Notice: Wiley Online Library will be unavailable on Saturday 27th February from 09:00-14:00 GMT / 04:00-09:00 EST / 17:00-22:00 SGT for essential maintenance. Apologies for the inconvenience.
Species distribution models (SDMs) generally use correlative relationships between the species location and the associated environment to project the species potential distribution under climate change. While projecting a future suitable climatic space is relatively simple using SDMs, predicting a species ability to occupy that space relies on understanding dispersal capacity; a lack of knowledge about species-specific dispersal ability, varying geographical contexts and technical constraints of simple SDMs has limited the consideration of dispersal in most studies. We review the current treatment of dispersal in SDM studies addressing the effects of climate change and explore how incorporating ‘partial-dispersal’ scenarios could lead to more realistic projections of species distributions into the future.
We consider the implications for projected distributions of incorporating full- and no-dispersal scenarios in SDMs and identify a range of methods and their associated information needs for implementing partial-dispersal scenarios.
While simplistic and easy to implement, full- and no-dispersal scenarios are only realistic in a few situations. Although implementing partial-dispersal scenarios may require information that is lacking for many species, we argue that even relatively simple partial-dispersal models, with fairly basic knowledge needs, improve projections of altered distributions under climate change. More complex models, using more sophisticated modelling approaches, have been tested in a few cases and provide robust projections.
While climate change SDM outputs have proved useful, we highlight that careful selection of dispersal scenarios, relevant to the particular questions being addressed, is necessary for appropriate interpretation of the model outputs when projecting into novel environments (e.g. future climates). A number of methods have been developed for incorporating partial-dispersal scenarios in SDMs; however, the data and computation requirements currently limit their application to large numbers of species, highlighting the need for other techniques and generic user-friendly modelling platforms.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Species distribution models (SDMs) are widely used to understand the potential consequences of climate change for the distribution of species (Pearson & Dawson, 2003; Araújo et al., 2005; Martinez-Meyer, 2005; Hijmans & Graham, 2006; Jimenez-Valverde et al., 2008; Thuiller et al., 2008; Elith & Leathwick, 2009; Fitzpatrick & Hargrove, 2009), to project future distributions of invasive species (Thuiller et al., 2005ab) and aid in conservation decision-making (Loiselle et al., 2003; Sinclair et al., 2010; Shoo et al., 2011). While mechanistic modelling approaches based on physiological processes exist (see Kearney & Porter, 2009; Kearney et al., 2010), most applications of SDMs are correlative (phenomenological) due to their accessibility, relative ease of use, minimal data needs and availability of modelling algorithms (Weller, 1999). Correlative SDMs require only occurrence data and information on the ‘environment’ at these locations to model the distribution of a species, which can be used to formulate hypotheses about a species niche (Jimenez-Valverde et al., 2008). This model can then be used to project the potential distribution, which correspond to areas in geographical space that have suitable conditions for a species (Araújo & Guisan, 2006).
Climate warming is predicted to be occurring at a mean global velocity of 0.42 km per year (0.11–0.46 km per year) (Loarie et al., 2009), and species are expected to respond to such changes by shifting their geographical distributions to follow suitable climatic conditions (Collingham & Huntley, 2000; Tingley et al., 2009). Although adaptation in situ does occur (Huntley et al., 2006), species have historically responded to rapid rates of climate change by shifting, expanding or contracting their geographical distribution (Huntley & Thompson Webb, 1989; Davis & Shaw, 2001; Huntley et al., 2010). Present-day climate change is already having a noticeable impact on the distribution of species climatic niche space (VanDerWal et al., 2013) and species distributions, with latitudinal and altitudinal range shifts increasingly documented (Parmesan & Yohe, 2003; Parmesan, 2006; Chen et al., 2011). However, while projecting a future suitable climatic space is relatively simple using SDMs, predicting a species ability to occupy that space relies on understanding inherent dispersal capacity as well as barriers and conduits to dispersal. Due to their accessibility and their frequent use, it is imperative that the assumptions of SDMs are fully considered in interpreting results. Failure to adequately consider an appropriate dispersal scenario is often highlighted as a limitation to SDM climate change projections (Weller, 1999; Thuiller et al., 2008).
Most SDM studies that make future predictions of species under climate change use full- or no-dispersal scenarios (Peterson et al., 2002; Araújo et al., 2004; Thomas et al., 2004; Guisan & Thuiller, 2005; Thuiller et al., 2005a,b; Pearson, 2006). A search of the ISI Web of Science using the key words ‘biodiversity’ and ‘climate change’ and ‘model*’ (restricted to 33 international journals most commonly publishing projections of biodiversity under climate change) revealed that the majority of SDM studies solely used full-dispersal scenarios in climate change projections (Fig. 1). There are a small fraction of species without any limitations on dispersal and colonization and, equally, there are some species that completely lack the capacity to disperse (Weller, 1999). In reality, species have varying degrees of dispersal ability, and observed rates of distribution shift vary greatly both among species and across taxonomic groups (Walther et al., 2002; Berg et al., 2010; Chen et al., 2011). Despite this, the ability to simulate realistic dispersal scenarios under climate change has been limited by: (1) lack of knowledge about the dispersal abilities of species; (2) varying historical, geographical, anthropogenic, temporal and ecological mechanisms that are unique to the region a given species will need to disperse to or within; and (3) technical constraints of the modelling (McConkey et al., 2012). As there is a no ‘one size fits all’ approach, and ecological knowledge is needed, most SDM studies fall back on all or nothing dispersal scenario approaches (Fig. 1). Recently, however, researchers have begun using a variety of approaches in an effort to incorporate more realistic dispersal into SDMs (Sorenson et al., 1998; Midgley et al., 2006; Gaston et al., 2008; Dullinger et al., 2012; Reside et al., 2012)– often termed ‘partial dispersal’ scenarios.
Here, we review how correlative dispersal scenarios have been applied to date in SDM research and provide examples for a range of climate change–specific topics where dispersal ability should be considered. Our review provides the reader with the broad range of options for consideration when choosing the best dispersal scenario for their predictive models. In addition, we discuss the current trend in the use of ‘hybrid’ models that incorporate demographic or life history factors into SDM studies and highlight the simple or complex methodologies these models require (Table 1). The implications of different dispersal scenarios and their associated information requirements are considered.
Table 1. Examples of studies incorporating partial-dispersal with SDMS for climate change modelling
No of species
Proteaceae plants in South Africa Midgley et al. (2006)
Full, none and partial (simple)
Migration rate per decadal time slice; dispersal distance determined broadly based on dispersal vector
Australian tropical savanna birds Reside et al. (2012)
243 (28 partial-dispersal scenarios)
Full, none or partial (simple)
Species assigned to a particular scenario depending on level of habitat specificity; dispersal distance assumed 3 km per year for all species in the partial-dispersal scenario
Selecting protected areas to accommodate species range shifts with climate change Hannah et al. (2007)
1695 (495 partial-dispersal scenarios)
Full or partial (simple)
Time step restricted dispersal further restricted by known dispersal distance radius, different assumptions for different taxa
Distance decay dispersal function driven by abundance near the range boundary and habitat beyond the range boundary; using SHIFT
Dynamics of range margins for meta-populations of Lagomorphs Anderson et al. (2009)
Dispersal modelled via exponential function based on distance between suitable patches; three dispersal rates for each scenario (high, medium and low); incorporates a spatially explicit meta-population model
Dispersal distance assigned to each species based on seven dispersal ‘types’; implemented in MigClim
Persistence of an Oak tree under climate change Conlisk et al. (2012)
None and partial (complex)
Dispersal rate dependent on distance between patches rate of dispersal declines with increasing distance between patches; average and maximum dispersal distances applied; within-patch and between-patch dispersal; 30 dispersal scenarios used in model, three in sensitivity analysis (no-dispersal, low, high) and dispersal as a function of carrying capacity of receiving patch
Full-dispersal scenarios in SDMs are the most commonly applied in climate change research (Fig. 1). This scenario assumes that a species (or taxa, community, etc.) could reach all geographical areas that are predicted to be suitable in future climate projections (Franklin, 2010). This scenario is generally the simplest to apply, often being the default output in SDM programmes. As a result, many studies provide no justification for applying the full-dispersal scenario, and it is common for no mention of dispersal processes to be made. Despite this, specific life history traits and biogeographical barriers can limit the ability of a species to migrate or disperse (Clark et al., 2003), and full-dispersal is likely to be unrealistic except for the most vagile species (Barbet-Massin et al., 2012). For instance, the failure to keep pace with shifting climates has been highlighted across many taxa including trees (Conway, 2008), reptiles and amphibians (Araújo et al., 2006), butterflies (Menéndez et al., 2006; Devictor et al., 2012) and birds in Europe (Roche et al.,; Farmer & Parent, 1997). Mobile marine species (Robinson et al., 2011), larger mammals and migratory birds (Sutherland et al., 2000) are more likely to keep pace with climate change via dispersal compared with plants, reptiles and amphibians. Despite this expectation that large ranging and mobile species will have the greatest dispersal ability, actual shifts exhibited by species in response to climate change may rely on many factors, such as availability of suitable habitat, degree of habitat fragmentation and biotic interactions. There is some empirical evidence from recent distributional range shifts, suggesting that variation within a taxonomic group can be large and that the specific ecology of each species should be assessed where possible (Chen et al., 2011).
Biogeographical barriers such as oceans, mountain ranges and deserts may limit dispersal (Pearson & Dawson, 2003; Guisan & Thuiller, 2005). The relative likelihood of dispersal across large natural barriers will depend on both the scale and magnitude of the barrier and its relation to the dispersal ability of the species. It might be unrealistic for a small reptile to cross an ocean, but reasonable for a migratory bird. Species specialized to a particular habitat type will be limited to the rate of change and establishment of that habitat in new areas (Huntley, 1995). Therefore, the use of a full-dispersal scenario can be detrimental to conservation applications where gains in range are overestimated, and losses underestimated (Midgley et al., 2006).
Although there are limitations to full-dispersal scenarios, they may be appropriate in applications where understanding the potential distribution is the desired output. This includes situations where pre-emptive conservation strategies are necessary, such as assisted migration (Hoegh-Guldberg et al., 2008; Willis et al., 2009) or the identification of future potential refugial areas, restoration sites and land acquisition (Shoo et al., 2011; Anderson et al., 2012). In addition, long-distance dispersal does occur and stochastic events such as hurricanes or altered circulation patterns could bring species long distances to areas they have not occurred in before (Hellmann et al., 2008), assuming that enough individuals will disperse to form a viable population. In this case, full-dispersal scenarios would allow for these potential sites to be identified to aid in future ecological surveys that may allow for the discovery of new populations (Raxworthy et al., 2003).
Understanding the potential range expansions of invasive species will likely require the use of full-dispersal scenarios. Some species introductions are facilitated by human movements (Hellmann et al., 2008) and thus have a greater chance of filling their potential distributions. Invasive species examples globally highlight this; the Asian house gecko (Hemidactylus frenatus) is an avid stowaway that is well adapted to human habitats and has undergone a large global range expansion due to human facilitated movement (Hoskin, 2010), while the introduced cane toad (Bufo marinus) is adept at using roads to speed up its range expansion in Australia (Brown et al., 2006). Purposeful movements of species outside of their native ranges by humans to ‘rescue’ them from climate change have also been documented (McLachlan et al., 2007). Thus, full-dispersal scenarios may be useful in such cases to determine the potential impact of these invasive species (Thuiller et al., 2005a,b).
To deal with the inherent limitation of the full-dispersal scenario, no-dispersal scenarios are often applied to future projections in conjunction with full-dispersal scenarios (Fig. 1). The justification being that the real outcome is likely to be somewhere between the no- and full-dispersal scenarios (Pitelka, 1997; Coetzee et al., 2009; Marini et al., 2009), and thus, the full range of possible distribution changes have been assessed (Pearson et al., 2006). Like full-dispersal, no-dispersal scenarios are a commonly applied method due to their simplicity, although they are also unrealistic for most species. No-dispersal scenarios are generated by restricting future predictions of suitable habitat to the extent of the current distributions. Thus, by applying a no-dispersal scenario, it is assumed that future sites that might become suitable for a species remain unoccupied. The assumption that a species will not move from its current range may generate overly pessimistic predictions of range losses in future climates for many species. For example, where no-dispersal scenarios have been applied to future predictions of birds (Jetz et al., 2007; Coetzee et al., 2009; Marini et al., 2009), the chance of over-stating extinction risk is high due to the recorded distributional plasticity of many bird species (Brown et al., 2001; Tingley et al., 2009).
No-dispersal scenarios have been used as the ‘worst case scenario’ in SDMs; however, they are appropriate for some analyses, mainly where species lack the capacity to disperse due to intrinsic factors, and when a species is unable to colonize a new area due to extrinsic factors. Poor dispersers may include reptiles and amphibians, for whom dispersal distances are often small (tens of metres), and dispersal events result in a high risk of mortality (Bonnet et al., 1999; Massot et al., 2008). Climate change and increased temperatures are predicted to further inhibit dispersal ability in these taxa (Massot et al., 2008). No-dispersal scenarios may also be appropriate for projecting future distributions of slow dispersers when using coarse grid scales. Many climate change prediction models use grid sizes up to 50 × 50 km; therefore, no-dispersal scenarios more realistically reflect the effect of climate change on the ranges of species that migrate slowly over smaller scales.
Extrinsic factors preventing dispersal include geographical barriers such as mountain peaks, where no higher altitude is available. Alpine species restricted to cooler, high altitude mountain tops are better represented by a no-dispersal scenario, as these species lose suitable habitat as climate space shifts upslope (Walther et al., 2002; Parmesan & Yohe, 2003; Parmesan, 2006; Dirnböck et al., 2011). For example, the mountain pygmy possum (Burramys parvus) is restricted to a small high altitude distribution in Australia that is predicted to disappear with small increases in global temperature (Brereton et al., 1995; Hughes, 2011). Mountain-top frog species already occupying the coolest refuges will be unable to disperse to cooler higher altitudes as climate space shifts upslope (Marra et al., 1998; Knudsen et al., 2011). In addition, species current distributions may be limited by factors beyond climate, and dispersal ability hampered by more than just landscape scale barriers (e.g. disturbances, habitat requirements interspecific interactions) (Huntley et al., 2010; Bateman et al., 2012). For example, habitat specificity limits dispersal where the habitat is clearly delineated. This is the case with soil or rock type for many mycophagous (fungus-eating) mammals in Australia (Laurance, 1997; Claridge, 2002), for reptiles such as the broad-headed snake (Hoplocephalus bungaroides) (Studds & Marra, 2007) and sand dune lizard (Uma inornata) (Barrows et al., 2010), as well as for salamanders restricted to single cave systems (Niemiller et al., 2010). In addition, human land-use and fragmentation may dramatically reduce habitat connectivity in some regions (Fischer & Lindenmayer, 2007), making no-dispersal scenarios appropriate for poor dispersers (Lu et al., 2012).
Where partial-dispersal models are applied in projecting changes in the distributions of species, ‘simple’ models that incorporate dispersal attributes to restrict or buffer an otherwise full-dispersal scenario are most common (Fig. 1, Table 2). More ‘complex’ models use relatively sophisticated modelling approaches that combine the outcomes of SDMs with one or a combination of dispersal capacity, demographic knowledge, habitat information, the temporal nature of climate change and other factors influencing capacity to occupy a changed distribution (Table 2). These complex models, sometimes referred to as ‘hybrid’ or ‘coupled’ models, have only been applied and tested in a few cases (Fig. 1), but are becoming increasingly accessible (Engler & Guisan, 2009; Franklin, 2010; Midgley et al., 2010; Engler et al., 2012).
Table 2. The degree to which full-, no- and partial-dispersal approaches consider important factors influencing the future distribution of species, where black dots indicate ‘almost always’ and grey dots indicate ‘sometimes’
Simple models of dispersal
There are several means by which basic attributes of dispersal can be incorporated into SDMs to improve projected outcomes compared with full- or no-dispersal models (Table 1). If maximum likely dispersal distance of a species is known, then this information can be used to determine whether a species will be able to reach a given projected potential distribution. This approach was taken by Lu et al. (2012), where known dispersal distances of threatened bird species were used to determine whether patches of suitable habitat in a fragmented landscape were able to be reached. If the species dispersal ability is unknown, other methods could be applied that account for estimated distances of dispersal, implemented in a time series to allow for a controlled dispersal projection. Future potential distributions of a species could be restricted by buffering a selected dispersal distance around the current distribution of that species for each future time period. Examples of this are presented in Midgley et al. (2006) and Williams et al. (2005) where different migration rates were selected for different plant seed dispersal agents, and these estimates were applied over several time slices throughout the projection. Potential migration rates for different species have been derived using paleo-data (Morin & Thuiller, 2009), or based on simple assumptions (Fitzpatrick et al., 2008). Hannah et al. (2002) and Reside et al. (2012) each modelled species distributions over several time steps, allowing species to reach suitable habitat within a dispersal radius based on ecological knowledge of the species being modelled. Although applying fixed dispersal distances may risk ignoring potential long-distance dispersal events, they are relatively simple to apply to many species (Fig. 1b) and can capture key elements of dispersal limitation over time.
The application of simple dispersal scenarios in predicting climate change outcomes for many species could be greatly facilitated by the development of new user-friendly modelling platforms. This software could combine correlative habitat suitability models, with simple dispersal models allowing automated application across many species, regardless of the type of taxa. It is possible that this objective could be implemented through adaptation of existing packages such as MigClim (Engler & Guisan, 2009; Engler et al., 2012) to be more taxonomically generic.
Complex models of dispersal
Several software packages are available to combine dispersal processes and demographic information with SDMs in projecting distributional changes under climate change (Table 1). The BioMove (Midgley et al., 2010) and MigClim (Engler & Guisan, 2009; Engler et al., 2012) packages are specifically designed for modelling terrestrial plant species, although MigClim is potentially general enough to be adapted to other taxa. BioMove incorporates SDM outputs with various demographic information (Midgley et al., 2010). Here, the dispersal ability of plant species is modelled by scaling demographic rates with the suitability of habitat, as well as incorporating disturbance and changes in vegetation structure. Iverson et al. (2004) incorporated SDMs with SHIFT, a programme that estimates dispersal abilities based on species abundances in the landscape and colonization probabilities. This model depends on the importance of species abundance at the boundary (termed source strength) and the density of the forest beyond the boundary (sink strength) to determine migration rates of trees. These modelling packages now allow the relatively simple incorporation of dispersal processes into climate change projections for individual species; however, the practical capacity of these approaches to evaluate more than a handful of species is unclear (Fig. 1b).
Some sophisticated modelling approaches have combined SDMs with spatially explicit meta-population models to project altered distributions of species over time, under climate change (Keith et al., 2008; Anderson et al., 2009; Engler & Guisan, 2009; Lawson et al., 2010; Conlisk et al., 2012; Dullinger et al., 2012; Fordham et al., 2012; Regan et al., 2012). These models project the abundance in one or more age classes for each grid cell at each time point, based on the abundance at the previous time point, dispersal from neighbouring grid cells and the predicted habitat suitability (from the SDM) (Clark et al., 2003; Franklin, 2010). By incorporating dispersal processes over time, these meta-population models provide both spatially and temporally explicit projections, allowing for the timing of major changes in the distribution of a species under scenarios of climate change to be clearly identified. Importantly, these models assume that the SDM predicted probability of occurrence for a species is directly linked to a number of important attributes, such as the initial population size in each grid cell, demographic rates and the carrying capacity for that species in each location over time (Keith et al., 2008; Dullinger et al., 2012). These key assumptions of spatially explicit meta-population modelling approaches remain largely untested.
Other recent advances in SDMs incorporate dynamic processes such as range expansions with demographic information to predict changes in relation to life history characteristics (Anderson et al., 2009; Brook et al., 2009). A significant drawback of these more complex models is that they require reliable information on demographic rates (e.g. reproduction, mortality) and the dispersal kernel for the species being modelled. Uncertainty regarding demographic parameters requires much greater modelling effort per species in exploring the possible range of outcomes for a species under climate change (e.g. Dullinger et al., 2012). In addition, important ecological factors may be oversimplified in demographic models, ignoring differential responses among individual life history stages, the effects of extreme and variable climate, regional differences in populations and demographics, the direct or indirect effects of climate on demography, and the influence of biotic interactions and how these factors affect dispersal success and population growth and establishment (Smolik et al., 2010; Conlisk et al., 2012; Dullinger et al., 2012; Fordham et al., 2012). The reliability of the models will therefore depend on the ecological relevance of the demographic information included (Fordham et al., 2012).
Although these coupled models – or hybrid models – show promise, caution is warranted; each modelling component contains its own assumptions and uncertainties. When these different modelling techniques are used in conjunction, the compounding effects of these uncertainties can put reliability of model outputs into question (Langford et al., 2011; Conlisk et al., 2013). Outputs from the different combinations of the components of a coupled model can yield vastly different results and predictions making it difficult for their use in management decisions (Fordham et al., 2012; Regan et al., 2012; Conlisk et al., 2013).
Modelling dispersal with knowledge shortfalls
There is a range of simple ways in which we can extend existing knowledge on dispersal abilities for a small number of species to the vast majority of species for which we have little information. A relatively simplistic approach is to use a realistic intermediate dispersal scenario for species within an assemblage based on average dispersal values for the taxa. For example, the development of allometric relationships between basic physiological properties (e.g. body size) and dispersal attributes generated for well-studied species could be used to estimate dispersal properties for poorly studied species (Thomson et al., 2011). In addition, well-studied species could be used to parameterize mechanistic models of dispersal for poorly studied species (Nathan et al., 2002; Murphy et al., 2008). Species could also be allocated to functional groups based on basic physiological dispersal attributes (e.g. seed size) or habitat specificity (Reside et al., 2012), with well-studied species in each functional group used as surrogates to provide key dispersal information for poorly studied species (Dennis & Westcott, 2006, 2007; Huntley et al., 2010). Alternatively, organisms can be grouped by factors such as common traits, life history or trophic position (Berg et al., 2010). For example, generalist species are often better dispersers than specialists, not based on dispersal capacity per se, but rather on the ability to occupy a broader ecological niche as generalists are less limited by the availability of specific resources (Berg et al., 2010; Sekar, 2012). Species in higher trophic positions are often better dispersers, as larger body sizes, and greater home ranges confer greater mobility than smaller, lower trophic position species (Berg et al., 2010; Angert et al., 2011). The next step would be to assess how these groupings could be paired with appropriate partial-dispersal scenarios. Meier et al. (2011) found that comparing projections from realistic dispersal kernels with those using full- or no-dispersal scenarios allowed for the identification of specific traits which could be used to suggest an appropriate dispersal scenario. Here, early-successional tree species were better at tracking climate change and did so at close to full-dispersal, while mid- and late- successional species responded slowly, and more in line with no-dispersal.
An appropriate surrogate for a dispersal scenario where actual dispersal capacity is uncertain could be generated by buffering the known distribution of a species by an appropriate distance from the range edge. The dispersal probability could be weighed by abundance or occupancy at the range edge, with higher abundances given a higher probability weighing. Iverson et al. (2004) demonstrated that abundance at range edge was more important than habitat availability beyond the edge in determining how far and fast North American trees migrate. If there is limited ecological knowledge about the species, then habitat suitability derived from SDMs could be used as a surrogate for abundance (Murphy et al., 2006; VanDerWal et al., 2009).
Dispersal and habitat contiguity
Most dispersal scenarios assume a uniform dispersal rate for all qualities of the matrix regardless of landscape factors (but see Meier et al., 2011). To account for this, distributions can also be clipped by biogeographical boundaries which are known to limit species (Williams et al., 2010) or by selecting only contiguous areas (Peterson et al., 2002; Peterson, 2003). Human-induced barriers, such as fragmentation, urbanization and anthropogenic habitat destruction, can also affect dispersal ability, as has been shown for butterflies and other insect species, reptiles and amphibians, and rainforest birds (Mattoni et al., 1997, 2001; Ojima et al., 1999; Collingham & Huntley, 2000; Araújo & Pearson, 2005; Russell et al., 2005; Araújo et al., 2006; Gaston et al., 2008; Southwood & Avens, 2010). For example, the movement to future climatically suitable locations for the endangered Quino checkerspot butterfly (Euphydryas editha quino) is blocked by anthropogenic land-use changes, such as urbanization and fragmentation, in the landscape (Mattoni et al., 1997, 2001). Dispersal ability can be particularly hampered when suitable patches are spaced widely apart or habitat availability is decreased (Collingham & Huntley, 2000). In such situations, land-use change data can be used to account for altered landscapes (Pearson, 2006), and connectivity via dispersal corridors (Graham et al., 2010) or chains of ‘stepping stone’ suitable habitat over time (Williams et al., 2005) can be accounted for (Collingham & Huntley, 2000; Meier et al., 2011). Here, probability of dispersal is influenced by neighbouring cells and the connectivity between them.
Here, we highlight that current climate change projections of species distributions often assume inappropriate dispersal scenarios, which can lead to unreliable projections of altered distributions. We examined how dispersal can be incorporated into SDMs in different ways, and our review outlines the appropriateness of different approaches to study species or systems with varying levels of knowledge. Partial-dispersal scenarios, uniquely tailored to each situation, are likely to provide the most robust projections of altered distributions under climate change. Although a number of methods have been developed for incorporating partial-dispersal in SDMs, the data and computation requirements can make their application to large numbers of species challenging. This highlights the need for continued development and refinement of techniques and generic user-friendly modelling platforms. Additionally, significant new research effort is required in collecting and collating dispersal data for a broad range of species, as well as the development of new methods of model parameter calibration to account for the lack of data availability that often prevents researchers from applying dispersal scenarios in SDMs. To provide the most useful models of climate change predictions for a species, selecting the most appropriate dispersal scenario is imperative.
HM and KM receive support from the CSIRO Climate Adaptation Flagship, and KM was additionally supported by a grant from the Australia–India Strategic Research Fund.
Brooke Bateman is a postdoctoral research associate in the SILVIS Lab (http://silvis.forest.wisc.edu/) at the University of Wisconsin-Madison and is a former member of and current collaborator with the Centre for Tropical Biodiversity and Climate Change Research (http://www.jcu.edu.au/ctbcc/index.htm) at James Cook University. Her current research interests are in spatial ecology and conservation, and the effect of extreme weather events on wildlife populations and communities.
Author contributions: J.VDW, B.L.B., A.E.R. and H.T.M. conceived the ideas; B.L.B., A.E.R. and K.M. led the literature search and analysis; B.L.B, H.T.M., A.E.R., K.M. and J.VDW were involved in writing and revising of the manuscript.