• Open Access

Spatial prioritization of conservation management

Authors


  • Editor
    Ana Rodrigues

Atte Moilanen, Department of Biosciences, Finnish Centre of Excellence in Metapopulation Biology, PO Box 65, FI-00014 University of Helsinki, Finland. Tel: +358 9 191 57753. Email: atte.moilanen@helsinki.fi

Abstract

We develop a high-resolution conservation prioritization analysis for New Zealand's rivers and streams that simultaneously consider both the present state (representation) of ecosystems, and the prioritization of management actions designed to mitigate ongoing human impacts on their expected future state (retention). As input we used information about the geographic distributions of river ecosystem groups and their compositional similarity, species richness, present condition as compared to their estimated pristine state, and upstream and downstream connectivity. Candidate management actions included riparian planting, establishment of wetlands on tile-drain outflows, and use of riparian buffer strips in plantation forests. The analysis, carried out at a 1-ha resolution for a study area of 22,000 km2 in Southland, New Zealand, demonstrates a credible range of options for management intervention, particularly in lowland streams under serious threat from agricultural intensification. The proposed analysis can be replicated elsewhere for terrestrial, freshwater, or marine systems using publicly available software.

Introduction

Although traditional approaches to conservation prioritization have focused on the analysis of present-day biodiversity pattern to find optimal sets of sites for protection, there is a growing realization of the need to also prioritize other kinds of conservation action (Pressey et al. 2004, 2007; Wilson et al. 2006, 2007). These might include habitat management to halt ongoing decline of ecosystems or restoration to improve ecosystem condition from a present partially degraded state (Hobbs 2007). Ideally, conservation should be focused on the difference that conservation actions make to biodiversity features: the idea is to either maximize conservation gains from management, or to minimize impacts when losses are inevitable. Habitat protection can be interpreted via the difference made as well: protection can remove threats that would otherwise lead to the decline of a presently high-quality area. However, focusing solely on the estimated difference made by conservation action may be dangerous, for example, if information about threats is incomplete (O’Hanley et al. 2007; Visconti et al. 2010), as it almost inevitably is. A natural or seminatural area of high intrinsic conservation value might be left without protection if there are no known threats that could impact the area in the near future. The area could, however, be impacted in the future by threats presently unrecognized. Consequently, some kind of balance needs to be found between focus on places with highest present-day natural values, and focus on areas where most cost-effective management or restoration can be implemented or greatest losses avoided.

Several studies have addressed optimal or near-optimal allocation of habitat management or restoration (Bryan & Crossman 2008; Lethbridge et al. 2010; Newbold 2005; Stralberg et al. 2009; Thomson et al. 2009). Here, we extend the traditional use of spatial prioritization tools for prioritizing protection by adding functionality that allows explicit balancing of present occurrence of biodiversity features (representation) against their future expected state conditional on the difference made by management action or the lack of it (retention). We demonstrate our approach at a high resolution (1 ha) across an extensive study area in southern New Zealand, using the Zonation approach. (Moilanen 2007; Moilanen et al. 2005, 2009, 2011) to analyze spatially explicit data layers describing variation in riverine ecosystem character, species richness, river condition, and expected change in those conditions resulting from the implementation of realistic catchment-level management (Allan 2004). We also include the use of small subcatchment-based planning units and connectivity settings to account for the inherent longitudinal connectivity of river systems (Hermoso et al. 2011; Moilanen et al. 2008). While all of these components may have been demonstrated individually in previous studies, our combining of them here demonstrates a substantial advance in the level of management-relevant analysis that can be feasibly implemented with publicly available conservation prioritization tools. An analysis similar to the one we propose could be implemented in other conservation planning software, including Marxan (Possingham et al. 2000), Marxan with Zones (Watts et al. 2009), ConsNet (Ciarleglio et al. 2009), and C-Plan (Pressey et al. 2009).

Methods

Balancing representation and retention in spatial prioritization

Zonation is a framework and software for spatial conservation prioritization that primarily operates on sets of raster grids that map the occurrence levels of biodiversity features across the landscape. In this conventional role, Zonation produces both a priority rank map for the entire landscape and a set of performance curves that describe the quality of the solution in absolute terms. The operational principle of Zonation can be summarized as iterative maximization of retention of weighted range-size normalized feature richness (Moilanen et al. 2011). Relevant details of Zonation have been summarized elsewhere (Moilanen 2007; Moilanen et al. 2005, 2009, 2011). Presently, our focus is on a combined analysis of representation and retention, which we achieve using duplicate sets of spatial feature layers as follows.

The first set of layers is the standard data that are used in systematic conservation planning or spatial conservation prioritization: it describes the current spatial distribution of each biodiversity feature and is used to prioritize representation. For these layers it is implicitly assumed that all conservation value is lost if conservation is not applied at a location. The second set of layers (for the same features or their subset) is constructed according to the analysis purpose; they describe the positive difference that management at each site would deliver, compared to what is expected without management. This second set of layers has nonzero values only where conservation action would make a difference. Combining present distributions and difference made layers allows the simultaneous prioritization of representation and retention (Figure 1). The balance between these two considerations is adjusted simply by changing the relative weights given to the two layer sets described above. This high-level strategy could be implemented also in other planning approaches than Zonation.

Figure 1.

Key components of the analysis.

Study area

Our analysis (Figure 1) demonstrates the joint prioritization of representation and management actions for an area of approximately 22,000 km2, located in the southern South Island of New Zealand. The environment there is dominated by a steep gradient of decreasing rainfall with progression away from the Fiordland mountains in the west to drier, lower elevation country to the east and south. Extensive alluvial plains occur in central and southern parts of the region, with low relief hill country dominating in the south-east. Pasture is the dominant land cover (45%), followed by indigenous forest (23%) and tussock grasslands (16%); plantation forests comprise 3% of the total area. Approximately 32,000 km of rivers and streams occur in the region. Around 34% of the region is designated as public conservation lands, predominantly in high elevation and/or steep parts of the landscape not suited to “productive” land uses such as agriculture or forestry. Agricultural activity in the region traditionally focused on a mix of cropping, and raising of sheep and beef cattle, but dairy farming has expanded rapidly with cow numbers rising from c. 38,000 to c. 600,000 over the last 20 years (Statistics New Zealand). Dairy farms occupied9.14% of all land by 1994, with significant negative impacts on water quality in rivers and streams (Environment Southland 2009).

Input data

Data describing the river ecosystems of this study area were derived from a GIS network topology representing New Zealand's river network (Table 1; Snelder & Biggs 2002). The importance of local-scale river connections was recognized by grouping river segments into 1921 catchment-based planning units (Allan et al. 1997), constructed by dividing all fourth or higher order catchments into their third (or higher) order subcatchments, and their main stems. Streams of first- to third-order were treated as individual planning units.

Table 1.  Main data inputs to the analysis
Input itemDerivation
Third-order planning unitsSubdivision of all fourth and higher order rivers into their third-order subcatchments and their main stem—additional information describes longitudinal flow linkages from upstream to downstream planning units
River ecosystemsTaken from the 200-ecosystem classification of New Zealand rivers and streams of Leathwick et al. (2011); 129 of these occur in Southland
River conditionEstimates of ecological condition for each river segment based on effects of catchment clearance, nitrogen inputs, introduced fish, mining, dams, industrial discharges, and roading (Leathwick et al. 2010)
Management differenceEstimates of the difference made by management given remedial action to mitigate the effects of sediment and nutrient discharges
Management costsEstimated cost of management calculated for each planning unit, based on its expected mix of land uses—see Table 2

Representation layers for the Zonation analyses were constructed from a multivariate environmental classification (generalized dissimilarity modeling) designed to capture variation in underlying biological patterns (Leathwick et al. 2011). One grid layer (100-m cell resolution) was used for each of 129 river ecosystem types (Figure 2B, Table 1), with each layer modified to take account of its similarity to all other ecosystem types as described in Leathwick et al. (2010). An additional layer provided spatially explicit estimates of current landscape biodiversity condition (Figure 2C, Leathwick et al. 2010) derived from an expert-system approach taking account of the likely effects of seven human-induced pressures on in-stream biodiversity condition, that is, catchment clearance of native vegetation, diffuse inputs of nitrogen from agriculture, introduced fish, urbanization, mines, dams, and industrial discharges.

Figure 2.

Input data for the Southland study area. (A) Topographic relief and distribution of public conservation lands; (B) river classification groups—colors were generated so that red, green, and blue components of a pixel were scaled by the first three principal components of environmental factors determining similarity of species composition between classification groups (Belbin et al. 1983); (C) average estimated biodiversity condition for planning units, 1.0 equals pristine and 0.0 completely degraded habitat; (D) average difference made for planning units; (E) estimated cost (NZ$) of management actions per 100 m of river length calculated at a planning unit level. A large lake in the west is shown in light blue in all panels.

Our management/retention feature layers described the estimated difference made to current biodiversity condition for each of the river ecosystems (Figure 2D), given a range of management actions. These were allocated planning unit by planning unit, taking consideration of the relative proportion of their extent occupied by intensive (dairy) or extensive (dry stock) pastoral farming, and plantation forests, as indicated by varying combinations of New Zealand's Land Cover Database (LCDB2: Ministry for the Environment 2004) and land use capability classification (LUC: National Water and Soil Conservation Authority (NWASCA), 1979) maps. Intensive dairy farming was assumed to be the dominant land-use on all pasture land cover occupying LUC classes I–III, and dry-stock farming was assumed to be the dominant use on all pasture occupying LUC classes IV–VII. Plantation forests were identified using the relevant LCDB2 class, irrespective of the LUC.

For dairy and dry stock farming our management interventions were aimed at reducing direct stock access to water-ways, restoring riparian vegetation cover to increase shading and reduce sediment and nutrient inputs, and establishing artificial wetlands on the discharge points of the tile drains that are used extensively on heavy soils in this region (Table 2). The action specified for plantation forests was the use of 5 m riparian protection zones or setbacks either side of all streams. Other actions of potential benefit to native biodiversity, such as the removal of large dams or reducing the populations of introduced fish species that prey on and/or compete with native species, were not considered given their low operational or social feasibility.

Table 2.  Management actions considered for the three landuses
LanduseActionCostCombined impact
DairyingThree-wire electric fencing$5/m fencing costs, or $1,000/100 m of stream length60% reduction in diffuse nitrogen inputs
Riparian planting 5 m either side of stream, including three years maintenance$20, 500/ha, or $2,050/ 100 m of stream lengthEffective increase in native catchment cover of 33%
Establish artificial wetland on tile drain discharges$550/ha applied across 1% of those parts of the planning unit in dairy productionEffective increase in native catchment cover of 33%
Dry-stockFive-wire fencing$18/m fencing costs, or $3,600/100 m of stream length30% reduction in diffuse nitrogen inputs
Riparian planting 5 m either side of stream, including three years maintenance$20, 500/ha, or $2,050/ 100 m of stream lengthEffective increase in native catchment cover of 33%
Plantation forest5 m set back from both sides of streams for all operationsAssumed opportunity cost of $5,000/ha or $500/100 m of stream length30% reduction in diffuse nitrogen inputs
  Effective increase in native catchment cover of 33%

Relative weights applied to the individual ecosystem layers in the representation set were calculated by averaging landscape-scale predictions of species richness for native fish species and macroinvertebrates (Leathwick et al. 2010); weights for the management/retention layers were calculated as described below, to reflect relative differences made by management actions, and to control for balance of representation and retention. One further layer described spatial variation in costs (Figure 2E); because management was allowed in only parts of the landscape, this was calculated as the ratio of expected change in condition divided by the expected cost and was treated as an additional, positively weighted feature layer in some of the Zonation analyses.

Zonation weights

Controlling the relative influence of representation versus management/retention layers is a key component of our approach; this was implemented as follows. Zonation allows for the weighting of individual layers to control their degree of influence on ranking outcomes, for example, weights could be set higher for endemic species or for ecosystems with high species richness (Leathwick et al. 2010). To begin, weights are set for the representation layers, denoting these as wj for environment j. We initially considered assigning the same set of weights to the management/retention layers (wrj), but this option was rejected for an important reason: features may differ markedly in their degree of change in condition consequent on management. We therefore construct weights for the management/retention layers to reflect both the weights applied to the respective representation layers, and the magnitude of changes resulting from management, with the latter expressed relative to the current condition (Figure 3).

Figure 3.

Factors relevant for the calculation of retention layer weights. There are two different cases in the weight calculation, corresponding to whether conservation action prevents further habitat degradation (A) or results in habitat quality improving from present (B). The weight of the retention layer is low when there is a small fractional loss predicted for the feature even in the absence of conservation action (C). In this case, the landscape-scale effect of management is minor. The retention layer weight becomes high when expected loss in the absence of action is a high fraction of what remains presently (D). The y-axis on the bars is in terms of fraction of the normalized distribution of the feature across the landscape.

When the goal is to prioritize management actions according to their expected gains (Figure 3B), the expected condition after management (rj) is higher than the current condition (cj), and higher weights should be allocated to those biodiversity features benefitting proportionally most from management (Figure 3D). In this mode, which we refer to as “management gain,” we set weights for the individual retention layers as

image

When management brings about large gains for a feature that is otherwise degraded, the management/retention layer weight increases accordingly, but when only a small relative gain is expected the weight becomes small (Figure 3C). Parameter β controls the balance between the representation and management/retention sets of layers; increasing β increases the emphasis given to management/retention.

Alternatively, final condition estimates (rj) may be lower than the present condition (cj), rj <cj, so that the prioritization accounts for expected future losses that can be prevented by protection (Figure 3A). In this mode, “loss prevention,” we set

image

Now, highest weights are given to those features for which a large relative loss is expected in the absence of protection. Note that in both cases, cj and rj are measured for the full landscape by summing across grid cells, and are expressed as proportions of the expected occurrences under some baseline landscape state (cj= 1), which can be the present state or some historical (near-pristine) reference state.

Analysis variants

Eight analyses were carried out applying standard Zonation methods (Moilanen 2007; Moilanen et al. 2005, 2009) on the analysis setup described in Figure 1. The neighborhood quality penalty technique (Moilanen et al. 2008) was used for all analyses to account for longitudinal connectivity along rivers—see Leathwick et al. (2010) for details. Other settings for these analyses varied as follows: (1) “representation only”—representation layers were individually weighted as described above, and the retention layers were included with a weight of zero, preventing them influencing prioritization outcomes but allowing their treatment to be assessed; (2)–(4) “representation and retention,” with weights for the retention layers calculated using values for β of 20, 50, and 100, respectively; (5) “retention only” in which the representation layers were rendered passive by giving them a weight of zero, and the retention layers were weighted as above, (6)–(8) “representation, retention, and costs,” in which β= 20, and the benefit/cost layer was included with weightings of 1, 2, and 5, respectively.

Results

The initial “representation only” analysis indicates those places on the landscape giving maximal representation of riverine ecosystems for any degree of geographic extent under protection, subject to spatial variation in current condition, and the desirability of maintaining longitudinal connectivity. The top 10% of sites (Figure 4A) are scattered throughout the landscape, although with a bias toward planning units in hilly terrain that are generally in better condition than those on the low-relief alluvial plains where biodiversity values have been heavily impacted by intensive land-uses.

Figure 4.

Landscape rankings for four scenarios derived by Zonation analyses using differing combinations of representation and retention layers. Map values indicate the top percentage of the landscape that the spatial units (catchments) belong to. Thus, category 0.0–10.0 indicates the top 10% of the landscape. (A) Representation layers alone; (B) and (C) representation and retention layers with weights for the latter calculated using a values for β of 20 and 50, respectively; (D) representation, retention (β= 20), and cost/benefit layer, with the latter having a weight of 2. For (D), selected groups of planning units showing either marked increases or decreases in priority compared to equivalent rankings without cost information (B) are indicated by a “+” and “-,” respectively.

Inclusion of retention layers into prioritization alters both the spatial pattern of high priority units and the average protection of ecosystems, with the magnitude of this effect varying according to the balance required between representation and retention, as defined via the value of parameter β. Using β= 20 gives a noticeable increase in the priority given to lowland planning units for which management is prescribed (Figure 4B); the resulting shift in emphasis also decreases slightly the average representation of ecosystems achieved at intermediate levels of geographic protection (Figure 5A). Increasing β to 50 (Figure 4C) further magnifies this effect, increasing both the priority given to lowland planning units and the unevenness of ecosystem representation delivered for any given level of geographic protection.

Figure 5.

Performance as a function of geographic protection of Zonation analyses using differing combinations of representation, retention, and cost layers. (A) Average ecosystem protection when using representation and retention layers; (B) average implementation of management when using representation and retention layers; (C) average ecosystem protection when using representation, retention, and cost layers; (D) budget expenditure when using representation, retention, and cost layers.

Varying the value of β also alters strongly the degree of management implementation implied by differing degrees of geographic protection (Figure 5B). The “representation only” analysis would result in the least management implementation; for example, only 12.5% of all proposed actions would be implemented within the top-ranked 20% of the landscape (Figure 5B, black line). However, this proportion steadily increases as the influence of the retention layers on the rankings is increased (Figure 5B); 19.7% of management actions would be implemented in the top 20% of the landscape when β= 20, increasing to 28.1% when β= 50, and to 31.3% when β= 100. The “retention only” analysis only considers the difference made layers in ranking, and results in 32% of management actions being implemented across the top 20% of the landscape.

Inclusion of cost information to the ranking process further alters the priority ranking (Figure 4D), the average levels of protection achieved for any given level of geographic protection (Figure 5C), and the budget required to implement the required management actions (Figure 5D). However, the magnitude of all of these changes is sensitive to the relative weighting given to the benefit/cost layer. Fixing β= 20, changes in average ecosystem protection are relatively muted when the benefit/cost layer weight is increased from 1 to 2, but there already is a marked decline when a weight of 5 is used, reflecting an increasing bias in selection toward planning units in hilly terrain where actions are less expensive, but producing gains for only some particular ecosystems.

Discussion

Our results provide an operational-scale demonstration of a novel analysis that allows ranking of potential biodiversity management projects across a landscape while taking account of variation both in project costs and the need to evenly spread management actions across a representative range of biodiversity values, in our case riverine ecosystems. This approach has particular value when prioritizing management projects for which there are marked differences in the costs between ecosystems. Here, an emphasis on cost-effectiveness alone might appear to maximize biodiversity gains, but those gains would not represent the full range of biodiversity assets. Our approach, with its easily specified tuning parameters, allows tradeoffs between potentially conflicting objectives to be explored at a landscape scale, providing explicit information about the decreased cost-effectiveness of management when representativeness-related considerations prevail, versus the loss of representativeness that occurs when greater emphasis is placed on maximizing cost-effectiveness.

Our results demonstrate a credible range of options for management intervention on the Southland lowlands with scenarios using intermediate weightings delivering outcomes that have minimal reductions in representativeness, while clearly identifying the most cost-effective projects for a range of different ecosystem types, including those in lowland hill-country and on the lowland alluvial plains. We emphasize that connectivity, condition, and retention all had significant effects on the analysis outcome, implying that these components might be highly relevant for similar assessments elsewhere. Also, given the up- and downriver connectivity responses applied, there was a tendency for our analyses to prioritize management at lowland and midelevation locations that are connected to high-quality headwater catchments of rivers that host environments that are rare and/or have elsewhere declined significantly in condition. Overall, we regard the β= 20, cost = 2 solution (Figure 5D) as striking the most sensible balance between achieving good representation, protecting areas in best condition, and delivering cost-effective management.

While our proposed method is applicable to any environment, including terrestrial planning, our application here concerned freshwater conservation. Freshwater systems themselves supply substantial ecosystem services (Schroter et al. 2005), but are globally impacted by multiple anthropogenic pressures (Abell et al. 2007). Continuing declines in the condition of freshwater systems demonstrate a need for increased conservation measures (Dudgeon et al. 2006; Kingsford & Neville 2005), and the demand for quantitative freshwater conservation approaches remains high (Linke et al. 2011).

The proposed analysis can be implemented with the publicly available version 2 of the Zonation software (Moilanen et al. 2005, 2009), but this requires extra effort in the preparation of input layers. Retention transforms and weight calculations are included in the recently released Zonation v3beta, greatly simplifying workflow.

Acknowledgments

A.M. thanks the Academy of Finland CoE program 2006–2011, grant 213457, and the ERC grant StG 260393 – GEDA for support. J.L. and J.Q. acknowledge support from New Zealand's New Zealand's Foundation for Research, Science and Technology under Contract C01×0305 and Department of Conservation.

Ancillary