Biodiversity and Conservation

, Volume 16, Issue 13, pp 3781–3802

Systematic landscape restoration in the rural–urban fringe: meeting conservation planning and policy goals


    • Policy and Economic Research UnitCSIRO Land and Water
    • School of Earth and Environmental SciencesUniversity of Adelaide
  • Brett A. Bryan
    • Policy and Economic Research UnitCSIRO Land and Water
  • Bertram Ostendorf
    • School of Earth and Environmental SciencesUniversity of Adelaide
  • Sally Collins
    • South Australian Urban Forest Biodiversity Program
Original Paper

DOI: 10.1007/s10531-007-9180-8

Cite this article as:
Crossman, N.D., Bryan, B.A., Ostendorf, B. et al. Biodivers Conserv (2007) 16: 3781. doi:10.1007/s10531-007-9180-8


Many landscapes that straddle the rural/urban divide suffer from low levels of species diversity following extensive clearing and fragmentation of native vegetation communities and conversion of land to agriculture. Further pressures are placed on remnant vegetation by encroaching urban expansion. These landscapes now exhibit a mosaic of small, patchy vegetation remnants that are under considerable pressure from housing and light-industrial development. Furthermore, agriculture in these landscapes tends to be of high economic value from uses such as intensive horticulture. Concerted and well-planned efforts are needed to balance the many conflicts of interest and competing demands for land with the need to restore landscapes for the protection of biodiversity. There has been a recent move in Australia toward regional biodiversity planning and goal setting, however specific detail on how to plan for achieving targets in complex landscapes is lacking. This paper applies a systematic landscape restoration model to a mixed-use, peri-urban landscape on the northern fringes of Adelaide, South Australia. The region contains fragments of remnant vegetation amongst a mosaic of high-value horticulture, light industry and urban development. Models produce maximally efficient solutions that meet comprehensive, adequate and representative conservation targets. Further constraints are added to the model to take into account the value of agricultural output, the biodiversity value of remnants, and property size and tenure. The effects on solution efficiencies as the number of constraints increase are investigated. This paper demonstrates the flexibility found in applying a systematic landscape restoration methodology. The process we present can be transferred to any rural–urban fringe region.


Integer programmingGeographic Information SystemsSpatial optimisationUrban landscapesLandscape restoration


A common feature of large urban centres in Australia and North America is the rural–urban fringe, which in Australia can extend up to 100 km around the mainland capital cities (McKenzie 1996), a space labelled as peri-urban or exurban. This region is characterised by strong population growth and consequent residential/industrial development, mixed with the presence of natural resources that are important from a strategic (e.g. urban water supplies) or conservation and habitat sense, as well as ‘prime’ agricultural land and high value heritage, landscape and environmental amenities (Bunker and Houston 2003; Stenhouse 2004). This milieu of multiple land uses and competing interests and demands poses many public policy challenges (Bengston et al. 2004). A growing body of literature points to the negative associations between urbanisation and natural ecosystems (Marzluff et al. 2001; Pickett et al. 2001; McKinney 2002; Miller and Hobbs 2002; Battisti and Gippoliti 2004; Turner et al. 2004).

South Australia (Fig. 1) has one of the most concentrated patterns of settlement in Australia, with 73.1% of the population residing in the Adelaide Statistical Division at the 2001 census (Australian Bureau of Statistics 2001a). While Adelaide’s rural–urban fringe (Outer-Adelaide Statistical Division, Fig. 1) accounted for only 7.5% of the state’s population at the 2001 census, it was exhibiting an average annual population growth rate of approximately four times that of the Adelaide Statistical Division and the state as a whole (2.1% compared with 0.52% and 0.56%, respectively) (Australian Bureau of Statistics 2001a). Similar patterns of population growth at the rural–urban fringe are found across all Australian capitals and large cities. A similar pattern is occurring in the United States. Heimlich and Anderson (2001) report that population growth on the edges of metropolitan areas in the USA exceeded 10% during the period 1992–1997, but was below 5% within core metropolitan areas during the same period. Theobald (2005) modelled a 2.2% growth in urban housing densities across the coterminous USA by 2020, compared to a 14.3% growth in the peri-urban region. The impact of rapid population growth in the peri-urban landscape is the conversion of agricultural and natural open space to higher intensity developed land uses. Theobald (2005) suggests this should be a concern to ecologists because of the potential impact on biodiversity and natural values.
Fig. 1

Location of the study area in South Australia, Australia. SD = statistical division

Explanations for the relatively rapid growth are generally lacking despite the repeated pattern across much of the developed world (Fisher 2003). However, the motivation for moving to peri-urban regions stems from the forces of suburbanisation and the processes of counterurbanisation (Hugo 1989). The former is defined simply as spillover growth from metropolitan areas into the adjacent peri-urban region to create new suburban estates (Ford 1999; Fisher 2003). The latter, on the other hand, is the move of suburbanites to rural fringes in seek of a rural idyll, i.e. to replace an urban lifestyle with a more rural one (Halliday and Coombes 1995).

A recent audit of the extent and condition of Australia’s native vegetation (National Land and Water Resources Audit 2001) stated that the Fleurieu Peninsula and Mt Lofty Ranges (encapsulated by the Outer Adelaide Statistical Division, Fig. 1), two regions that contain a significant proportion of Adelaide’s rural–urban fringe, are amongst the list of Australia’s bioregions that contain the most highly fragmented remnant vegetation. Other studies support this finding (Bryan 2000; Paton et al. 2000; Westphal et al. 2003a; Paton et al. 2004). Respectively, the two regions contain 10.3% and 15.7% of their pre-European (pre-1788) vegetation cover, and 85.0% and 33.8% of the remaining vegetation is fragmented into patches of less than 1,000 ha (National Land and Water Resources Audit 2001). Much of this cover has disturbed understorey (e.g. through weed invasion) that reduces its integrity and value. As a whole, only 8.8% of Adelaide’s rural–urban fringe is under native vegetation cover (unpublished data). This is well below the broad target of 30% habitat cover loosely recommended for native species conservation (Andren 1994; Freudenberger et al. 2004). Although this target is somewhat arbitrary because it cannot be applied to all species, it is a target that is gaining some currency in the conservation planning literature.

Policy, planning and coordinated action are needed to redress the acute shortage and highly fragmented status of native vegetation in the rural–urban fringe of Australian cities. There are the policies and plans, but not necessarily the processes to implement them. There has been a recent nationwide shift to regional integrated natural resource management, of which biodiversity management is a significant component. Regional planning for natural resource management in Australia is now primarily the responsibility of regionally based agencies (e.g. Catchment Management Authorities, Natural Resource Management Boards) that are charged with planning and administering funds for natural resource management. This regionally based structure of natural resource management in Australia is reinforced by national funding programs such as the Natural Heritage Trust (NHT; Natural Resource Management Ministerial Council 2002) and National Action Plan for Salinity and Water Quality (NAP; Council of Australian Governments 2000).

The requirement for sound research and planning within a clear objectives/targets-based framework is highlighted across much of the recent integrated natural resource management (INRM) literature (e.g. Slocombe 1998; Bellamy et al. 1999, 2001; Walker et al. 2001; Edvardsson 2004). Many regional INRM agencies in Australia have developed or are in the process of developing plans to identify the major environmental assets and threatening processes operating in their region. The centerpiece of these regional INRM plans and investment strategies is a set of aspirational (long-term) targets and associated resource condition (medium-term) and management action (short-term) targets that are used as measures of progress to attaining the former.

A review of the two natural resource management plans applicable to the peri-urban area of Adelaide, namely the Mount Lofty Ranges and Greater Adelaide (Mount Lofty Ranges Interim Integrated Natural Resource Management Group 2003a) and the Northern and Yorke Agricultural District (Northern and Yorke Agricultural District Integrated Natural Resource Management Committee 2003) Integrated Natural Resource Management Plans, reveals a consistent approach to strategic biodiversity conservation. Both documents contain general broad-scale targets for protection, restoration, and the establishment of comprehensive, adequate and representative samples of biodiversity. For example, the Mount Lofty Ranges and Greater Adelaide Integrated Natural Resource Management Plan proposes an increase of 3,000 ha of habitat under protection by June 2005, and a minimum of 7,000 ha of re-constructed habitat by June 2007 (Mount Lofty Ranges Interim Integrated Natural Resource Management Group 2003a). The Northern and Yorke Agricultural District Integrated Natural Resource Management Plan proposes a 20% increase of native vegetation currently under protection by June 2005, and 5,000 ha of revegetation over a five-year period (Northern and Yorke Agricultural District Integrated Natural Resource Management Committee 2003). However, methods for identifying priority sites for meeting revegetation and general conservation targets are generally lacking, except for reference to the regional biodiversity plans that contain broad-scale and qualitative attempts at biodiversity prioritisation. Detail is also limited on how to include in prioritisation processes any existing and potential conflicts of interest that arise in the rural–urban fringe as a consequence of suburbanisation and counter-urbanisation.

Similar gaps can be inferred from critiques of catchment plans outside of Australia. For example, Brody et al. (2004), in a comparative study of 22 watershed plans in Florida, USA, scored many regional plans low on the quality of ‘goals and objectives’ they contain, and the ‘factual’ content underpinning them. Whilst Brody et al. (2004) limit the dissection of the plans to a comparative analysis of summed scores for different attributes of the plans, it can be inferred that a plan that scores low on ‘goals and objectives’ will be deficient in processes for prioritising, protecting and restoring important habitat. Low scores for ‘factual basis’ (Brody et al. 2004) underpinning plans would suggest that resource inventories are poor as is the associated understanding of human impacts on natural resources, a significant attribute in prioritisation of management of biodiversity. However, it appears that similar critiques of the content of catchment plans are rare.

This paper describes a quantitative modelling process to identify priority sites for restoring biodiversity within a region that is a part of Adelaide’s rural–urban fringe. This study complements the wide literature on systematic conservation planning (Kirkpatrick 1983; Margules et al. 1988, 2002; Bedward et al. 1992; Pressey et al. 1993; Underhill 1994; Csuti et al. 1997; Haight et al. 2000, 2005; ReVelle et al. 2002; Cabeza and Moilanen 2003; Arthur et al. 2004; Cabeza et al. 2004; Snyder et al. 2004) by focusing purely on restoration (although see Westphal et al. 2003b; Newbold and Eadie 2004; Williams and Snyder 2005) in a real-world applied planning context, as opposed to a theoretical context. In this paper the restoration of biodiversity is defined as the re-establishment of indigenous woody vegetation through direct planting and natural regeneration following changes in land management practices. Using the systematic landscape restoration principles and methodology developed in Crossman and Bryan (2006), this paper proposes an approach for identifying priority sites that takes into account biodiversity planning targets and caters for constraints that may arise when resolving planning conflicts. Outputs from the systematic landscape restoration methodology can be fine-tuned with ground truthing and discussion with local conservation and biodiversity experts and landowners. The proposition is based on a real landscape in the northern outskirts of Adelaide.


Study area

The focus of this study is a 23,000 ha region 15 km north of the Adelaide central business district (Fig. 1). This region is the focus of a local tree planting and remnant vegetation protection pilot project that has a requirement to be underpinned by sound planning and prioritisation. Topography is flat, with maximum elevations of 10 m above sea level in the far-eastern edge of the study area. The climate of the region is typical of coastal Mediterranean-type ecosystems characterised by warm dry summers and cool wet winters. Mean daily maxima range from 30°C in February to 15°C in July and mean daily minima range from 16°C in February to 6°C in July. Annual precipitation in the region is approximately 440 mm/year (Bureau of Meteorology 2006).

The study area contains a multitude of land uses, ranging from relatively low-value, low intensity grazing and cereal crops to high-value intensive horticulture, light industry (predominantly salt evaporation pans and sewerage treatment works) and urban development. The median land parcel size is 2.3 ha (mean 13.9 ha) and 70% of all parcels are less than 5 ha. Approximately 5,375 ha (23%) of the study area supports remnant native vegetation (Fig. 2).
Fig. 2

Dominant plant associations and buffered conservation rated flora within the study area

Remnant native vegetation communities were delineated through a combination of aerial photography and field survey and grouped according to overstorey dominants (Department for Environment and Heritage 2005). Although this proportion of cover is relatively high in a fragmented landscape context (McIntyre and Hobbs 2000), the majority (4,705 ha) is coastal mangrove (Avicennia marina; 1,970 ha), samphire (Halosarcia spp. and Sarcocornia quinqueflora; 1,920 ha), nitre bush (Nitraria billardierei; 535 ha) and coastal daisybush (Olearia axillaris; 280 ha) associated with unproductive and often highly saline soils toward the western edge of the study area (Department for Environment and Heritage 2005; Fig. 2). The remaining vegetation covers only 670 ha (4%) of what could be considered the arable portion of the study area. The river redgum (Eucalyptus camaldulensis) communities are restricted to the two major east–west watercourses in the centre and north of the study area, while the Eucalyptus gracilis mallee woodland is found within a number of small scattered fragments. This latter proportion is significantly lower than the recommended target of 30% to satisfy species conservation goals (Andren 1994; McIntyre and Hobbs 2000; Freudenberger et al. 2004). There is a clear need for the expansion of existing remnants, particularly across the arable, highly fragmented landscapes. Several species of conservation-rated flora would benefit from habitat restoration (Fig. 2).

Systematic landscape restoration

The ever-increasing volume of conservation planning studies (Kirkpatrick 1983; Margules et al. 1988, 2002; Bedward et al. 1992; Pressey et al. 1993; Underhill 1994; Csuti et al. 1997; Haight et al. 2000, 2005; ReVelle et al. 2002; Cabeza and Moilanen 2003; Arthur et al. 2004; Cabeza et al. 2004; Snyder et al. 2004) aim to identify networks of existing high value ecosystems that conserve the most biodiversity with limited resources, i.e. conserve biodiversity as efficiently as possible (Pressey and Nicholls 1989). This generally involves expansion of existing reserve systems to incorporate unprotected ecosystems. However, restoration may be the only available option to satisfy conservation planning principles in heavily disturbed landscapes that contain few examples of pre-European settlement vegetation and ecosystems (Saunders and Hobbs 1995; Dobson and Bradshaw 1997; Gilbert and Anderson, 1998). Crossman and Bryan (2006) apply a systematic landscape restoration approach that implements conservation planning principles (Margules and Pressey 2000) within an integer programming optimisation framework to identify geographic priorities for revegetation and/or restoration. They use physical and environmental spatial data to describe a region’s ecosystems, which in turn act as surrogates for biodiversity in highly fragmented landscapes.

Impedance layer

The models include a series of impedance surfaces (Crossman and Bryan 2006) that give spatial meaning to solutions, and perform in a similar fashion to the ‘cost surfaces’ of MARXAN (Ball and Possingham 2002; Stewart and Possingham 2005; Richardson et al. 2006). However, of most relevance to this paper, the impedance surfaces have the potential to guide solutions towards sites that minimise conflicts in landscape restoration activities as well as meeting multiple restoration priorities. For example, restoration activities should take place on sites adjoining remnant (and priority remnant) vegetation and in riparian zones (Hobbs and Norton 1996; Lindenmayer et al. 2002) to maximise species population viability (Wood and Pullin 2002) and restoration success (Sänger and Jetschke 2004; Suding et al. 2004) and create corridors to improve dispersal (Saunders and Hobbs 1991). Four impedance surfaces were developed to account for these priorities, respectively: (i) straight-line distance to remnant vegetation; (ii) straight-line distance to major watercourses; (iii) straight-line distance to public tenure, and; (iv) straight-line distance to priority remnant vegetation. The first three surfaces were calculated using basic GIS functions and warrant no further explanation. The third surfaces required the identification of priority vegetation, the methodology of which is described in the next section.

An additional impedance was developed in an attempt to minimise the likelihood of rejection of selected sites by private landowners. Hobbs and Harris (2001) and Lindenmayer et al. (2002) suggest that it is important to consider social and economic realities if restoration exercises are to succeed. Therefore, the value of primary production per parcel impedance surface was produced. Calculation of primary production value is described in a later section.

Each impedance surface was rescaled to the range 1–1,000 to minimise the effects of uneven distribution across features. Values closer to 1 represent shorter distances and lower primary production values. Impedance surfaces were combined to a single impedance layer (l) using an n-component mixture model:
$$ l = p_1 f(1) + \cdots + p_n f(n) $$
where f(1)...f(n) are impedance surfaces and are weighting coefficients that are determined by the user. The weighting coefficients can be adjusted to reflect the priorities of the decision maker, but must sum to 1. Equal weighting coefficients were applied to all impedances, the value of which was determined by the number of components.

Model description

The objective of systematic landscape restoration is to minimise:

$$ \sum\limits_{i\, = \,1}^m {x_i l_i } $$
where, for i = 1...m within a landscape of m equal size sites, li is the impedance layer (l) value and xi is a single site with a value of ‘1’ if they are within a particular class or ‘0’ otherwise. Areal constraints are applied that are a function of the area of each class of constraint data, the proportional target and the minimum area target. A solution vector is defined such that its elements xi are given a value of ‘1’ if a site is selected for restoration or ‘0’ if not selected. The objective is to minimise the number of sites to be restored subject to the constraints.

Vegetation prioritisation

Spatial datasets were developed for inclusion in the priority vegetation impedance surface in an effort to prioritise for biodiversity, and hence prioritise for restoration. Planning for the management of biodiversity in fragmented landscapes should focus on high value remnants (McIntyre and Hobbs 1999, 2000). Remnant vegetation was assessed for quality based on local expert (Todd Berkinshaw, Greening Australia; Ben Moulton, Urban Forest Biodiversity Program) knowledge of the region. Each patch of vegetation was qualitatively scored using a five-point continuum scoring system describing condition of the remnant (Table 1). The scores range from 1 (very poor), in which overstorey is severely reduced and weeds and/or bare ground cover up to 50%, to 5 (excellent), representing patches of maximum diversity, no weeds and few, if any, signs of anthropogenic disturbance.
Table 1

Scoring continuum for assessing vegetation condition (adapted from Moritz and Moss 2003)

Condition class

Criteria for assigning condition class

Plant species composition


Soil erosion/disturbance

5 = Excellent (near intact)

Maximum diversity of annual and perennial species possible for the association. No environmental weeds

Perennial species of various ages

No erosion (other than natural features or processes). Plant and litter cover protects soil from wind and water erosion in all seasons. Very few signs of human disturbance

4 = Good

Minor reduction in density of palatable and susceptible perennials. Increased proportion of shorter-lived species of environmental weeds, but at very low densities OR other weeds present which are not threatening biodiversity

Perennial species of various ages.

No accelerated erosion, or only minor/slight erosion evident. Possibly increased susceptibility of soils to erosion in dry seasons. Good litter or lichen or moss cover over most of ground

3 = Fair

Significantly reduced cover and density of palatable species. Environmental weeds present and possibly widespread but not threatening long-term ecological integrity if controlled. Weed cover and or proportion of weed species up to approximately 25% of native species in localised areas

Significantly reduced regeneration of palatable species. Establishment of less preferred or unpalatable species

Moderate erosion may be evident. Reduced density and cover of perennial species and litter increases susceptibility of soils to erosion in most seasons. Bare patched of soil

2 = Poor

Dominance of annual and ephemeral species and perennials with relatively low palatability (i.e. high proportion of “increaser” species). Weed cover and/or weed diversity up to 50% of native species in localised areas, threatening long-term ecological integrity

No regeneration of desirable perennial species. Existing stands degenerating

High susceptibility of soils to erosion in all seasons. May be severe erosion present. Extent of past erosion renders site susceptible to further soil movement if grazed at any level. Litter, moss or lichen crust much reduced

1 = Very Poor

Seasonal cover of only pioneer ephemerals or unpalatable species. Weeds and/or bare ground cover over 50% in several areas. Overstorey may be lacking (past clearance) or much reduced

No regeneration occurring

Unstable. May be severely or very eroded. Large areas of bare ground

Patches were further scored using a 1–5 scale for each of shape, degree of landscape fragmentation, distance to existing protected areas and proportion of pre-European vegetation remaining. The vegetation quality and four landscape criteria scores were summed to produce a relative measure of priority for patch management and expansion.

Shape complexity was calculated using the mean patch fractal dimension function available in FRAGSTATS (McGarigal et al. 2002). Mean fractal dimension approaches one for shapes with simple perimeters and approaches two for more complex shapes. This index was used in preference to the common perimeter-area ratio measurement of complexity because it can be applied at multiple spatial scales. Mathematically, mean patch fractal dimension (F) equals:
$$ F = \frac{{2\ln (1/4p_{ij} )}} {{\ln a_{ij} }} $$
where pij is the perimeter (in metres) of patch ij and aij is the area (in square metres) of patch ij. Range cut-offs for categorisation into five classes was made using 0.2 increments between one and two.

To calculate fragmentation, a neighbourhood function was passed over a grid of remnant vegetation cover within the region. The neighbourhood function was parameterised to sum the amount of vegetation cover in a 5 km radius around each site in the study area. The output was then converted to a percentage to represent the percentage cover of vegetation within a 5 km-radius circular neighbourhood extending from each site. This grid was reclassified into fragmentation categories based the four states of landscape alteration defined by McIntyre and Hobbs (1999): 1, relictual (<10% cover); 2, fragmented (10–59.9%); 3, variegated (60–89.9%), and; 4, intact (90% +). No sites are within ‘intact’ landscapes because of the limited extent of remnant vegetation.

A straight-line distance function was used to calculate the distance that each site containing remnant vegetation is from already protected vegetation (National Parks and Wildlife Service reserves, heritage agreements, local government reserves). Range cut-offs for categorisation into 5 classes are arbitrary but designed to give greater importance to remnant vegetation communities closest to existing protected areas.

The proportion of pre-European vegetation remaining was calculated for those remnant vegetation communities within the area covered by the pre-European mapping. The ratio of pre-European against remnant was categorised into five classes using arbitrary cut-offs designed to giver greater importance poorly represented communities in the study area.

Economic prioritisation

Choosing sites for systematic landscape restoration should take into consideration economic factors, particularly the opportunity cost of revegetating arable land (Hobbs and Harris 2001; Lindenmayer et al. 2002; Stoms et al. 2004). Income from primary production will be forgone when converted to a treed landscape for biodiversity. Therefore, it makes sense to prioritise systematic landscape restoration activities toward properties with lower value primary production relative to the region as a whole. The Australian Bureau of Statistics conducts 5-yearly censuses of agricultural production data for establishments with an estimated value of agricultural operations of $5,000 or more. The most recent census was in 2001 (Australian Bureau of Statistics 2001b). The Australian Bureau of Statistics database contains the total value of output for each agricultural commodity and is compiled to the statistical local area. We linked this database to the cadastral database of the study area to derive an estimate of the value of primary production at the property scale. The cadastral database distinguishes properties and the type of land use on each. Value is measured as output (AUS$) per hectare per property. This information formed an impedance layer.


Biodiversity surrogates

Surrogate data for biodiversity in complex topographical landscapes would typically include variables derived directly from a digital elevation model (e.g. slope, aspect and elevation), as well as indices describing variation in local climate (e.g. temperature, precipitation, solar radiation) (Faith et al. 2001; Margules et al. 2002; Bryan 2003). However, topography in this study area is flat and the extent of the region is limited. Therefore, variations in topographic and climatic data will be minimal at the scale of this study. The surrogates in this study are limited to descriptions of soil, geological strata, and an estimate of pre-European (pre-1788) native vegetation (Table 2). These are treated as separate inputs.
Table 2

Summary list of spatial data used in this study





Soil land systems



External, Government Agency

Geological Units



External, Government Agency

Pre-European vegetation



External, Government Agency

Remnant vegetation



External, Government Agency

Conservation-rated flora



External, Government Agency

Distance to major watercourses



Internal, straightline distance function

Distance to remnant vegetation



Internal, straightline distance function

Distance to high priority remnant vegetation



Internal, combination of vegetation quality and landscape criteria

Distance to low value agricultural output



Internal, classified landuse and ABS agriculture statistics database

Distance to public land tenure



Internal, straightline distance function

Small (<5 ha) parcels



Internal, spatial query

a R = raw; D = derived

b B = biodiversity surrogate; C = constraint

The soils data were derived from a long-term soil survey by the South Australian Department of Primary Industries and Resources. At the finest level of detail the soil database contains units of homogenous soil type on similar geology and characteristic topography. At the coarser level, these soil landscape units have been amalgamated into soil land systems that contain a repeating pattern of geology, topography, soils and vegetation. The latter are considered to be the most appropriate unit for surrogacy in this study area because of the relatively small variation in topography. A total of 10 soil land systems exist in the study area. The Department of Primary Industries and Resource’s geological strata database, while partly correlated to the soils database, contains descriptions of relatively homogenous strata. Geology explains broad-scale ecosystem patterns. Soil, however, explains finer-scale patterns. Seven geological units are present in the study area. Estimates of pre-European vegetation have been made for the study area south of the Gawler River (Kraehenbuehl 1996). Distribution of vegetation communities thought to exist prior to clearance was estimated from historical botanical and field naturalist notes, combined with extensive remnant vegetation field checking and known associations between vegetation and soil and landscape types. Five vegetation communities are thought to have been present in the southern part of the study area.

Proportional and areal targets

A fixed areal target of 15 ha of each surrogate data class and proportional targets in the range of 1–50%, in 1% increments, were applied to each model. This 15 ha is an arbitrary figure because there is no functional research indicating that this area (size) maximises biodiversity or guarantees species viability in this particular environment. However, general studies into species/area relationships suggest patch sizes of 15 ha are a minimum for conservation (Loyn 1987; Freudenberger et al. 2004). An additional target of 100% of the area within a 500 m buffer of conservation-rated species recordings was included to ensure all species and surrounding habitat is selected for restoration in all model solutions. Again, this area (79 ha) is an arbitrary figure because there are no studies into whether this is sufficient to maintain species populations. But excluding this information creates the risk that such important sites will be overlooked when planning for restoration.

Existing native vegetation

The models take into account the presence of remnant native vegetation. Specifically, sites that support remnant native vegetation (Fig. 2) had to be selected in all solutions and were given a value of 1 for xi (Eq. 2). Thus, the aim of the models was to select non-vegetated sites in the study area that, together with existing remnant vegetation, met conservation targets most efficiently.

Parcel size

The fourth constraint was designed to reduce planning complexity by eliminating land parcels smaller than 5 ha. Some degree of urban subdivision will already exist in the rural–urban fringe to meet the demands of rapidly growing populations. Planning and on-ground action will be expedited if the number of potential landowners targeted for systematic landscape restoration is minimised. Therefore, sites within parcels of area less than 5 ha in the present study area were excluded from solutions and were given a value of 0 for xi (Eq. 2). This had the effect of masking those parcels smaller than 5 ha.

Model implementation

All input datasets were converted to 1 ha resolution grids. All spatial modelling and analyses were completed in the ESRI ArcGIS 8.3 suite of GIS software (ESRI 2002). ILOG’s OPL Studio 3.6 and CPLEX 9.0 optimisation engine were used for all optimisation problem-solving (ILOG 2003). CREDOS (Crossman et al. 2007), an optimisation spatial decision support tool, facilitated data transfer between ArcGIS and CPLEX. The number of sites (m) of 1 ha resolution in the study area is 22,983. The total number of classes of biodiversity surrogate data equals 22 (10 soil land systems, seven geological units, and five pre-European vegetation communities). All models were run on a Pentium 4, 3.0 GHz PC with 1 Gb of RAM.

Cabeza et al. (2004) demonstrate that the inclusion of additional constraints in systematic conservation planning, contrary to earlier studies (Possingham et al. 2000), does not increase the cost of solutions in terms of the number of sites or the total area selected. Although Cabeza et al. (2004) were referring to protection of existing habitats, it is logical to expect that the spatial inflexibility of site selection found in conservation planning would apply to landscape restoration. The Cabeza et al. (2004) findings were examined in relation to landscape scale restoration by comparing solution size as the number of constraints was increased.


The vegetation prioritisation process identifies patches of remnant vegetation to target for systematic landscape restoration planning (Fig. 3). An arbitrary cut-off priority score of 17 (Fig. 3) was applied to the remnant vegetation total priority score. Although this is an arbitrary value, it is justified for demonstration purposes.
Fig. 3

Inputs into the assessment of vegetation priority (inputs) and the total priority score

The impedance surfaces (Fig. 4), when combined to produce an impedance layer (Fig. 5), force model solutions to sites of low impedance cost. However, increasing the complexity of the impedance layer by adding more surfaces has the effect of smoothing impedance values across the study area. For example, the impedance layer with only distance to remnant vegetation as an impedance surface has a mean score of 173.25 (SD 194.01), while the impedance layer with all five combined impedance surfaces has a mean of 205.81 (SD 140.35).
Fig. 4

The five impedance surfaces used in construction of an impedance layer. 1: distance to remnant vegetation; 2: distance to major watercourses; 3: agricultural value; 4: distance to high priority remnant vegetation; 5: distance to public tenure land parcels
Fig. 5

The five impedance layers and masked parcels (exclusion of parcels smaller than 5 ha) used in the six systematic landscape restoration models. All models include remnant vegetation as a constraint. A: 1; B: 1 + 2; C: 1 + 2 + 3; D: 1 + 2 + 3 + 4; E: 1 + 2 + 3 + 4 + 5; F: 1 + 2 + 3 + 4 + 5 and masked parcels. See Fig. 4 for impedance surface code descriptions

Regardless of the dynamics within impedance layers, their inclusion produces solutions that are practical and manageable, and can be used to assist with planning landscape restoration activities for biodiversity conservation (Fig. 6). Solutions in Fig. 6 were produced under the four-constraint model (i.e. includes all five impedance surfaces and the masking of parcels smaller than 5 ha). Selected sites adjoin and connect remnants, are absent from small parcels, and are located on lower-value primary production land. Furthermore, it is the high value remnants located along watercourses that tend to be buffered. It is worth noting that the majority of selected sites are in the arable, extensively cleared parts of the study area. These areas have very limited remnant vegetation cover (less than 5%).
Fig. 6

Solution set in 5% incremental increases of proportional target. Areal target is fixed at 15 ha. Impedance layer F (see Fig. 5) applied in this model

Increases in impedance layer and constraint complexity has little effect on solution efficiency (Fig. 7). The increase in the number of sites required to satisfy the various constraints shows little variation between models. All models, apart from A (Fig. 5), behave in a relatively steady manner within site selection space.
Fig. 7

Increase in solution size for the six systematic landscape restoration models A to F. All models include remnant vegetation as a constraint. A: impedance surface 1; B: 1 + 2; C: 1 + 2 + 3; D: 1 + 2 + 3 + 4; E: 1 + 2 + 3 + 4 + 5; F: 1 + 2 + 3 + 4 + 5 and masked parcels. See Fig. 4 for impedance surface code descriptions

As the proportional target increases, solution size increases, but not linearly (Fig. 7). When low proportional targets are set, remnant vegetation will contribute significantly to achieving systematic landscape restoration goals. Very few additional sites in the study area were required on top of those supporting remnant vegetation to meet proportional targets of up to 10% representativeness (Figs. 6, 7). Beyond 10%, solution size increases steeply until a stable state is reached, in this case 15%. The steepness of the slope between 10% and 15% may be an artefact of the existing area of remnant vegetation in the surrogates, however, this warrants further investigation. Beyond 15%, an increase in proportional target of 1% costs approximately only 0.8% of the study area (Fig. 7). Therefore, greatest efficiencies in site selection for representative goals are found at targets lower than 15%.


The need to plan for biodiversity by setting landscape restoration objectives, as opposed (or in addition) to achieving the conservation requirements of a particular species or suite of species (Lambeck 1997), has recently been recognised in the literature (Lindenmayer et al. 2002; Lindenmayer and Fischer 2003). Consequently, landscape planning is beginning to appear in regional biodiversity plans. In Australia these plans are increasingly more explicit and specific about their goals for vegetation management and conservation. This has been driven not just by landscape ecology principles and theory, but also by the nature of top-down funding programs. It is now relatively common for these plans to contain general broad-scale targets for protection, restoration, and establishment of comprehensive, adequate, and representative samples of biodiversity. Examples include 7,000 ha of reconstructed habitat by 2007 (Mount Lofty Ranges Interim Integrated Natural Resource Management Group 2003a), and 5,000 ha or revegetation over a five-year period (Northern and Yorke Agricultural District Integrated Natural Resource Management Committee 2003). Methods for spatial planning to achieve specific conservation targets in the rural–urban interface have received minimal attention in both the scientific literature (except Ruliffson et al. 2003; Haight et al. 2005) and the local/regional plans.

This paper presents an approach that can be used to systematically plan for conservation with the ability to consider the complex makeup of landscapes and the socio-economic interests of its inhabitants. Efforts to meet targets, such as the examples in the previous paragraph, will need to be guided by an adaptive plan. For example, a hypothetical target in our rural–urban study area could be 30% representation of surrogate data (Fig. 6) by a certain time, e.g. 2010. To meet this target, 3,956 ha of restoration (in this case revegetation) are required (Fig. 6). This would take total vegetation cover to 40% of the region (5,375 ha of remnant plus 3,956 ha of revegetation), but conservation planning goals would be given the greatest chance of success with the knowledge that 30% of environment types have been restored in the region.

This study considers relatively simplified and generalised ecological processes within a restoration context. Planning for ecological restoration would ideally investigate how an ecosystem works through improved understanding of composition, structure, function, heterogeneity and resilience (Hobbs and Harris 2001). Specific consideration should be given to biotic factors that influence the state of the ecosystem, and therefore have a direct bearing on the success of meeting restoration goals. These factors include exotic species invasion, altered herbivory regimes, landscape connectivity and seed sources, and long-term climate changes (Suding et al. 2004). Many of these ecosystem state variables that describe the ecosystem and its chances of restoration success are specific to the site level, whereas this is a landscape scale study. Therefore, only minimal consideration has been given to the site-based variables via the efforts to assess the value of remnant vegetation qualitatively for prioritising management. This study considered the condition of the remnant vegetation (from a qualitative assessment of native species diversity, presence of exotic species, and herbivory impacts) and landscape context (through FRAGSTATS and neighbourhood analyses) to rank remnant vegetation from high to low quality. Also included in the systematic landscape restoration models was a measurement of landscape connectivity, and its influence on dispersal of local seed, by incorporating impedances that encourage solutions to expand and connect existing remnants. But the solutions identified by the model should be ‘fine-tuned’ with expert ground-truthing and site-by-site assessments of restorability and ability to meet desired restoration goals.

Practical implementation of solutions could involve an iterative process whereby bottom-up planning is blended with this study’s top-down approach, such as in active adaptive management (Possingham 2001). Preliminary outputs such as those in Fig. 6 would be a first pass and could guide early planning activities. Community-level consultation would identify sites within a solution that could be managed to reduce threats and revegetated in the short-term, and sites that could not be. This would be based on a landowner’s willingness to participate. Such consultation may also identify sites outside of a solution that could be revegetated. This information would go back into the model in the form of extra constraints. The systematic landscape restoration models would be rerun with the updated information, and the new solution(s) would be presented at further community-level forums. The iterative process continues until targets are reached. Iterations might be annual or biennial. This systematic approach is transparent, but most importantly for complex landscapes such as those at the urban–rural interface, is inclusive and would allow for easy scheduling of on-ground revegetation activities.

Although not directly investigated in this paper, it would be relatively simple to measure costs of on-ground expenses and opportunity cost to various landowners of revegetation activities. The models presented here can identify sites to the nearest hectare that meet any target defined by the user. Assuming the cost of restoration/revegetation per hectare is known, planners at the local and regional level can estimate total cost of works to meet a specific conservation target and apportion to budgetary periods, or if budgets are predefined, can identify a systematic landscape restoration target that will be achieved within the budgetary period. Similarly, the Australian Bureau of Statistics agricultural statistics were queried to estimate output per hectare per property across the study area. It was then a simple GIS operation to estimate the forgone income per property for any systematic landscape restoration output. If the impedance surface that favours properties with lower value output has been included, then theoretically the set of selected sites should be of lower opportunity cost than if it had not been included. Landowners with low value output could be targeted for on-ground works, and in broader terms, the systematic landscape restoration outputs will be more saleable because opportunity cost has been explicitly addressed.

Model solutions without an impedance layer are impractical and could never be presented to regional planners, let alone implemented. While they meet areal and proportional targets, and hence contribute to regional biodiversity planning and conservation, they are inadequate for species protection because site arrangement is irregular, resulting in increased levels of fragmentation.

However, several questions arise in relation to the impedance layers. First, does the inclusion of an increasing number of impedance surfaces in the construction of an impedance layer add any value to the models? While it appears advantageous to include several impedance surfaces and constraints, the trade-off may be a loss in specificity. Second, is the additive model used to combine impedance surfaces the best approach? And third, will different weighting coefficients have a significant effect on reducing the smoothing effect? It has been demonstrated in this paper that increasing the number of constraints through an impedance surface with more impedance layers and masking (excluding) parcels smaller than 5 ha has little effect on the cost of solutions. This supports the Cabeza et al. (2004) findings for reserve selection planning. There remains the need to investigate the sensitivity of impedance layers to variations in impedance surface inputs from two angles: (i) how sensitive is the impedance layer and site selection to increasing number of impedance surfaces, and; (ii) how sensitive is the impedance layer and site selection to changes in the weighting component values. There is also value in investigating alternatives to the n-component additive mixture model, such as a multiplicative model or some other form of multi-criteria data combination.

McIntyre and Hobbs (2000) argue that protection and improvement of existing remnants provides the best value for money in the management of degraded landscapes. They suggest that the priority should always be to maintain any existing good quality habitat by removing exogenous disturbances, and improve degraded remnants by also removing disturbances and implementing active management such as weed removal and re-introduction of understorey species. Revegetation is expensive and is unlikely to result in the restoration of habitat back to its unmodified state (Diamond 1987). Therefore, protecting and improving existing remnants will be the most efficient approach to meeting conservation planning targets (Margules and Pressey 2000). Remnants in this study contribute up to 10% representativeness. Of course this is highly variable across regions and is dependent on distribution and presence of remnant vegetation. However, in fragmented landscapes restoration through revegetation will be the only option to achieve higher targets such as 30% (Andren 1994).

Many of Australia’s agricultural regions have been subjected to widespread clearing of vegetation and subsequent fragmentation of habitat. Reserve selection in these areas will not be sufficient to conserve biodiversity because remnant habitats are small, isolated and subject to disturbance via edge-effects. It is in these landscapes that restoration is urgently required to begin the halt of further species decline. However, land in these regions is in high demand from a variety of land uses, particularly in peri-urban regions experiencing rapid population growth, and restoration is expensive. Restoration needs to be carefully planned and prioritised to gain maximum ecological benefit whilst having minimal adverse economic impact. Careful planning is of most necessity in regions where urban and rural land uses collide.


This research was conducted with the support of an Australian Research Council Discovery grant DP0343036. We thank Ross Oke, Matt Turner, Ben Moulton and Darcy Peters at UFBP, and Todd Berkinshaw at Greening Australia for kind assistance with data supply and general support. Thank you to Kris Rothley for reviewing an earlier version of the manuscript. We thank an anonymous reviewer for improving the manuscript.

Copyright information

© Springer Science+Business Media, Inc. 2007