et al. 2003).
Evaluating predicted outcomes for biodiversity provides a basis for comparing alternative land use scenarios. This comparison might be made between the do-nothing scenario and a proposed change involving development, conservation, or a mix of both. Alternatively, the comparison can be made between competing proposals. It may be tempting to prioritize on the basis of current conservation values alone. This, however, takes no account of the trajectory of a scenario when currently active or foreseeable processes are considered. Where conservation values are under little threat, for example, it would emerge through scenario modelling that added protection of such areas makes little improvement to the future conservation outcome. In such a case, it might be better to allocate scarce conservation resources elsewhere, where conservation values may currently be less, but where these values are subject to more threat. Similarly, it might be shown to be of greater benefit to actively restore areas where the underlying factors leading to the currently degraded state require remedial work (Hobbs & Harris 2001) rather than to allocate resources to areas that are regenerating passively.
Ignoring the dynamics of habitat change is to implicitly assume that changes in land use and management will produce changes to habitat that are both certain and instant, which is clearly false, and in the context of land use decision-making is also likely to lead to a misallocation of scarce conservation resources. For example, a scenario that includes the clearing of mature forest and the regeneration of a similar area that has been previously cleared could result in an assessment that there is no change to the overall conservation outcome. Factoring in the dynamics of vegetation and habitat should lead to a more realistic assessment, i.e. that the above scenario would result in a net conservation loss at least in the short to medium term.
The development of spatially explicit landscape scenario analysis greatly enriches the decision support capabilities of modelling. A scenario may comprise a range of actions that vary in substance and location (and may overlap spatially). Assessing the combined effect of a given set of actions may require more than simply summing the effects of individual actions, and may need to consider interactions between the various actions over the landscape (Margules & Pressey 2000).
The relative outcomes of competing scenarios will depend on the timeframe chosen. Short timeframes favour the retention of existing values and actions that are likely to yield rapid benefits; longer timeframes favour the values that may take longer to materialize but may eventually provide greater benefits. To view the full picture it may be necessary to model outcomes over a range of timeframes.
Issues of complexity
Complex adaptive systems, such as ecological systems, are characterized by non-linearity, self-organization, diversity, aggregation and flows (Levens 1998). Loehle (2004) identified complexity as having at least six dimensions in ecology: spatial, temporal, structural, process, behavioural and geometric. Due to this inherent complexity, modelling exercises will always be confronted by uncertainty surrounding the very nature of the system and how it works (Burgman 2005a,b). In many ways such systems are unknown and unknowable.
A model can be extensive in scope without being highly complex; or it can be intensive and highly complex while focusing on a portion of the environment (e.g. single-species Population Viability Analysis). Modelling will always involve some simplification and unavoidable trade-offs must be made between generality, realism and precision (Levens 1966). Models need to be crafted for a particular purpose, in such a way as to adequately meet the objectives of the modelling exercise while avoiding undue model complexity (Levens 1998; Lansing 2003). This is achieved through strategies such as reduction, scaling and coarse-graining (Loehle 2004; Buchanan 2005).
Vegetation condition can be modelled generically (i.e. a number of attributes are incorporated into a single score). Alternatively, modelling can be applied separately to a number of vegetation condition components such as tree height, tree hollow density and density of coarse woody debris. In this way, vegetation condition dynamics and habitat can be more closely related to the landscape processes of regeneration and threats. A single score for vegetation condition can be calculated by combining the individual scores for vegetation condition components or a habitat score can be calculated by combining the scores according to the habitat requirements of individual faunal species.
Hilderbrand et al. (2005) has pointed out that the assumptions surrounding the dynamics of vegetation are based on a set of restoration myths. The climax model of succession (Clements 1916, 1936; Watt 1947; Holling 2001), based on the ‘carbon copy’ myth, has been recognized as useful but unrealistically static (Holling 1992). Whereas myth making and usage may be unavoidable, we should be explicit about the assumptions we draw on. The single climax model has been largely superseded by recognition of multistate, cyclical and continually dynamic ecosystems (Westoby et al. 1989; Levens 1998; Hobbs & Harris 2001; Holling 2001; Wallington et al. 2005). The state-and-transition model characterizes ecological systems as possessing a set of alternative states, and a set of transitions between states (Westoby et al. 1989). Some systems have been described as reaching ‘quasi-equilibrium’ states (Shugart & Smith 1996). This recognition of the dynamic nature of ecosystems highlights the limitations of adopting rigid benchmarks to define natural states.
Three approaches to landscape scenario modelling of vegetation condition
Three approaches to landscape scenario modelling of vegetation condition are described below. The approaches are based on case studies that the authors have been involved with. They are not an exhaustive set but rather a subset of possible modelling approaches. They represent a progression from simple and static approaches to more complex and dynamic ways of modelling changes to vegetation condition. For any of the three approaches below, vegetation condition can be modelled generically or for any subset of vegetation condition components (Box 1).
The land-use–condition approach
The land-use–condition approach (New South Wales National Parks and Wildlife Service 2002) includes the simplest manifestation of vegetation condition dynamics where changes to land use translate directly and instantly to changes in vegetation condition. Changes to vegetation condition therefore reflect the expected long-term outcome, after all transitions have completed. The timeframe is effectively set to infinity – the transition phase is ignored. Threats are not explicitly recognized but are implicit within each land use class. Uncertainties associated with regeneration can be accommodated by reducing the vegetation condition expected from the land use change. At its simplest, this approach requires only a land use map and a table setting out the vegetation condition values associated with each land use.
The land-use–regeneration approach
The land-use–regeneration approach (Resource and Conservation Assessment Council 2004) introduces the dynamics of vegetation condition change into the modelling while continuing to link the equilibrium level of vegetation condition directly to land use. This approach accommodates aspects of climax theory and the state-and-transition model. It recognizes a range of states, determined by land use and gives form to the transitions between states. Stable climax states are modulated by land use, allowing only limited recovery within production systems. Although other functional forms could be applied, we have developed a variant of the logistic function (s-shaped curve) to represent transitions between states (Fig. 1). The general shape of the regeneration transition curve has been chosen to reflect an initial sluggishness to regeneration (the rebuilding of soil and seed bank) that is followed by relatively rapid change (plant growth) with the finishing touches (e.g. the development of tree hollows) proceeding at a slow pace. The function is configured through the selection of transition time and equilibrium vegetation condition parameters (Table 1).
Table 1. Example of table used to define vegetation condition transitions for the land-use–regeneration approach
Land use A
Land use B
Transition time – regeneration
Transition time – degradation
This approach requires a map of current vegetation condition. Mapping current vegetation condition is a separate modelling challenge (e.g. New South Wales Department of Environment and Conservation 2004) (see other papers in this issue). With this approach a site could at any point in time, either be in a stable state or somewhere within a transition between two states. The model for each land use is divided into two separate functions. If the current vegetation condition is below the equilibrium vegetation condition, the regeneration function applies; if the current vegetation condition is above the equilibrium level, the degradation function applies. At its simplest, the approach requires a minimum of three parameters: the equilibrium vegetation condition for the stable state, the transition time for regeneration (primary succession) and the transition time for degradation. Degradation, especially actively aided by mechanical means, is often much faster acting than regeneration. The transition function can be further refined to include time delay parameters to accommodate such phenomenon as extinction debt and the reliance on climatic events and species colonization.
The stage at which a site has progressed along its transition is the point where the current vegetation condition intersects either the regeneration or the degradation curve for the land use applying to the site. The predicted future vegetation condition resulting from a land use for that site can then be derived by moving along the relevant curve to the right by adding the relevant timeframe to the x-axis. At any point, a site can be switched to an alternative land use that will place it at new location on an alternative transition curve. The process is repeated for each site in the region to derive a surface of predicted vegetation condition.
The threat–regeneration approach
The threat–regeneration approach (New South Wales Department of Environment and Conservation 2004) is based on the interaction between regeneration and a range of threats. The likelihood of threats activating is spatially described by a surface of annual probability for each threat. The approach also requires information on the consequence of each threat (a single value), should it become active, on vegetation condition. Land use, although likely to remain a factor in deriving threat levels, can be superseded in this approach by more subtle, customised management actions that relate directly to threats. These may include management actions such as controlled grazing, weed control and fire management.
Annual probability surfaces for threats, such as vegetation clearing, logging, land degradation, weed invasion and firewood collection, are the products of separate modelling exercises based on a range of methodologies using a range of environmental and cultural variables (New South Wales Department of Environment and Conservation 2004) (Fig. 2). Threat consequence information is typically obtained from expert judgement. As in the second approach, transition functions are used in this approach to describe the dynamics of regeneration (Fig. 3 and Table 2).
Table 2. Example of table used to describe influence of threats in the threat–regeneration approach
Vegetation condition consequence
Canopy (max. 45)
Understorey (max. 45)
Coarse woody debris (max. 10)
The annual probabilities associated with each threat for the instances of highest (PHigh) and lowest (PLow) threat are shown. Instances of intermediate threat are interpolated between these.
A decision-tree methodology has been adopted as a means of combining multiple threats with regeneration to derive a prediction of the future vegetation condition at a site (Fig. 4). Each column of nodes on the decision-tree represents a threat; the branches represent all possible combinations of active (black nodes) and inactive threats (white nodes) at a site. In Fig. 4 the vegetation condition consequence of each relevant threat (values ranging from zero condition to one hundred, for maximum condition) is indicated at each active node. The vegetation condition consequence for an entire branch (furthermost right column) equals the lowest vegetation condition consequence found along the bran (including its current vegetation condition). For each cell in the region, the expected future vegetation condition is calculated as follows.
1Calculate the threat probabilities for the given timeframe t (in years) as: Pt = 1– (1 –Pa)t, where Pa is the annual probability extracted from the threat surfaces. The probability of a threat not acting equals one minus the threat probability.
2Calculate the product of the probabilities along the length of each branch to obtain the probability of that combination of threats eventuating (across all branches these should add up to one).
3Multiply the results from 2 (above) by the vegetation condition consequence for the branch to obtain the expected future vegetation condition for each branch of the tree.
4Sum the result from 3 (above) across all rows to obtain the expected vegetation condition.
At each site, there is a probability that no threat will eventuate (the lowest branch in Fig. 4). In such a case, the vegetation condition consequence is based on regeneration where the current vegetation condition determines the starting point on the regeneration function (see Fig. 1). The process is repeated for all sites (grid cells) in the region to produce a map of predicted future vegetation condition (Fig. 5).
The relatively simple vegetation condition modelling within the first approach provides a non-dynamic, easy to explain regional overview, where people can quickly see the long-term results of a scenario. With this approach all the complexity, including uncertainty and vegetation condition dynamics, which would otherwise add layers of complexity to the analysis, are distilled into vegetation condition scores. For example, restoration may be given a lower vegetation condition score where restoration success is expected to be slow and uncertain. This approach therefore relies heavily on judicious selection of these scores.
The second approach refines the first approach by including a transition phase between stable states. It recognizes that much of the current landscape is not at equilibrium and that land use change will often alter the direction of vegetation condition change but will not necessarily lead to an equilibrium state, at least in the short term. A weakness of this approach for decision support is that, as with the first approach, it assumes that threats are uniform within a land use type and that therefore all areas within a given land use will eventually revert to an equal vegetation condition regardless of their unique situation. Although this may be a realistic simplification in respect to a land use such as cropping, where the effects on vegetation condition are uniformly devastating, this is not the case with other land use classes. Within rangelands, for example, remote, rugged and infertile sites tend to retain relatively high vegetation condition because of the low intensity of grazing. Although such limitations could be addressed by further dividing the land use classes according to intensity of land use, there comes a point where it becomes more efficient, in terms of the management of complexity, to address the threats directly.
The explicit inclusion of threatening process within the third approach further refines the analysis by linking outcomes more directly to causal factors. In this way, the variegated nature of landscapes can be recognized and subtle changes to management scenarios can be explored, unconstrained by rigid land use classes (Fig. 6). However, the approach does require more data, more complex models and slower computation.
The second and third approaches provide snapshots of predicted vegetation condition at specific future times. Where a complete long-term view of landscape changes is required, these approaches must be run over a range of time intervals.
We have attempted to demonstrate the imp-ortance of scenario modelling of vegetation condition as an aid to land use decision making in the context of biodiversity and have stressed how the issues of complexity and model formulation must be addressed.
Modellers are often constrained by a lack of knowledge and understanding of system response to land management. Practitioners (ecologists, land managers and regenerators), having more hands-on field experience, can assist in the refinement of models. To this end specialist modellers and practitioners should be encouraged to form collaborative teams.
We have described, as examples, three approaches that we have developed for particular projects within unique circumstances: the land-use–condition approach, the land-use–regeneration approach and the threat–regeneration approach. The three approaches continue to be refined and customized for use in a range of applications as part of tool sets for evaluating scenarios in terms of biodiversity outcome. In these broader modelling frameworks vegetation condition modelling becomes one of a number of other modelling components that include vegetation pattern mapping, spatial configuration analysis and population viability analysis.
Although the approaches described here are not intended to be exhaustive, they represent a range of available approaches to scenario modelling of vegetation condition. We do not recommend or prescribe any particular approach, but rather recommend the tailoring of modelling approaches to suit the requirements as well as the limitations associated with individual applications.
The work presented here is likely to be complemented by ongoing developments in the mapping of current vegetation condition and site-based assessment of vegetation condition (see other papers in the issue). We recognize a number of priorities for further refinement of vegetation condition modelling. These include the integration of vegetation scenario modelling into a learning environment. We also see the incorporation of approaches to better deal with spatially incomplete data, and uncertainty generally, as a priority for further work.
Box 1. Three approaches to scenario modelling of vegetation conditionApproach 1. The Land-use–condition approach. This, the simplest of the three approaches links predicted future vegetation condition directly to land use.Approach 2. The Land-use–regeneration approach. This approach utilizes a transition function to describe the dynamics of vegetation condition changes following land use change. As with Approach 1, equilibrium vegetation condition values are linked directly to land use.Approach 3. The Threat–regeneration approach. This, the most realistic and complex of the three approaches, is used to predict vegetation condition outcomes independently from land use by modelling changes to current vegetation condition based on threats and regeneration.
We gratefully acknowledge the ideas and contributions made by John Westaway, Philip Gibbons, Roderic Gill and the Centre for Ecological Economics and Water Policy Research UNE, Dave Robson, David Keith, Glenn Manion, Kellie Mantle, Brendan Rennison, Jill Smith, Richard Thackway, Louise Drielsma. We would also like to thank the paper's referees for their valuable and constructive contributions that vastly improved the paper's clarity and readability.