SEARCH

SEARCH BY CITATION

Keywords:

  • benefit function;
  • conservation planning;
  • cost-effectiveness;
  • Lepidoptera;
  • mesic grasslands;
  • site-selection algorithm;
  • vascular plants

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • 1
    Conservation management often encompasses multiple, alternative management actions on a given site, involving habitat restoration and maintenance for example. Which actions are preferable depends on the conservation goals, the expected outcomes of actions, and their associated costs. When actions affect habitat quality differently, species that vary in habitat requirements will not respond to the actions in the same way. When all these species are of conservation concern, trade-offs between them are inevitable and the selection of appropriate actions becomes less straightforward. Although this is a common planning problem, it has received little attention in the conservation planning literature.
  • 2
    We demonstrate how to obtain cost-effective planning solutions for a set of species with contrasting requirements, when multiple alternative conservation actions are available for each site. We investigate the strength of trade-offs between species with different habitat preferences, when planning optimal management of a set of semi-natural mesic grassland sites. A community of vascular plants and Lepidoptera species depends on such grasslands, which are maintained by cattle grazing. The various species differ in their responses to grazing intensity. We apply an algorithm that selects a grazing intensity for each site to maximize the benefit over all species, under a given budget constraint.
  • 3
    The optimal grazing intensity for sites, and consequently, the expected representation of species, was sensitive to the relative values assigned to species (weights) and the budget available. The outcome also depended on assumptions regarding species representation under suboptimal management. A sensitivity analysis showed that the trade-offs between species were strong, illustrating the potentially significant consequences of conservation decisions.
  • 4
    Synthesis and applications. Maximizing conservation benefit over all species may result in high representation of some species at the cost of others. Although this is a natural consequence of budget limitations and conservation priorities, understanding these consequences is essential if planning is to accommodate species with different conservation needs. Our methodology is a novel extension of conventional reserve selection methods and can support planning for species with contrasting requirements, leading to more robust and cost-effective conservation decisions.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

In the struggle to conserve biodiversity, increasing amounts of effort are directed towards improving the population sizes of vulnerable species, and towards restoring degraded ecosystems (e.g. Dobson, Bradshaw & Baker 1997). Large budgets are devoted to these restoration attempts; the European Commission allocated EUR 1·3 billion over the period 1992–2005, for example, to maintain and restore the natural habitats and populations of endangered species (EC 2007). The conservation toolbox contains a wide array of restoration and maintenance measures. Agri-environment schemes, for example, have been introduced in most European countries to mitigate the negative trends observed in species dependent on agricultural environments (e.g. Kleijn & Sutherland 2003; Whittingham 2007). A particular cost and impact is associated with any given measure, and this will often differ according to location, since site conditions are typically spatially heterogeneous. In fact, the effects of each measure on different species can also vary widely because habitat requirements are species-specific (see Pöyry et al. 2005; Pykälä 2005). When species which require conservation attention have contrasting requirements, a trade-off must occur between management actions if all the focal species are to benefit. Consequently, it is no longer straightforward to solve such a planning problem cost-effectively.

Optimization tools which can aid in this planning process have received considerable attention in the context of reserve selection during the past 20 years (Cabeza & Moilanen 2001; Margules & Sarkar 2007). Reserve selection tools enable the selection of efficient sets of sites which together fulfil a particular conservation goal, such as maximizing the representation of a set of species for a particular budget. Indeed, it seems that optimization tools have become an essential tool in conservation planning whenever cost-effective planning solutions for large sets of sites and species are required.

Surprisingly, however, this approach to conservation planning problems has been relatively underused in areas other than reserve selection. The few examples are found in the planning of forest management (Hof et al. 1994; Bevers et al. 1995), habitat restoration (e.g. Newbold & Eadie 2004; Drechsler et al. 2006; Westphal, Field & Possingham 2007), landscape configuration (Holzkämper, Lausch & Seppelt 2006) and threat reduction (Wilson et al. 2007).

The selection of cost-effective sets of sites and actions for multiple species can be complex for computational reasons (a large number of combinations of sites and actions, Van Teeffelen & Moilanen 2008). However, the primary reason for the complexity of the selection process is that strong trade-offs have to be made between species with different requirements. In such cases, predicting which sets of sites and actions may lead to a cost-effective solution is difficult. In multi-action planning problems, where actions can benefit certain species while adversely affecting others, it is important to investigate to what extent trade-offs occur between species. The few previous studies which have aimed at identifying optimal sets of conservation actions for multiple species have not directly addressed this issue. In the study by Holzkämper, Lausch & Seppelt (2006), landscape configurations optimized for one species were not harmful to others. Hof et al. (1994) investigated the trade-offs between two species only at a theoretical level, while Bevers et al. (1995) and Wilson et al. (2007) did not present trade-offs between species groups. Furthermore, except for Wilson et al. (2007), these studies did not consider the cost implications of the actions involved. Since conservation planning is characterized by limited budgets and solutions can differ substantially according to the budget available, we consider it imperative to incorporate the costs of the actions and the budget limitation into the optimization when evaluating the trade-offs between species.

In this study, we demonstrate a novel method for planning conservation actions cost-effectively. This method can incorporate multiple alternative actions and their associated costs, as well as species-specific responses to those actions. We apply this method to a system of mesic semi-natural grasslands in Finland, where several plant, butterfly and moth species have contrasting requirements with respect to the intensity of cattle grazing allowed on the sites (see Pykälä 2004; Pöyry et al. 2005). Using habitat modelling, we predict species responses to three alternative grazing management actions, while the cost of grazing varies with grazing intensity and patch size. We investigate the sensitivity of solutions (in terms of the actions selected, species representation and cost) to: (i) budget size, (ii) assumptions regarding species representation under suboptimal management, and (iii) the relative priorities (weights) assigned to species. Since species have different habitat requirements, the relative weighting of the species will influence which action is preferred at individual sites. Together, these analyses will demonstrate how conservation planning problems with multiple alternative actions can be handled in situations where trade-offs between species with contrasting requirements need to be made.

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

action-selection method

To allocate management actions to sites, we used a forward stepwise heuristic algorithm, which finds optimal or near-optimal solutions for multi-action selection problems (see Van Teeffelen & Moilanen (2008) for a detailed description of the algorithm). The algorithm uses the concept of a ‘benefit function’ (Bevers et al. 1995; Arponen et al. 2005) to attribute a value to species representation in a set of sites. A benefit function f[Rj(X)] is an increasing function of the representation of a species j, Rj(X). This function specifies how the conservation value of a set of site–action pairs, X, changes when the representation level of a species changes due to the addition or removal of a site–action pair x to/from X[for a graphical illustration, see Van Teeffelen & Moilanen (2008)]. The term ‘site–action pair’ is used to indicate a site to which a particular management action has been assigned. The total value of the set of site–action pairs is calculated by summing over species-specific values, multiplied with a species-specific weight wj:

  • image(eqn 1)

Weights can be varied between species when some species are given conservation priority over other species, because they are more vulnerable, for example. By assigning a higher weight to one species than to another, changes in the representation of that species have a larger effect on the value F(X). As a consequence, site–action pairs that improve the representation of this species have a higher marginal gain, which increases their chances of being selected by the algorithm (see equation 3).

Arponen et al. (2005) and Van Teeffelen & Moilanen (2008) tested various shapes of benefit functions (linear, ramp, concave and sigmoid). In this study, we used a concave benefit function because it tends to push species representation to higher levels and results in higher representation levels for rare species than other functions (Arponen et al. 2005). The following function f was used for all species (note however that different functions can be used for different species if preferred):

  • image(eqn 2)

where rj(xia) is the representation of species j at site i when management action a is assigned to site i, resulting in site–action pair xia. At every iteration, the algorithm calculates the marginal gain Δx for each site–action pair, this being the change in total value divided by the cost of the site–action pair, cx, conditional on the value from already selected site–action pairs F(X):

  • Δx = (F(X + x) – F(X))/cx.(eqn 3)

The site–action pair that returns the highest marginal gain is then added to the set of selected site–action pairs F(X). The algorithm terminates either when no further improvements are possible, or when no further improvements can be afforded within the remaining budget. The result is a set of site–action pairs with the associated cost and species representation levels. Note that although multiple actions are available for each site, only a single action can be chosen for each site.

material

Site data

We use data on plants, butterflies and moths in semi-natural grasslands in Finland.

Two data sets relating to species occurring in semi-natural grassland sites were used in this study, and were described in detail in Pöyry et al. (2004, 2005) and Raatikainen, Heikkinen & Pykälä (2007). The first data set was collected during the year 2000 from 47 mesic semi-natural grassland sites situated in southern Finland. The sites were divided into three categories according to their management history: intensive pastures (n = 20), extensive pastures (n = 12) and abandoned pastures (n = 15). The intensive pastures were grazed by cattle during the study year, and they had a recorded history of grazing management dating back several decades. In the extensive pastures, the last grazing event had occurred 1–9 years prior to the field study. In the abandoned pastures, grazing management had ceased ≥ 10 years prior to the field study, and these sites had been subject to gradual overgrowth. All the sites included in the study were unforested, however. The second data set was collected in 1999 and 2000 and consists of 20 sites classified as intensive pastures (n = 9), extensive pastures (n = 1) and abandoned pastures (n = 10).

Site area varied from 0·25 ha to approximately 6 ha. To obtain comparable samples of species and environmental data, a plot of 0·25 ha was placed at each site (see Raatikainen, Heikkinen & Pykälä 2007). Nectar plant abundance was visually estimated three times during the year 2000 on a scale ranging from 0 (no nectar plants) to 10 (several abundant nectar plant species) in the 0·25 ha plot (see Pöyry et al. 2004). Soil phosphorus (P) and pH were analysed from samples collected in 45 sites included in the first data set. Fifteen 1 m2 plots were placed within each 0·25 ha plot with stratified random sampling (Raatikainen, Heikkinen & Pykälä 2007). The mean vegetation height of a site was calculated on the basis of direct measurements (Stewart, Bourn & Thomas 2001) from the 1 m2 plots and used as a surrogate for the intensity of disturbance caused by grazing animals.

Plant data

We used the number of 1 m2 plots per site where the species was present as an estimate of the abundance of each species per site (with a maximum of 15 plots). Plant species were classified a priori by their management preferences as preferring either intensive grazing, extensive grazing or no management (Pykälä 2005). A subset of 27 species (with varying management preferences) was used in habitat modelling, for the 45 sites for which soil data was available. These species were classified into positive, neutral and negative indicator species of semi-natural grasslands in Finland following Pykälä (2001).

Lepidoptera data

We used abundance data relating to 21 species (Pöyry et al. 2004), with varying management preferences (following Pöyry et al. 2005). All those sites for which larval host plant data was available (56 sites) were used for habitat modelling (1 abundance data point per species per site). Species were classified as having increasing, stable or declining distributional trends based on Huldén et al. (2000) and Kuussaari et al. (2007).

Management actions

For each meadow, we considered three simple grassland management actions which differ in grazing intensity: annual grazing, biennial grazing (every second year), or no grazing. These management actions mimic the disturbance levels observed for the three management categories of intensive, extensive and abandoned pastures. Both the vegetation height and the abundance of nectar plants are influenced by the grazing intensity or by the time elapsed since the last grazing event. In turn, species respond to changes in these variables as a result of grazing. For example, vegetation height influences the amount of light available, which affects the ability of the plant species to compete with others. Lepidoptera species are also affected by the grazing pressure: intensive (annual) grazing increases mortality of lepidopteran larvae due to trampling, for example. In order to predict species’ responses to each grazing action, we needed estimates for vegetation height and nectar plant abundance under each of the current management intensities. We computed mean vegetation height and nectar plant abundance for groups of sites having the same management type (29 sites for intensive, 13 sites for extensive and 25 for abandoned, from the two data sets). We used these averages as the expected outcome for each of the management actions, assuming that the outcome of the management action was independent of the current management intensity.

Management cost

As an estimate of management cost, we used the average annual allowance of EUR 300 per hectare provided by the Ministry of Agriculture and Forestry for the maintenance of semi-natural grasslands in the Finnish agri-environment scheme (Schulman, Heliölä & Pykälä 2006). These grasslands are not typically maintained by farmers unless subsidies are provided, since it is less profitable to keep cattle on low-productive semi-natural grasslands than in ordinary field pastures. The grasslands are unsuitable for cultivation due to the presence of rocky outcrops, poor soils or steep slopes. Hence, we do not consider other costs and benefits than the allowance from the agri-environment scheme. In this scheme, the typical period of a contract for management support is 5 years (Schulman, Heliölä & Pykälä 2006), for which we assume the cost for the annual grazing of a patch to be 5 × EUR 300 = EUR 1500 per hectare. Biennial grazing is assumed to cost half of this amount on average: EUR 750 per hectare. No subsidy is granted when a meadow is not maintained, but for computational reasons (division by cost in equation 3), we assume a cost of EUR 1 per site. Managing all 45 sites by annual grazing would be the most expensive solution: EUR 73 950.

Model fitting

In order to predict species abundances in each meadow under different kinds of management, we used a generalized additive model (GAM, Hastie & Tibshirani 1990) for each species. For the plant models, the available explanatory variables were vegetation height, management intensity, patch area, phosphorous level and pH. For the butterfly and moth models, the available explanatory variables were vegetation height, management intensity, patch area, nectar plant abundance and (for host plant specific species) larval host plant density. We did not consider variable interactions but we tested for correlations and allowed variables with a correlation of rp < |0·6| (Pearson's correlation coefficient).

We used the grasp package (version 3·3, Lehmann, Overton & Leathwick 2002) for S-PLUS version 6·1 (2002) for fitting, selecting and evaluating the models. We used the bi-directional stepwise model selection option, on the basis of the Akaike's Information Criterion (AIC). We allowed smoother degrees of freedom for each continuous variable to vary between 1 and 3. Models where both the ‘vegetation height’ and ‘management intensity’ variables were included were not accepted, since these variables were highly correlated (rp = 0·81). If grasp selected such a model, we manually reran the model selection procedure, subsequently omitting one of these two variables, and retaining the model with the highest deviance explained. We also tested for more parsimonious models using grasp's F-test for a quasi-Poisson model, retaining the model with the better (lower) AIC. The model calibration statistics included deviance explained, correlation and fourfold cross-validated correlation between the predicted and observed abundance values.

Predicting species abundance

We used the set of 45 sites for which we had soil samples in the action–selection analysis. We made three predictions of abundance for each species in each site, one for each management action, by varying the mean vegetation height and the abundance of nectar plants. For our predictions, the management action (annual grazing, biennial grazing, no grazing) was used to represent the variable for management intensity (intensive management, extensive management, abandoned pastures), which was used for model fitting. We thus obtained the predicted abundance of each species in each site under each management action; this served as the input for the action–selection method.

In principle, the predicted abundance levels could be used directly as representation rj(xia) in the action–selection method. However, as the range of predicted abundance levels can vary substantially between species, we scaled the predicted abundance levels for each species by dividing the predicted abundance k of species j at each site i under each action a [kj(xia)] by the maximum benefit that could be obtained for a species in the study area. The maximum benefit was the summed abundance of that species over all sites when each site was managed by Aj: the management action that generated the highest abundance level for species j. Therefore, rj(xia) in equation 2 is represented by:

  • image(eqn 4)

Hence, rj(xia) is a proportion of the maximum representation level that can be obtained for species j in this set of sites. For example, when all sites are managed by annual grazing, this will produce a representation level equal to one for species which are predicted to prefer annual grazing. When sites are not managed by the preferred action, the species representation will be lower, depending on the strength of the fitted species response to changes in management intensity.

analysis

Action–selection was run for a variety of budgets and weights for different species groups. As a baseline analysis, all species obtained equal weights, and the analysis was performed for a series of budgets increasing in increments of EUR 2500 to a maximum of EUR 75 000. The smaller the budget, the more sites were expected to be assigned to the ‘no grazing’ management, since this option is virtually without cost. As the budgets increase, those species depending on more intensive management will subsequently benefit from one of the more intensive management actions. We therefore expect a trade-off between those species which prefer more intensive grazing and those species which prefer no grazing. To analyse the effect of changing the relative priority of species, we systematically increased or decreased weights by one for particular groups, while maintaining the weights of other species constant at one, until the results converged. For each combination of weights, we conducted optimizations for a series of budgets from EUR 0 to 75 000, with increments of EUR 2500.

Although there is always one management action which can be considered the optimal management action for each species (the action that generates the highest predicted abundance for that species, Aj), the habitat models predict a certain level of representation for every species under every management action. It is possible, however, that suboptimal management will not sustain the species over the long term. Allowing representation for suboptimal management could thus lead to an overly optimistic picture. We therefore also performed an analysis whereby the representation of species under suboptimal management was set at zero, assuming that only optimal management can support that species.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

management actions

The mean vegetation height increased as the management intensity decreased: 7·8 cm (± SE 0·6 cm, n = 29) for intensively managed sites; 28·6 cm (±1·2 cm, n = 13) for extensively managed sites, and 39·1 cm (±2·3 cm, n = 25) for abandoned sites. The average abundance of nectar plants was lowest for intensively managed sites (nectar plant abundance index 5·16 ± 0·37, n = 29), increased with extensive management (7·54 ± 0·30, n = 13) and decreased again at sites that were no longer grazed (6·32 ± 0·40, n = 25). Both the mean vegetation height and the nectar plant abundance differed significantly between the management intensity levels (P < 0·001, one-way anova).

habitat modelling

We were able to model 13 plant species and 16 Lepidoptera species successfully (Table 1). Selected model variables and calibration statistics are provided in Supplementary Material Table S1. The predicted responses of species to the different management actions varied widely (Fig. 1). Nineteen species were omitted from the conservation planning analysis for one of the following reasons:

Table 1.  Species grouped according to indicator type (plants) and population trend (Lepidoptera), for all species and successfully modelled species separately. The number of modelled species predicted to prefer each of the grazing actions is also provided
 Indicator type/trendNo. of species availableNo. of species modelledModel-predicted management preference
Annual grazingBiennial grazingNo grazing
PlantsPositive50
Neutral147412
Negative866
LepidopteraDeclining411
Stable108152
Increasing7752
Total 482951212
image

Figure 1. Predicted responses of four species to grazing management. Responses are scaled as a proportion of the maximum attainable representation in the 45 sites (i.e. the responses of the best management type sum up to one). Note that y-axes have different scales.

Download figure to PowerPoint

  • 1
    The variable ‘vegetation height’ or ‘management intensity’ was not selected: these species were predicted to be insensitive to different management options (two plant species).
  • 2
    No biological interpretation for the modelled relationship between abundance and the variable ‘vegetation height’ or ‘management intensity’ could be provided (four plant species).
  • 3
    Weak or negative cross-validated correlation (< 0·1), which indicates a weak or overfitted model, rendered predicted values unreliable (eight plant species and five Lepidoptera species).

Among the species that could not be successfully modelled were all positive indicator plant species and three out of four declining Lepidoptera species (Table 1). Hence, it was not possible to test the effect of prioritizing these groups in the conservation planning analysis.

action–selection results

Budget

First, we will describe the scenario where all species were assigned equal weights. Without a budget, none of the sites could be grazed (Fig. 2d), resulting in an average representation of species of 0·69 (Fig. 2a). For the maximum available budget, 36 out of 45 sites (80%) were assigned to biennial grazing instead of no management, with an average species representation of 0·78 and solution cost EUR 31 374. Although this increase in average representation appears minimal in view of the amount of money spent, the relative representation of different species groups changed considerably between smaller and larger budgets (Fig. 2a). As expected, the mean representation of those species which prefer grazing improved, especially for those species preferring biennial management. As a consequence, the species group preferring no grazing decreased in average representation; however, this group still had an average representation of 0·75, which is higher than the average representation of those species which prefer annual grazing (0·51).

image

Figure 2. Results of site–action selection for increasing budget, optimized for all species, with increasing weights for species preferring annual grazing, weight for other species = 1. Panels (a–c) show the average species representation for all species (solid line), and for species grouped per management preference (other lines). Panels (d–f) show the proportion of sites that is assigned to each management action. Weights of species preferring annual grazing equalled one in panels (a, d), five in panels (b, e) and 40 in panels (c, f).

Download figure to PowerPoint

Prioritizing species

We tested the effect of applying increasing weights to particular species groups using the group of species that prefer annual grazing. This group contains the fewest species and annual grazing is the most expensive management action. This action is thus the least preferred for selection by the algorithm, resulting in lower representation levels for species which depend on this management action. We tested how much representation was lost for other groups when the weight assigned to the group of species preferring annual grazing was increased from one to 40 (Fig. 2a–c). The results were equal for weights of 40 and larger. Increasing the weight for this group of species promoted the selection of annual grazing management (budget permitting, Fig. 2d–f). With an unlimited budget, a weight of 5 was required to increase average representation to above 0·7 for species preferring annual grazing (0·72). Doing this caused only a minor reduction in the representation for species preferring biennial grazing (representation 0·78) and species preferring no grazing still obtained a representation of 0·53 on average. Higher weights and larger budgets increased the number of sites assigned to annual grazing even further (Fig. 2f), yielding the maximum benefit for those species preferring this type of management, but at the cost of the representation of species with other preferences (Fig. 2c). For those species preferring biennial grazing, both the annual grazing and no grazing options are suboptimal management. In this analysis, weights of over 15 caused the management assigned to sites to switch from no grazing to annual grazing, as soon as the budget allowed (see Fig. 2f). This switch decreased the representation of species preferring biennial grazing from 0·53 to 0·30 on average (Fig. 2c). This decrease indicates that most of these species are predicted to have higher representation levels under no grazing than under annual grazing (see Ochlodes sylvanus in Fig. 1). Indeed, for eight out of 12 species which prefer biennial grazing, the no-grazing option generated intermediate levels of representation, which explains the decreasing mean trend in Fig. 2c.

Some of the plant species in the data set are negative indicators for semi-natural grasslands (Pykälä 2001). We therefore tested the effect of decreasing the weights assigned to these species from 0 to –10 (results were equal for weights of –10 and smaller). Applying a weight of zero for these species (which all prefer no grazing; see Table 1) caused the algorithm to select biannual grazing rather than no grazing (budget permitting, Fig. 3d). At still lower weights, annual grazing became more beneficial (Fig. 3e–f), as it kept the representation of species preferring no grazing at the lowest level (as in Fig. 1, Polypogon tentacularius). The optimization results confirmed the expectation of a strong preference for annual grazing management, yielding results comparable to those where annual species received higher weights (Fig. 2).

image

Figure 3. Results of site–action selection for increasing budget, optimized for all species, with decreasing weights for negative indicator plants, weight of other species = 1. Panels (a–c) show the average representation of species (legend as in Fig. 2a–c). Panels (d–f) show the proportion of sites that is assigned to each management action (legend as in Fig. 2d–f). Weights of negative indicator plants equalled zero in panels (a, d), –5 in panels (b, e) and –10 in panels (c, f).

Download figure to PowerPoint

Sensitivity to suboptimal management

The assumption of zero representation for suboptimal management resulted in the allocation of sites to each of the three management actions (Fig. 4c). The average representation dropped as a consequence of this assumption (Fig. 4a as compared to Fig. 2a), indicating that species were predicted to benefit from suboptimal management (see also Fig. 1). When in addition the weight for negative indicator plant species (which all prefer no grazing) was set to zero, the number of sites assigned to no grazing was reduced, in favour of the number of sites assigned to biennial and annual grazing (Fig. 4d). This resulted directly in an improved representation for the two species groups that preferred grazing management (Fig. 4b as compared to 4a).

image

Figure 4. Results of site–action selection for increasing budget, optimized for all species, with species representation for suboptimal management set to zero. Panels (a, b) show the average representation of species (legend as in Fig. 2a–c). Panels (c,d) show the proportion of sites that is assigned to each management action (legend as in Fig. 2d–f). Panels (a, c): all species equal weight. Panels (b, d): Weight of negative indicator plants = 0; Weight of all other species = 1.

Download figure to PowerPoint

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

In this study, we have demonstrated that the optimal allocation of management actions to sites for the purpose of species conservation is by no means straightforward. When species have contrasting requirements, trade-offs between the different species are inevitable. We have shown that the strength of the trade-offs depends on the requirements that the species have, the relative priority assigned to species, and the budget available. Those species requiring annually grazed grassland were the most sensitive to the management applied; they had an average representation of only 0·34 when no budget was available, and thus, none of the sites could be grazed (compared to a representation of 1·0 when all sites were grazed annually).

The tolerance of the species to suboptimal management varied (Fig. 1). Whereas some species showed very steep changes in the predicted abundance between different management strategies (e.g. Trifolium repens, Fig. 1), others showed only slight responses to changes in management (e.g. Hypericum maculatum, Fig. 1). The species which were predicted to be the most sensitive were strong drivers of the optimal solution because they would benefit less from suboptimal management.

Although the models predicted the species to be present under suboptimal management, it remains to be seen to what extent these species would in fact persist under suboptimal habitat conditions. It is possible that species are still present at a site, although the conditions for that species to reproduce successfully are no longer present. Since butterflies and moths have short generation times, they are expected to respond relatively quickly when conditions become unsuitable (although they may be found in suboptimal sites due to source–sink dynamics). In plants, by contrast, adult individuals and seed banks can survive in suboptimal conditions, although the seeds might be incapable of germinating. As a result, a species can be recorded as present although the conditions are no longer suitable for species persistence, a phenomenon known as the ‘extinction debt’ (Tilman et al. 1994; Hanski & Ovaskainen 2002). By setting the expected representation levels of species to zero for suboptimal management, we assumed that species could not survive under suboptimal management. As a result, it became optimal to select each management action a given number of times, and species representation of the different preference groups became more directly related to the number of sites managed in the preferred way (Figs 2 and 4). Omitting species with low conservation priorities (negative indicator plant species) from the analysis allowed the needs of more sensitive species to be addressed more effectively (Fig. 4).

We have demonstrated how weighting can be used to increase (or decrease) the emphasis on particular species. In our analysis, the results were very sensitive to the weights applied and the budget available, which highlights the importance of investigating the sensitivity of the results to such decisions. Those species which prefer annual grazing required larger weights in order to improve their representation level. This had several causes:

  • 1
    The set of modelled species contained only five species with this preference, compared to 12 species for both biennial and no grazing (Table 1).
  • 2
    Annual grazing was the most costly option, and was therefore the least preferred when optimizing for cost-effectiveness.
  • 3
    Species with a preference for biennial grazing had mostly no grazing as the second-best option (eight species), rather than annual grazing (two species).

Consequently, annual grazing was not automatically selected. Most of the priority grassland species (declining Lepidoptera and positive indicator plants) were expected to prefer more intensive grazing (Pöyry et al. 2005; Pykälä 2005), but could not be modelled satisfactory (see below). The negative indicator plants, by contrast, were expected and predicted to prefer ‘no grazing’ (Table 1; Pykälä 2005). Due to the abandonment of traditional agricultural practices, mesic semi-natural grasslands are disappearing in Finland and the quality of the remaining grasslands is deteriorating (Luoto et al. 2003). In this context, our analysis suggests that investment in the maintenance of semi-natural mesic grasslands in the form of different grazing regimes seems justified in order to conserve priority species, as well as to control the numbers of negative indicator species.

In this case study, we were unable to model 19 out of 48 species successfully, and for this reason they were omitted from our analysis. Among these species were eight out of nine species of conservation concern (Lepidoptera with declining distributions and positive indicator plant species). Our analysis demonstrated that the composition of the set of species to which action–selection was applied had a considerable influence on the outcome. In real-world conservation applications, it is therefore undesirable to omit species of conservation interest from the analysis. For species with too few data to model them individually, it may be possible to collect additional empirical or expert data, or use other modelling techniques such as community modelling (Elith et al. 2006; Ferrier & Guisan 2006). A combination of different approaches could be used to assess the robustness of species predicted responses to conservation actions, but this was beyond the scope of our study.

Although our study exemplifies many conservation planning problems, it is among the first to combine contrasting species requirements, multiple available actions and action cost simultaneously in an optimization framework. Potential applications of the approach we describe are found in planning for habitat preservation, restoration and maintenance (Drechsler et al. 2006; Van Teeffelen 2007), forest harvesting (Hof et al. 1994; Bevers et al. 1995), landscape configuration planning (Holzkämper, Lausch & Seppelt 2006) and invasive species management.

This study demonstrates that planning conservation management for species with contrasting requirements is far from straightforward, and that careful consideration should be given to trade-offs between species. We have shown how the optimization tool used in this study can be employed to investigate the strength of trade-offs and guide conservation decision-making. This can produce a set of robust and cost-effective conservation actions, maximizing representation for those species with conservation priority. We believe that our approach could be applied effectively to many conservation planning problems in management, restoration and protection.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

We thank Juha Pykälä for providing a part of the plant data. We are grateful to Risto Heikkinen, Varpu Mitikka, Mark Burgman and two anonymous reviewers for providing valuable comments to an earlier version of the manuscript. The authors were financially supported by the Academy of Finland, project no. 202870 (A.v.T) and project no. 209017 (M.C.), the Finnish Cultural Foundation (J.P.) and the Maj and Tor Nessling Foundation (K.R.). Collection of plant and butterfly/moth data was funded by the Finnish Ministry of Environment (for the project ‘Maintaining biodiversity in traditional rural landscapes – optimal management and area networks’ through the Finnish Biodiversity Research Programme FIBRE coordinated by the Academy of Finland).

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information
  • Arponen, A., Heikkinen, R.K., Thomas, C.D. & Moilanen, A. (2005) The value of biodiversity in reserve selection: representation, species weighting, and benefit functions. Conservation Biology, 19, 20092014.
  • Bevers, M., Hof, J., Kent, B. & Raphael, M.G. (1995) Sustainable forest management for optimizing multispecies wildlife habitat: a coastal Douglas-fir example. Natural Resource Modeling, 9, 123.
  • Cabeza, M. & Moilanen, A. (2001) Design of reserve networks and the persistence of biodiversity. Trends in Ecology & Evolution, 16, 242248.
  • Dobson, A.P., Bradshaw, A.D. & Baker, A.J.M. (1997) Hopes for the future: restoration ecology and conservation biology. Science, 277, 515522.
  • Drechsler, M., Johst, K., Wätzold, F. & Westphal, M.I. (2006) Integrating economic cost into the analysis of flexible conservation management strategies. Ecological Applications, 16, 19591966.
  • Elith, J., Graham, C.H., Anderson, R.P., Dudik, M., Ferrier, S., Guisan, A., Hijmans, R.J., Huettmann, F., Leathwick, J.R., Lehmann, A., Li, J., Lohmann, L.G., Loiselle, B.A., Manion, G., Moritz, C., Nakamura, M., Nakazawa, Y., Overton, J.M., Peterson, A.T., Phillips, S.J., Richardson, K., Scachetti-Pereira, R., Schapire, R.E., Soberon, J., Williams, S., Wisz, M.S. & Zimmermann, N.E. (2006) Novel methods improve prediction of species’ distributions from occurrence data. Ecography, 29, 129151.
  • EC (2007) LIFE-Nature. European Commission, Brussels. Available from http://ec.europa.eu/environment/life/funding/life3/index.htm (accessed April 2008).
  • Ferrier, S. & Guisan, A. (2006) Spatial modelling of biodiversity at the community level. Journal of Applied Ecology, 43, 393404.
  • Hanski, I. & Ovaskainen, O. (2002) Extinction debt at extinction threshold. Conservation Biology, 16, 666673.
  • Hastie, T. & Tibshirani, R. (1990) Generalized Additive Models. Chapman & Hall, London.
  • Hof, J., Bevers, M., Joyce, L. & Kent, B. (1994) An integer programming approach for spatially and temporally optimizing wildlife populations. Forest Science, 40, 177191.
  • Holzkämper, A., Lausch, A. & Seppelt, R. (2006) Optimizing landscape configuration to enhance habitat suitability for species with contrasting habitat requirements. Ecological Modelling, 198, 277292.
  • Huldén, L., Albrecht, A., Itämies, J., Malinen, P. & Wettenhovi, J. (2000) Suomen Suurperhosatlas – Finlands storfjärilsatlas. Lepidopterological Society of Finland, Finnish Museum of Natural History, Helsinki, Finland.
  • Kleijn, D. & Sutherland, W.J. (2003) How effective are European agri-environment schemes in conserving and promoting biodiversity? Journal of Applied Ecology, 40, 947969.
  • Kuussaari, M., Heliölä, J., Pöyry, J. & Saarinen, K. (2007) Contrasting trends of butterfly species preferring semi-natural grasslands, field margins and forest edges in northern Europe. Journal of Insect Conservation, 11, 351366
  • Lehmann, A., Overton, J.M. & Leathwick, J.R. (2002) grasp: generalized regression analysis and spatial prediction. Ecological Modelling, 157, 189207.
  • Luoto, M., Rekolainen, S., Aakkula, J. & Pykälä, J. (2003) Loss of plant species richness and habitat connectivity in grasslands associated with agricultural change in Finland. Ambio, 32, 447452.
  • Margules, C.R. & Sarkar, S. (2007) Systematic Conservation Planning. Cambridge University Press, Cambridge, UK.
  • Newbold, S. & Eadie, J.M. (2004) Using species-habitat models to target conservation: a case study with breeding mallards. Ecological Applications, 14, 13841393.
  • Pöyry, J., Lindgren, S., Salminen, J. & Kuussaari, M. (2004) Restoration of butterfly and moth communities in semi-natural grasslands by cattle grazing. Ecological Applications, 14, 16561670.
  • Pöyry, J., Lindgren, S., Salminen, J. & Kuussaari, M. (2005) Responses of butterfly and moth species to restored cattle grazing in semi-natural grasslands. Biological Conservation, 122, 465478.
  • Pykälä, J. (2001) Maintaining biodiversity through traditional animal husbandry. The Finnish Environment, 495, 1205. (in Finnish with English summary).
  • Pykälä, J. (2004) Cattle grazing increases plant species richness of most species trait groups in mesic semi-natural grasslands. Plant Ecology, 175, 217226.
  • Pykälä, J. (2005) Plant species responses to cattle grazing in mesic semi-natural grassland. Agriculture, Ecosystems & Environment, 108, 109117.
  • Raatikainen, K., Heikkinen, R. & Pykälä, J. (2007) Impacts of local and regional factors on vegetation of boreal semi-natural grasslands. Plant Ecology, 189, 155173.
  • Schulman, A., Heliölä, J. & Pykälä, J. (2006) Quality and management of contract areas of the agri-environment support – the special support for management of traditional rural biotopes and for enchantment of biodiversity. The Finnish Environment, 3, 187 (in Finnish with English summary).
  • S-PLUS 6.1 Release 1 (2002), Insightful Corp., Seattle, WA, USA.
  • Stewart, K.E.J., Bourn, N.A.D. & Thomas, J.A. (2001) An evaluation of three quick methods commonly used to assess sward height in ecology. Journal of Applied Ecology, 38, 11481154.
  • Tilman, D., May, R.M., Lehman, C.L. & Nowak, M.A. (1994) Habitat destruction and the extinction debt. Nature, 371, 6566.
  • Van Teeffelen, A.J.A. & Moilanen, A. (2008) Where and how to manage: optimal selection of conservation actions for multiple species. Biodiversity Informatics, 5, 113.
  • Van Teeffelen, A.J.A. (2007) Where and how to conserve: extending the scope of spatial reserve network design. PhD Thesis, University of Helsinki, Helsinki, Finland.
  • Westphal, M.I., Field, S.A. & Possingham, H. (2007) Optimizing landscape configuration: a case study of woodland birds in the Mount Lofty Ranges, South Australia. Landscape and Urban Planning, 81, 5666.
  • Whittingham, M.J. (2007) Will agri-environment schemes deliver substantial biodiversity gain, and if not why not? Journal of Applied Ecology, 44, 15.
  • Wilson, K.A., Underwood, E.C., Morrison, S.A., Klausmeyer, K.R., Murdoch, W.W., Reyers, B., Wardell-Johnson, G., Marquet, P.A., Rundel, P.W., McBride, M.F., Pressey, R.L., Bode, M., Hoekstra, J.M., Andelman, S., Looker, M., Rondinini, C., Kareiva, P., Shaw, M.R. & Possingham, H.P. (2007) Conserving biodiversity efficiently: what to do, where, and when. PLoS Biology, 5, e223.

Supporting Information

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  9. Supporting Information

Table S1. List of species used in the analysis and their habitat modelling statistics

Please note: Blackwell Publishing are not responsible for the content or functionality of any supplementary materials supplied by the authors. Any queries (other than missing material) should be directed to the corresponding author for the article.

FilenameFormatSizeDescription
JPE_1514_sm_TableS1.pdf21KSupporting info item

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.