Mapping a new future: using spatial multiple criteria analysis to identify novel habitats for assisted colonization of endangered species


  • M. C. Dade,

    1. School of Earth and Environment, The University of Western Australia, Crawley, WA, Australia
    2. School of Animal Biology, The University of Western Australia, Crawley, WA, Australia
    Search for more papers by this author
  • N. Pauli,

    1. School of Earth and Environment, The University of Western Australia, Crawley, WA, Australia
    Search for more papers by this author
  • N. J. Mitchell

    Corresponding author
    1. School of Animal Biology, The University of Western Australia, Crawley, WA, Australia
    • Correspondence

      Nicola Mitchell, School of Animal Biology, The University of Western Australia, Crawley, WA 6009, Australia. Tel: +61 8 6488 4510


    Search for more papers by this author

  • Editor: Earl McCoy


Many species with restricted ranges or poor dispersal ability will occupy suboptimal habitats under climate change. The practice of translocating populations to more suitable areas outside the known historical range (assisted colonization) has been advanced as a solution to reduce the risk of extinction in the wild. Due to the high-risk and bureaucratic complexities of assisted colonization, it is imperative that a systematic process is used to select release sites that have a reduced likelihood of translocation failure. Here we demonstrate how a spatially explicit, three-stage multiple criteria analysis (MCA) can be used to identify potentially suitable sites for assisted colonization of an endangered species. We employ this method as an initial screening process, prior to final selection of sites for assisted colonization of the critically endangered western swamp tortoise (Pseudemydura umbrina). This species occurs naturally in two small reserves in southwestern Australia, and is currently threatened by a shift to a drier climate and consequent changes to hydrological regimes. A literature review, characteristics of remnant P. umbrina habitat, and expert knowledge were used to create a composite index of habitat suitability, mapped across the entire south-west bioregion of Australia. The most suitable sites were 150 to 250 km south of the known range of P. umbrina, in areas of high landscape connectivity and low human population density. A subset of sites were examined in further detail and ranked using weighted summation. Careful use of MCA, taking into account data uncertainties and differences in expert opinion, can be a valuable tool when evaluating novel habitats for threatened species.


The distributions of many species are changing as an adaptive response to altered local and regional climates (Parmesan et al., 1999; Hoegh-Guldberg et al., 2008; Seddon, 2010; Moreno-Rueda et al., 2012). Species that are unable to change their distribution because they are poor dispersers are especially disadvantaged by changing climatic conditions. In such instances, assisted colonization – the translocation for conservation purposes of a species outside of its known historical range – has been proposed as a means of reducing extinction risk (Hoegh-Guldberg et al., 2008; IUCN/SSC, 2013). Assisted colonization is controversial, largely due to the risks inherent with introductions of species to novel habitats (Ricciardi & Simberloff, 2009; Seddon, 2010; Burbidge et al., 2011), but has been used successfully as a conservation tool in New Zealand, where threatened birds and reptiles have been translocated to offshore islands to avoid predation by introduced mammals (Clout & Craig, 1995; Thomas, 2011). To date, assisted colonization has not been widely used as a method to reduce the risk of extinction due to climate change, but is increasingly being considered as a more assertive approach to conservation (Willis et al., 2009; Seddon, 2010; Burbidge et al., 2011).

Many past attempts at assisted colonization were conducted under crisis conditions, and little information has subsequently been available on the reasons behind successful or failed attempts (Towns & Ferreira, 2001). Previous translocations of herpetofauna have shown that without suitable high-quality habitat, translocations have a low chance of success (Germano & Bishop, 2009). To ensure that a potential translocation site can support a particular species, habitat evaluation should include characteristics that allow for the survival of the target species, as well as ensuring that the translocation site can maintain ecological integrity (Griffith et al., 1989). Identification of suitable habitat requires understanding of a species' fundamental niche and consideration of management, logistical and socio-economic factors that could influence the likelihood of success (Heaton et al., 2008; Burbidge et al., 2011; Knight et al., 2011; Thomas, 2011). Given the potential risks associated with any species translocation, and particularly with assisted colonization, systematic and transparent methods for prioritizing release sites need to be developed.

Despite habitat evaluation being critical to translocation success, a systematic process for site selection is rarely performed when planning species translocations. Evaluating the suitability of potential habitat involves assessing a range of diverse biophysical and management criteria, which are often expressed in different units of measurement. This can distort assessments, leading to inaccuracy in habitat evaluation and reduce the chance of a successful translocation (Lahdelma, Salminem & Hakkanen, 2000). Multiple criteria analysis (MCA) is a process that provides structured, transparent ranking of decision options, against multiple criteria measured in various units (Hajkowicz, 2008). When applied in the context of site selection, suitable regions can first be identified with geographic information systems (GIS), and alternative sites then assessed using the straightforward and transparent method of weighted summation (Hajkowicz, 2008).

Identifying alternative suitable translocation sites is essentially a spatial prioritization problem (Knight et al., 2011). A common approach to identification of sites based on multiple spatial criteria involves using GIS to create composite maps with different layers of environmental information for each criterion to be considered (Clifton & Boruff, 2010; Dermol & Kontić, 2011). Indicators for each criterion are developed, scored and combined to create an overall composite index of site suitability (Graymore, Wallis & Richards, 2009). This approach has been used to identify suitable areas for translocation for species within their current range. For example, Heaton et al. (2008) used a spatially explicit MCA model to identify areas for the translocation of the Mojave Desert tortoise (Gopherus agassizii) in California, USA, and a similar method was used by Smith et al. (2007) to identify habitat for the Julia Creek dunnart (Sminthopsis douglasi) in Northern Australia. Despite its obvious applications in assisted colonization, spatially explicit MCA has not previously been used to identify suitable habitat outside of a species' current or historical range.

Here we demonstrate how wetlands that are potentially suitable for the assisted colonization of a threatened species can be identified using a systematic and transparent MCA approach, combined with expert knowledge. Our example species is the western swamp tortoise (Pseudemydura umbrina) – a critically endangered reptile from southwestern Australia for which assisted colonization is planned for the near future (Mitchell et al., 2013). Little is known about the distribution of P. umbrina prior to European settlement; for example, it is unknown whether current habitats represent a portion or the whole of their fundamental niche. Anecdotal evidence suggest that P. umbrina may be a reasonably adaptable species, as captive-bred individuals use artificial aestivation refuges, nest in sand next to holding ponds and eat food that is not part of the diet of wild individuals (S. Arnall, pers. comm., 2012). Hence, attempts to replicate the exact habitats found at existing sites may be unnecessarily restrictive when planning assisted colonization, but factors important to long-term survival of P. umbrina and to maintenance of the ecological integrity of the release site should both be considered. Using GIS methodologies, we combine quantitative and qualitative data in a spatially explicit MCA to produce a mechanism for identifying suitable new habitats for endangered species.

Material and methods

Case study

Reptiles are particularly vulnerable to climate change as many of their physiological processes are dependent on ambient temperature (Huey et al., 2012; Moreno-Rueda et al., 2012). A recent assessment indicated that 19% of the world's reptiles are threatened with extinction, including about 50% of assessed species of tortoises and turtles (Böhm et al., 2013). The western swamp tortoise (P. umbrina) is critically endangered (Burbidge et al., 2010; IUCN, 2012) due to extensive land clearing and a naturally restricted range. This species now occurs in only four wetlands in southwestern Australia; the Ellen Brook and Twin Swamps Nature Reserves, where this species naturally occurs, and the Moore River and Mogumber Nature Reserves, where captive-bred individuals have been released (Fig. 1).

Figure 1.

Study area indicating the IBRA boundaries for the South West Botanical Province, together with the extent of detailed wetland mapping available in this region and current habitat for Pseudemydura umbrina.

Due to long-term declines in rainfall and reduced groundwater recharge, the time during which the wetlands hold water (the hydroperiod) is decreasing in length (Mitchell, Jones & Kuchling, 2012; Mitchell et al., 2013). P. umbrina feeds and breeds in water, and the viability of wild populations depends on the continuity of suitable hydroperiods. Additionally, the fragmented nature of the current habitat prevents individual P. umbrina from seeking out more suitable areas (Burbidge et al., 2010; Mitchell et al., 2013). As the current habitat is unlikely to sustain the species in the wild over the long term, assisted colonization is being planned for this species (Burbidge et al., 2010; Mitchell et al., 2013). Mapping of the thermodynamic niche of P. umbrina, under-current climates and projected climates for 2030, has recently been completed (Mitchell et al., 2013), and indicates that translocations sites well south of the current range are likely to provide the most climatically suitable habitat in the future.

Study area

The study area encompassed the South West Botanical Province (SWBP) in the southwest of Western Australia (Beard, 1980). The SWBP falls within one of the world's 34 biodiversity ‘hotspots’, where regions of high endemism are experiencing high rates of habitat loss (Myers et al., 2000). Most of the 7000 plants described from the hotspot (Hopper & Gioia, 2004) have evolved on nutrient-poor, highly weathered soils; the vegetation is dominated by sclerophyllous species forming associations including forests, woodlands, mallee and kwongan (heath). The majority of the SWBP is characterized by Mediterranean climates (Köppen classifications Csa and Csb) with warm to hot summers and mild, moist winters (Kottek et al., 2006; Peel, Finlayson & McMahon, 2007).

The SWBP comprises seven bioregions according to the Interim Biogeographic Regionalization for Australia (IBRA; DSEWPaC, 2012), covering approximately 298 577 km2 (Fig. 1). The extant populations of P. umbrina are found within the Swan Coastal Plain IBRA bioregion, at the western edge of the SWBP. There are a diverse range of wetlands throughout the SWBP; typologies have been developed to differentiate wetlands according to geomorphic form, hydroperiod, degree of inundation, vegetation type and conservation status (Semeniuk, 1988; Semeniuk et al., 1990; Semeniuk & Semeniuk, 1995; Government of Western Australia, 1997). Detailed mapping of wetlands has only been undertaken in some parts of the SWBP. The initial stages of this study assessed the entire SWBP for habitat suitability for P. umbrina, while the final stage focused on two areas where detailed, higher precision wetland mapping was available. These areas were the Swan Coastal Plain (Davis, Townley & Balla, 1996; Hill et al., 1996), and the area between the towns of Augusta and Walpole in the southern coastal region of Western Australia (V&C Semeniuk Research Group, 1997).

Development of a spatially explicit MCA model

This research combined spatial analytical tools with MCA methods to create a spatially explicit MCA approach for identifying and ranking potential sites for the assisted colonization of P. umbrina. The development of the spatially explicit MCA model comprised three stages: (1) identification of habitat criteria and assessment of data availability using a hierarchical decision problem approach; (2) mapping of all potentially suitable habitat in the SWBP and creation of a composite index to determine the degree of suitability; and (3) weighted summation, ranking and sensitivity analysis of a sample of the most suitable sites identified in stage 2 in two high-priority regions: the Swan Coastal Plain and the Augusta to Walpole region. The methods for each stage are described below.

Identification of habitat components, indicators and data availability

To aid in identification and assessment of model criteria, the decision problem was mapped out as a conceptual model using a hierarchical structure (Fig. 2), following the example of Kontos, Komilis & Halvadakis (2005). The first level represents our overall objective to identify suitable translocation sites for P. umbrina. The second level represents habitat components that would contribute to the suitability of the host site, either in terms of biophysical characteristics or reduced risk of threatening processes. The third level details attributes of each habitat component that could be used to quantify habitat suitability, based on existing spatial and non-spatial data.

Figure 2.

Hierarchical structure of the decision problem for evaluating site suitability of potential translocation sites for Pseudemydura umbrina.

The habitat components and indicators used in the MCA model were identified through a review of relevant literature, observations from a site visit in May 2012 to Ellen Brook Nature Reserve, and through consultation with four experts on the species. Expert consultations were based loosely on the expert elicitation structure described in Martin et al. (2012). A semi-structured interview was conducted with all individuals to identify the habitat components regarded as important and potential indicators. At a later date, a final list of habitat components was provided to each expert, who was asked to rank the components based on their opinion of how critical each would be to successful translocation of P. umbrina to a novel site. The habitat components included soil type, wetland type, vegetation structure and composition, presence of native vegetation, wetland salinity, presence of roads, size of the site, food availability, current land use, land use change and presence of feral predators (Table 1). As comprehensive spatial data were only available for some of the habitat components, they were divided into two groups: (1) those components that were mapped and used to create a suitability mask and composite index of habitat suitability; and (2) those components that were researched after mapping, using a literature review and online databases, to further assess their suitability (Table 2).

Table 1. Assessed habitat components and their importance to future assisted colonization of Pseudemydura umbrina
Habitat componentaRationale for inclusion
  1. aHabitat components assessed in this study were limited to those for which reliable and/or detailed data were available for either the whole study area, or a significant component of the study area.
Soil typePseudemydura umbrina aestivates underground over summer, requiring water-holding soils. Current habitat occurs on the Gilgai clay complex, or on sandy duplex soils (Schoknecht, 2002; Burbidge et al., 2010). Peat soils may be suitable, due to high water-holding capacity (A. Burbidge, pers. comm., 2012).
Wetland typeThe biology of P. umbrina is directly linked to wetland hydrological characteristics (Burbidge et al., 2010). Pseudemydura umbrina inhabits freshwater wetlands with a hydroperiod of around seven months, with shallow (< 80 cm) water depth to meet feeding, breeding and temperature requirements (Burbidge, 1981).
Presence of native vegetationMany wetlands in the South West Botanical Province (SWBP) of Western Australia have been cleared or suffered severe degradation (Davis & Froend, 1999). Intact native vegetation can be used as an indicator of disturbance and potential habitat suitability.
Vegetation structure and compositionPseudemydura umbrina requires medium density vegetation for aestivation and protection from predators, but will avoid very dense vegetation. Vegetation canopy height must be low enough to prevent excessive shading and reduction in water temperature (A. Burbidge, G. Kuchling, S. Arnall, D. Bradshaw, pers. comm., 2012). Current habitat is characterized by Banksia woodland and Melaleuca shrubland (G. Kuchling, pers. comm., 2012).
Current land useLand use is an indicator of disturbance and ease of future management. Urban and industrial areas exert pressure due to dogs, cats and human activity (G. Kuchling, pers. comm., 2012). Agricultural areas can include suitable habitat but have high initial and ongoing costs for land purchase and rehabilitation (A. Burbidge, G. Kuchling, S. Arnall, D. Bradshaw, pers. comm., 2012). Crown land is the most compatible, secure and least costly land use type.
Land use changeTranslocating a species to areas within projected development zones is not a viable long-term strategy if habitat is later developed (Heaton et al., 2008). Note that in special circumstances it is possible for the state Government to purchase and set aside land for conservation purposes within projected development zones.
Food availabilityPseudemydura umbrina requires abundant, diverse prey in spring and summer. Prey includes aquatic invertebrates, including crustaceans and insect larvae, as well as tadpoles (Burbidge et al., 2010; Gilbert, 2010). Captive-bred individuals can adapt to dietary change (A. Burbidge, G. Kuchling, S. Arnall, D. Bradshaw, pers. comm., 2012).
Site sizeExisting reserves for P. umbrina are 80 and 141 ha (Burbidge et al., 2010). Individuals have been recorded moving outside reserve boundaries, suggesting that larger sites of at least 200 ha are required (G. Kuchling, pers. comm., 2012).
Distance from main roadsIndividuals have been killed on roads near existing reserves, regardless of fencing (A. Burbidge, G. Kuchling, S. Arnall, D. Bradshaw, pers. comm., 2012).
PredatorsPredators include foxes, cats, pigs, raptors, rats, waterbirds, bandicoots (Burbidge et al., 2010). Baiting for feral animals has reduced deaths through predation (A. Burbidge, G. Kuchling, S. Arnall, D. Bradshaw, pers. comm., 2012).
SalinitySecondary salinization threatens the ecological integrity of freshwater wetlands in the SWBP (Burbidge et al., 2010; Wallace et al., 2011)

Development of a composite index to map habitat suitability

Indicators were developed for each habitat component that could be mapped using GIS (Table 2). The resultant habitat indicators were classified as either ‘constraint’ or ‘factor’ criteria. Constraint criteria were considered necessary to site suitability and were regarded as Boolean criteria, so that potential sites that met the requirements for a particular criterion were given a value of 1, and sites that did not meet the criterion were scored as 0 (Eastman et al., 1995). Factor criteria define some degree of suitability for all areas, giving an area a value ranging between 0 and 1.

Table 2. Criterion and data sources used to define habitat suitability across the study area for Pseudemydura umbrina
Habitat componentCriteria typeaData sourcesConstraint criteria: prerequisites for site suitability bFactor criteria: scores for determining relative suitability of areas satisfying all prerequisitesc
  1. aCriteria for each of the habitat components were classified as ‘constraint’ (denoting criteria that were regarded as prerequisites for site suitability, and which were used to exclude all unsuitable areas), ‘factor’ (denoting criteria for which a range of habitat attributes were considered suitable, with some attributes more suitable than others) or ‘both’ (in these cases some habitat attributes were considered constraints, and others as factors).
  2. bDuring the first stage of spatial analysis, areas that did not satisfy all of the constraint criteria were deemed unsuitable and excluded from further analysis.
  3. cFor each habitat component, areas that did not satisfy any of the factor criteria received a score of zero for that component.
  4. dEach soil-landscape unit mapped by (Schoknecht, 2002) represents a mosaic of two or more soil types. The approximate composition of the mosaic is given as percentages for each soil type.
  5. eComprehensive spatial data on land use changes, food availability and baiting for predators were not available for the whole study area. These factor criteria were scored and assessed only for a small selection of wetlands identified as having the highest potential site suitability, which were assessed in the final stage of weighted summation. Similarly, site size was calculated only for the final selection of wetlands included in weighted summation.
Habitat components with data in GIS format (used to create composite suitability index)
Soil typeBoth
  • Soil-landscape maps at systems level (Schoknecht, 2002)
Presence of wet or waterlogged soil types within soil-landscape unitd.
  • 1 = Soil-landscape unit contains sandy duplex soil types

    0.5 = Soil-landscape unit contains loamy duplexes, clay, sandy or loamy earths, but not sandy duplex soil types

WetlandsConstraintGeomorphic wetlands Swan Coastal Plain (Hill et al., 1996), Augusta to Walpole (V&C Semeniuk Research Group, 1997) and statewide (GA, 2003.)Area lies within a mapped dampland, sumpland, palusplain or floodplain wetland typeN/A
Native vegetationConstraint
  • Native vegetation extent dataset (Beeston, Hopkins & Shepherd, 2002)
Native vegetation presentN/A
Vegetation structure and compositionBoth
  • State Pre-European type vegetation dataset (Beeston et al., 2002)
Area contains vegetation with suitable structure: woodland, shrubland, heathland, sedgeland or grassland
  • 1 = Vegetation composition dominated by Banksia or Melaleuca

    0.66 = Vegetation structure is shrubland, sedgeland or heath dominated by species other than Banksia and Melaleuca

    0.33 = Vegetation structure is woodland

Current land useBoth
  • Land use by Shire dataset (Beeston et al., 2002)
Land use of the area is not urban or rural residential, mining, Indigenous uses, manufacturing, industrial, services, transport, communication, airports, or ports
  • 1 = Area is designated for conservation

    0.66 = Area designated for natural feature protection, habitat management or natural resource management

    0.33 = Area designated for groundwater or minimum intervention

Distance from main roadsFactor
  • Geodata Topographic 1:250 000 dataset (GA, 2003)
  • 1 =  > 5 km from main roads

    0.5 = 2 to 5 km from main roads

Site sizeeFactor
  • Analysis completed within ArcGIS
  • 1 = site > 200 hae

    0.5 = site 50 to 200 hae

  • Soil-landscape maps at systems level (Schoknecht, 2002)
Soil-landscape unit contains > 10% by area of saline soil types.
  • 1 = Soil-landscape unit contains < 3% by area of saline soil types

    0.5 = Soil-landscape unit contains 3 to 10% of saline soil types

Habitat components without data in GIS format (researched after GIS analysis)
Land use changeeFactor
  • Local Town Planning Scheme maps (WAPC, 2011)
  • 1 = Area not within proposed development areae
Food availabilityeFactor
  • DEC WetlandBase (Storey, 1998; Halse et al., 2002; Lane et al., 2010; DEC, 2012a)
  • 1 = 10 or more macroinvertebrate species present including 10 or more species recorded as part of P. umbrina diete

    0.5 = 10 or more macroinvertebrate species present, but without 10 or more species previously recorded in P. umbrina diete

  • Fox and cat baiting location maps (Western Shield, 2011)
  • 1 = Baiting undertaken for foxes and catse

The constraint criteria were used to create a suitability mask in ArcGIS 10.1 (ESRI, 2012), following a similar approach to that outlined by Clifton & Boruff (2010). Spatial data for each of the six constraint criteria (presence of native vegetation, vegetation structure and composition, soil type, salinity, wetland type and current land use) were converted into a raster grid with a 30 × 30 m cell size. For each criterion, raster cells were given a value of either 1 (suitable) or 0 (unsuitable). Raster grids for all six constraint criteria were combined using cell statistics in ArcGIS 10.1. The output raster grid was reclassified so that cells with a value of 1 for all six criteria were given a value of 1 and all other cells were given a value of 0. The output grid was used as a suitability mask for the identification of potentially suitable habitat in the SWBP and raster cells that did not return a positive value for all six constraint criteria were eliminated from further consideration.

A composite suitability index was created to show the degree of site suitability for all raster cells not eliminated from the study, using the five factor criteria for which spatial data were available (vegetation structure and composition, soil type, salinity, current land use and distance from main roads). Comprehensive spatial data were not available for food availability, land use change or predators; site size was assessed at a later stage (see below). For each of the five factor criteria, a raster grid was created with a 30 × 30 m cell size. Each raster cell was assigned a value between 0 and 1, with 0 being least suitable and 1 being most suitable, according to the suitability of the habitat criteria mapped for that cell (see Table 2 for further details of scoring). The five raster grids were combined using cell statistics to create a composite index of site suitability across the masked area within the SWBP, with the highest scoring cells receiving a score of 5 out of a possible 5.

Weighted summation, ranking and sensitivity analysis

The final stage was to assess and rank a selection of the most suitable translocation sites identified through the spatially explicit GIS model. As 15–30 captive-bred P. umbrina individuals are available for release each year (Burbidge et al., 2010), only a small number of sites could ultimately be used for assisted colonization. In addition, detailed wetland mapping and information on potential food sources for P. umbrina were only available for a subset of regions within the SWBP. For these reasons, only wetlands that received the highest possible composite suitability index score, which were located within the boundaries of detailed wetland maps, and that were known to contain suitable food sources for P. umbrina (Storey, 1998; Halse, Scanlon & Cocking, 2002; Lane, Clarke & Winchcombe, 2010) were considered for further assessment and ranking.

All nine factor criteria (Table 2) were incorporated into the combined ranking of suitable wetland sites and were weighted according to expert opinion. First, the most suitable wetlands identified through mapping were scored between 0 and 1 for each of the nine criteria in the analysis, with a score of 1 indicating maximum suitability. Second, the rankings of the nine factor criteria drawn from the four experts were converted into weights using the naïve method of Hajkowicz, McDonald & Smith (2000). Finally, the criteria scores and criteria weights derived from expert rankings were converted into a single score for each site using weighted summation (Heaton et al., 2008):

display math

with the following constraints:

display math
display math

where the overall site score (ui) equals the sum of all criteria weights (wj) multiplied by their respective criteria score (xi,j), provided that the sum of all weights equals 1 and each weight has a value between 0 and 1.

A sensitivity analysis was conducted to assess the robustness of the criteria weights and the degree of uncertainty in the site scores. Graymore et al. (2009) suggest sensitivity analyses for MCA should assess small changes in the weights. In this study, criteria weights were changed, one at a time, by 10% to test whether site rankings were altered. Any change in ranking would suggest a level of uncertainty in the value of the weights (Graymore et al., 2009). The final site scores were used to identify the most suitable wetlands for assisted colonization of P. umbrina, based on the habitat components assessed.


Spatial representation of composite habitat suitability index

Applying the suitability mask and composite habitat index across the SWBP using low-precision wetland data indicated that the greatest concentrations of potentially suitable habitats were located on the Swan Coastal Plain and the south coast of southwestern Australia. For this reason, only detailed maps based on high-precision wetland mapping from the Swan Coastal Plain (Davis et al., 1996) and the Augusta to Walpole region, encompassing much of the potentially suitable southern coastal area (V&C Semeniuk Research Group, 1997) are presented here.

On the Swan Coastal Plain, 25 808 ha of potentially suitable habitats were mapped, consisting of 15 682 discrete locations (Fig. 3). Mapping the composite index for the Augusta to Walpole region highlighted 39 727 ha of potentially suitable sites, comprising 6424 discrete sites (Fig. 4). Wetlands with the maximum composite habitat suitability index of 5 included 2906 discrete sites on the Swan Coastal Plain, 1166 discrete sites in the Augusta to Walpole region and 689 additional sites outside the boundary of the two areas with detailed wetland mapping, but within the SWBP study area. Three of the four current locations at which P. umbrina is found were rated as 5 out of 5 for habitat suitability, while one site scored 4 due to its proximity to a major road (Fig. 3). Many of the most suitable sites identified on the Swan Coastal Plain were small and highly fragmented, with an average size of 1.76 ha compared with 17.01 ha for the most suitable sites in the Augusta to Walpole region.

Figure 3.

Index of habitat suitability applied to wetlands on the Swan Coastal Plain.

Figure 4.

Index of habitat suitability applied to wetlands between Augusta and Walpole.

Twelve wetlands with the highest suitability scores and that did not already contain populations of P. umbrina were selected for the third stage of the analysis (sites shown on Figs 3 and 4). These sites were selected based on the total continuous wetland area, and on the availability of data on food resources, feral predators and projected land use change. One site (Owingup Swamp) was outside the boundaries of our detailed wetland mapping but was included because it appeared highly suitable and food, predator and land use data were available.

Criteria weighting, weighted summation and sensitivity analysis

There was appreciable variation in the criteria rankings performed by the P. umbrina experts and in the average weight attributed to each criterion (Table 3). For example, ‘distance from main roads’ was ranked highest by one expert, but lowest by another. When weightings were averaged across the four experts, ‘food availability’ was regarded as most important and ‘distance to main roads’ as least important. There was only a slight difference (5.9%) between the weights of the highest and lowest priority criteria, which made it difficult to identify any particular criterion as most important (Table 3). However, ‘land use change’, ‘soil type’ and ‘distance from main roads’ had the highest variation in weights across the four experts, indicating disagreement on their relative importance.

Table 3. Rankings and weights attributed to habitat components by four experts
Expert numberRankings attributed to habitat components assesseda
Soil typeVegetation compositionCurrent land useLand use changeFood availabilitySite sizeDistance to main roadsPredatorsSalinity
  1. aFour experts independently ranked defined habitat components and associated criteria in their perceived order of importance to the survival of P. umbrina at a release site. The most important factor(s) received a ranking of 1, and the least important a ranking of 9.
  2. bAverage criteria weights calculated based on the naïve method outlined in Hajkowicz (2008).
Average criterion weightb0.1260.1050.0850.1290.1410.0840.0820.1180.118

The final suitability scores for the final 12 sites, based on nine factor criteria and calculated using the weighted summation method are shown in Table 4. The Chesapeake Road wetland was ranked as the most suitable translocation site based on the criteria assessed in this study. The three sites on the Swan Coastal Plain (Kalgup Road, Fernwood Farm and Gingin Brook) had the lowest rankings of the 12 selected ‘highly suitable’ sites. The sensitivity analysis performed on the criteria weights showed that changing the criteria weights by ± 10% led to minimal change in the order of site rankings. A change in the weight of the ‘vegetation composition’ criterion did result in a difference in the ranked order of sites, as two sites (Alamein Track and Doggerup Creek) switched ranking order. However, as this was the only change in ranking, there was generally high certainty in the ranking outcomes.

Table 4. Results of weighted summation for 12 potential translocation sites
 Habitat criteria scoresa and weights based on expert ranking
Site nameSoil typeVegetation compositionCurrent land useLand use changeFood availabilitySite sizeDistance from main roadsPredatorsSalinityTotal site scoreSite rank
  1. Values in bold are the weighted importance of each criterion calculated using expert opinion, 1 being the most important and 0 being the least important. These were used in the weighted summation equation to calculate the final site score.
  2. aScores for individual habitat criteria range from 0 (least suitable) to 1 (most suitable).
  3. bThe higher the weight, the more important the criterion to site suitability.
Chesapeake Road11111111111
Kodjinup Melaleuca Swamp111110.51110.962
Marron Road1111110.510.50.903
Lake Poorginup10.331110.51110.894
Owingup Swamp10.66110.510.5110.855
Alamein Track0.510.661100.5110.787
Doggerup Creek10.33110.50.50.5110.788
Meerup River10.330.3310.501110.729
Fernwood Farm10.330.331100.5010.6210
Kalgup Road10.330.3310.500.5010.5511
Gingin Brook10.3301101000.5112
Criterion weightsb0.1260.1050.0850.1290.1410.0850.0820.1300.118  


Identification of potential habitat for assisted colonization of P. umbrina

Assisted colonization is a recent addition to the conservation toolbox, and instances where it may provide the only means of maintaining wild populations will become increasingly common. Our results show that assisted colonization is a potential conservation measure for P. umbrina as the most suitable wetlands identified were several hundred kilometres south of the species' documented historical range. The few wetlands on the Swan Coastal Plain ranked as highly suitable included existing habitats at Twin Swamps, Mogumber and Moore River Nature Reserves. This cross-validation suggests that our MCA provides a realistic assessment of habitats that would give the greatest chance of a successful translocation.

The question of whether species will persist within ecosystems depends not only on the species, but also on the characteristics of individual ecosystems and their connectivity (Roe & Georges, 2007). In our case study, regions with high wetland connectivity would allow individual P. umbrina to move between wetlands without risking desiccation during transit (Burbidge, 1981). Wetland connectivity was not directly assessed as part of our MCA, but is an important consideration when deciding on the suitability of individual identified sites for assisted colonization. Few intact interconnected wetlands remain on the Swan Coastal Plain – an area that has experienced significant increases in human density and land clearing in the past century. Translocations of P. umbrina within this region may require private or government-owned land to be revegetated and restored to increase connectivity.

In contrast, a number of suitable wetlands are clustered together in the Augusta to Walpole region. This region retains a high proportion of remnant vegetation, including important conservation areas such as the D'Entrecasteaux National Park and the Ramsar-listed Muir-Byenup wetland system (DEC, 2012b). In this setting, long-term maintenance of hydrological and ecological connectivity is likely. The high conservation significance of many of the potentially suitable wetlands in the Augusta to Walpole region means that final site selection may favour wetlands that occur outside of a National Park or other significant habitat (Harris et al., 2013).

Several habitat components viewed by experts as important to successful translocations were not incorporated in our analyses due to a lack of data at the required spatial scale. These included factors important to ecological integrity, such as the presence of competing species (e.g. the oblong tortoise Chelodina colleri) or a threatened ecological community (DSEWPaC, 2012). Possibly the most important elements omitted from this study were the hydrological characteristics of the wetlands. Appropriate wetland hydroperiods and water temperatures will be critical to translocation success for P. umbrina because they influence tortoise activity and growth (Mitchell et al., 2012, 2013). Screening of the ecohydrological suitability of the SWBP under projected climate change scenarios (Mitchell et al., 2013) identified geographical regions that were similar to those we identified using the MCA approach. As more detailed spatial analysis of wetland hydroperiods becomes available, it will be possible to incorporate hydrological characteristics within the composite index of site suitability.

Combining quantitative and qualitative data in multiple criteria analysis to identify translocation sites

The desktop GIS-based analysis demonstrated here can be used as a first step in narrowing down a suite of potentially suitable areas that warrant further investigation. For quantitative data, we relied heavily on low-cost (or free), publicly available spatial data analysed within a GIS framework. The use of GIS to map species habitats can be cost-effective relative to conducting extensive field surveys for developing species distribution models (Smith et al., 2007). This is particularly important for species that are specialized to habitats that are difficult to find using broad-scale field surveys (Burbidge, 1981).

To confirm the accuracy of mapped data, some of which may have been derived from remote sensing or inferred from other data sources such as geological mapping, site visits are essential for confirming that an identified site constitutes suitable habitat. Ground-truthing or verification with recent satellite imagery can be used to test the accuracy of a model using an error matrix (Smith et al., 2007). An error matrix can compare suitability identified in the field with that identified in the suitability index, giving an idea of mapping accuracy (Poulin et al., 2002). This is a logical future step for the MCA model used in this study, as it would give a clearer indication of the potential of this method for identifying future translocation sites for P. umbrina.

In cases where published or long-term monitoring data are scarce, expert opinion may be an important source of information for environmental management decisions; in some cases, it may be the only available source of information. Using expert opinion in decision making can increase the inaccuracy of results as opinions may be biased or poorly calibrated (Martin et al., 2012). For example, Hajkowicz (2008) found high levels of disagreement among stakeholders who were asked to rank environmental problems in the Whitsunday Islands in Queensland. In the current study, our sensitivity analysis indicated that the overall rankings of sites were consistent, but there were discrepancies in the weighting allocated to different habitat criteria by the four experts. One expert ranked proximity to main roads as one of the most important habitat criteria, which may have been related to personal experiences of finding dead tortoises on roads adjacent to an existing reserve. The other experts, having little or no exposure to this experience, ranked this criterion as far less important to species survival.

Martin et al. (2012) suggest a mechanism to deal with uncertainty among experts, by allocating different weights to the opinions of different experts, depending on how accurate a decision maker believes an expert's views to be. In the case of identifying translocation sites for P. umbrina, experts could be weighted by their years of experience working with the species, or by the nature of the research they have conducted. An additional difficulty with using expert opinion for calibrating criteria weights in studies similar to ours is that for many range-restricted species for which assisted colonization might be considered, there are likely to be only a small pool of experts available for canvassing, as was the case in this research.

An alternative MCA method, other than weighted summation, could have been applied to our decision problem. The analytical hierarchy process is another widely applied method that uses pairwise comparisons and compares criteria and options in every unique pair to produce criteria weights and scores (Hajkowicz & Collins, 2007). However, the tangible benefit of applying weighted summation to the MCA model in this study is not strictly in identifying an optimum site for assisted colonization, but in providing a transparent process that can be understood by a range of stakeholders who will be affected by the model's results (Hajkowicz, 2008). Indeed, for controversial projects such as translocating species to habitats of high conservation value, extensive and transparent stakeholder consultation will be an important step in the selection of final release sites (Burbidge et al., 2011; Harris et al., 2013).


The combination of GIS-based analysis, informed by expert opinion and weighted summation in a spatially explicit MCA can identify potential translocation sites that are well matched to our current understanding of habitat requirements for a species. Our results emphasize the importance of considering multiple criteria in the decision-making process. These could include criteria relevant to the anticipated impact of the introduced species to the recipient community, which would be critical to minimising the risks posed by assisted colonizations. Further, when considering the translocation of species, regardless of whether it is within its historical range or to a new area, it is important to consider the suitability of the site based not only biotic factors, but also on socio-economic considerations. Such additional factors are readily incorporated into a MCA framework.

Our results should not be taken as final recommendations for sites for assisted colonization of P. umbrina as they have not been ground-truthed or subjected to logistical considerations such as cost-benefit analysis. However, as sites that provide a good match for existing habitat have been identified, it would be logical to consider these sites for assisted colonization over any sites within highly novel habitat. Before assisted colonization can occur, it will be important to project the effect of future climate scenarios on ecological and hydrological processes, which are key to the long-term survival of translocated species (Carroll et al., 2009). Fortunately, many habitat components (e.g. soil type, land use) will not be affected by climate change, and the advance of spatially explicit models of the biological impacts of climate change will allow climate-sensitive criteria to be evaluated and incorporated into MCA models. Rapid improvement in the resolution of GIS data is also likely in many parts of the world. The potential of a method similar to that demonstrated here in planning assisted colonizations is substantial, and when coupled with ground-truthing can allow more strategic deployment of species translocations than has previously been possible.


We thank the Western Australian Department of Parks and Wildlife and the Department of Agriculture and Food, particularly Geoff Banks, Ryan Vogwill and Phil Goulding for access to datasets used in this project. We are also grateful to the western swamp tortoise experts who participated in criteria selection and weighting (Andrew Burbidge, Gerald Kuchling, Don Bradshaw and Sophie Arnall) and to Hasnein bin Tareque, Cathy Cao and the Western Swamp Tortoise Recovery Team for their advice and support. Two anonymous reviewers provided valuable feedback that improved the paper. Our research was funded by an Australian Research Council (ARC) Linkage grant (LP0990428) to N.J.M., and by the Schools of Earth and Environment and Animal Biology at The University of Western Australia.