Notice: Wiley Online Library will be unavailable on Saturday 30th July 2016 from 08:00-11:00 BST / 03:00-06:00 EST / 15:00-18:00 SGT for essential maintenance. Apologies for the inconvenience.
Animal dispersal is highly non-random and has important implications for the dynamics of populations in fragmented habitat. We identified interpatch dispersal events from genetic tagging, parentage analyses and assignment tests and modelled the factors associated with apparent emigration and post-dispersal settlement choices by individual male agile antechinus (Antechinus agilis, a marsupial carnivore of south-east Australian forests).
Emigration decisions were best modelled with on data patch isolation and inbreeding risk.
The choice of dispersal destination by males was influenced by inbreeding risk, female abundance, patch size, patch quality and matrix permeability (variation in land cover). Males were less likely to settle in patches without highly unrelated females.
Our findings highlight the importance of individual-level dispersal data for understanding how multiple processes drive non-randomness in dispersal in modified landscapes. Fragmented landscapes present novel environmental, demographic and genetic contexts in which dispersal decisions are made, so the major factors affecting dispersal decisions in fragmented habitat may differ considerably from unfragmented landscapes. We show that the spatial scale of genetic neighbourhoods can be large in fragmented habitat, such that dispersing males can potentially settle in the presence of genetically similar females after moving considerable distances, thereby necessitating both a choice to emigrate and a choice of where to settle to avoid inbreeding.
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.
Dispersal is a key process in the ecology and evolution of natural populations and is a major determinant of the viability of populations in fragmented habitat (Reed 2004). Animal dispersal decisions are made in response to diverse influences, such as environmental and demographic conditions at natal and settlement locations, landscape connectivity, local kinship patterns (which relate to inbreeding risk and kin cooperation), and individual state variables like sex, body condition and heterozygosity (Barbraud, Johnson & Bertault 2003; Bowler & Benton 2005; Clobert et al. 2009; Selonen & Hanski 2010). Fragmentation changes the context for dispersal decisions in many ways, by modifying not only the spatial configuration of habitat, but also the demographic and genetic landscapes in which dispersal decisions are made (Dick, Etchelecu & Austerlitz 2003; Sumner 2005; Keyghobadi 2007). Thus, understanding complex, non-random movement patterns in fragmented landscapes requires consideration of multiple factors.
Key reproductive influences on mammalian dispersal include mate availability and inbreeding avoidance. In many mammals, males are more likely to disperse when per capita female abundance is low or relatedness to local females is high (Pusey & Wolf 1996; Lehmann & Perrin 2003; Gilroy & Lockwood 2012). Mate availability has also been demonstrated to influence the choice of post-dispersal settlement location by males (Fisher 2005; Glorvigen et al. 2012), although the role of inbreeding risk in settlement decisions has received less attention. Here, we hypothesize that males should favour settling in sites containing females to whom they are less related. This is most likely to occur when females prefer less-related males as sires (Pusey & Wolf 1996; Mateo & Johnston 2000; Parrott, Ward & Temple-Smith 2006; Bonadonna & Sanz-Aguilar 2012) and when the spatial scale of non-random genetic structure among potential mates (i.e. male-female relatedness) is large in relation to the scale of dispersal (Dick, Etchelecu & Austerlitz 2003; Banks et al. 2005a). The latter scenario might commonly occur in fragmented landscapes, where landscape-wide population density can be reduced, and gene flow is concentrated between a small number of sources and destinations (Gustafson & Gardner 1996).
In addition to reproductive factors, environmental variation at the patch and landscape level can influence dispersal decisions. Commonly, individuals are more likely to disperse from patches that are smaller, and those where resources are limiting (Kuussaari, Nieminen & Hanski 1996; Bowler & Benton 2009). In contrast, patches that are larger, or provide greater per capita resource availability, are often preferred as dispersal destinations (Kuussaari, Nieminen & Hanski 1996; Remy et al. 2011). Dispersal rates are also influenced by the landscape context of patches. Both immigration and emigration are often negatively correlated with patch isolation, which, in turn, can be mediated by geographic isolation from other patches (Bender, Tischendorf & Fahrig 2003) and landscape permeability to movement (Revilla et al. 2004; Prevedello & Vieira 2010).
In this study, we investigated how biotic and abiotic factors influence dispersal decisions by male agile antechinus (Antechinus agilis, a small Australian marsupial carnivore with predominantly male dispersal; Cockburn, Scott & Scotts 1985) in a highly fragmented landscape (Fig. 1). We collected data on individual dispersal decisions from genetic tagging, parentage analysis and assignment tests, and tested whether the decision to emigrate was influenced by inbreeding avoidance (relatedness to local females) or female abundance, patch size, patch quality or patch isolation. Subsequently, we tested whether inbreeding avoidance, female abundance, patch size and quality, or matrix permeability, influenced the choice of settlement location.
Materials and methods
Study area and species
The agile antechinus is a small marsupial carnivore (males 20–45 g, females 14–25 g). The species is semelparous and polygamous, with females giving birth to 8–10 young in September. Post-natal dispersal typically occurs in January and February and is strongly male-biased (Cockburn, Scott & Scotts 1985; Banks & Peakall 2012). The mating season occurs in August, after which all males die.
We sampled agile antechinus in 2002 and 2003 from a network of 27 sites forest within the Buccleuch State Forest, a plantation of over 50 000 ha of exotic Pinus radiata in southern New South Wales, Australia. The sites included 23 eucalypt forest patches within the plantation and four sites in a larger block of eucalypt forest on the plantation edge (Fig. 1). The agile antechinus is found in many of the patches (which range in area from less than 0·5 ha to over 100 ha), but not in the pine plantation itself, due to the absence of tree hollows that are required for shelter (Lindenmayer, Cunningham & Pope 1999). Although the extant patch populations show few genetic signals of isolation, the occupancy of remnant patches by agile antechinus is strongly influenced by patch isolation (distance to the nearest occupied patch) and the basal stand area of trees, with larger populations occurring in larger patches dominated by Eucalyptus viminalis or E. radiata (Banks et al. 2005b).
Sampling and identification of dispersal events
We used data on the natal and breeding locations of 472 individuals (333 females and 139 males, from a total of 811 sampled adults) to build a data set of interpatch dispersal events with three genetic approaches, including:
Genotype matching of adults and juveniles: We generated 15-locus microsatellite genotypes from DNA samples extracted from ear tissue (adults) or tail skin samples (pouch young). We used the Microsatellite Toolkit software (Park 2001) to match the genotypes of 341 adults to those of pre-dispersal juveniles of that generation. We sampled 620 juveniles as pouch young of females trapped approximately 1 month after the birth of that generation (October of the previous year; Banks et al. 2005c). The microsatellites used (Banks et al. 2005b) had high power to distinguish individuals, with an estimated probability of a genotype match between full siblings of 1·05 × 10−7. The method used in API-CALC (Ayres & Overall 2004) accounted for the genetic structure in the population (FST = 0·029, FIS= −0·019).
Maternity assignment: We used Cervus 3.0 (Kalinowski, Taper & Marshall 2007) to assign maternity of 53 adults for whom matching juvenile genotypes were not identified. This was conducted on the adults sampled in 2003 only, as we did not sample candidate mothers for the 2002 individuals. We conducted 10 000 simulations to identify critical delta LOD scores to distinguish candidate mothers. We estimated that we had sampled 80% of candidate mothers because our sample numbers represent 80–100% of the mark–recapture population size estimates for each patch (Banks et al. 2005b). The proportion of loci genotyped was 0·996, and the proportion of loci mistyped was estimated at 0·019 based on mother–offspring mismatches. We accepted assignments using a 95% confidence threshold for delta LOD scores. The average probability of non-exclusion of an incorrect mother using this approach was 4·5 × 10−6. We double-checked all genotype matches and maternity assignments involving mismatches at one or two loci by rescoring all loci for those individuals.
Assignment tests: We used GeneClass2 (Piry et al. 2004) to assign natal patches for a further 78 adults by comparing genotype likelihoods across the set of candidate source patches sampled. We used the assignment test algorithm of Rannala & Mountain (1997), with allele frequencies estimated from the genotypes of females sampled in that patch in that year. (We excluded males because they were likely to have been immigrants.) An individual was assigned to a patch if its genotype likelihood in that population was greater than 95% of the summed likelihoods in the five most likely populations. If the true population of origin was not sampled, it is possible that an individual could be incorrectly assigned to the most likely population. To minimize this risk, an assignment was not accepted if the genotype was excluded from that population with a probability of 95% or greater according to the Monte Carlo resampling test of Paetkau et al. (2004), that creates a distribution of genotype likelihoods of simulated philopatric individuals.
Our data on individual movement histories (Banks, Lawson & Finlayson 2013) can be considered to represent ‘apparent dispersal’, similar to the concept of ‘apparent survival’ in mark–recapture demographic analyses, in that they represent dispersal events, conditional on the sampled individuals having survived and been captured in the study area. Note that we did not consider intrapatch dispersal, as most patches are smaller than a single home range (agile antechinus can forage outside eucalypt forest patches).
Statistical analysis of apparent emigration
We investigated what influenced the probability that a sampled individual had emigrated from its natal patch. The list of candidate explanatory variables is presented in Table 1. Because our raw data showed that less than 2% of females (6 of 333 individuals) dispersed from their natal patch, we conducted this analysis for the data on males only, as we considered the dispersal records for females to be insufficient for robust analysis.
Table 1. Variables used in statistical modelling of emigration and dispersal destination decisions. Variables in the same ‘set’ were not fitted in the same model but evaluated separately, as they were alternative representations of the same underlying factors
Minimum relatedness (Queller & Goodnight 1989) of a focal male to the females in that patch
Patch demography variables
Estimated patch population size from mark–recapture modelling (Banks et al. 2005b)
Estimated number of females in that patch from mark–recapture modelling (Banks et al. 2005b)
Sex ratio of juveniles (pouch young) sampled in that patch in the same generation
Patch habitat metrics
Basal stand area of trees in m2 ha−1
1. Eucalyptus camphora gully
2. Mixed species gully
3. Mixed species slope/ridge
Patch isolation metrics
Patch area in hectares
Distance (m) to the nearest occupied patch
Summed patch population sizes within a 2830 m buffer of the focal patch. 2830 m was the mean distance moved by dispersing males.
Summed patch population sizes within a 1392 m buffer of the focal patch. 1392 m was the mean distance moved by all males.
Summed patch population sizes within a 2830 m buffer of the focal patch, weighted by the inverse of the distance to each patch. 2830 m was the mean distance moved by dispersing males.
Summed patch population sizes within a 1392 m buffer of the focal patch, weighted by the inverse of the distance to each patch. 1392 m was the mean distance moved by all males.
Interpatch connectivity metrics
Euclidean distance between each pair of patches.
CCS1 – CCS36
Pairwise ‘resistance distances’ between sites generated in Circuitscape (McRae 2006). 36 variables were generated representing distances between sites in landscapes where native Eucalyptus forest had a resistance (to movement) value of 1, and pine plantation, cleared farmland and roads had all possible combinations of resistances of 1, 2 or 10.
Candidate explanatory variables relating to reproductive factors included the number of females at the natal site and relatedness to females at the natal site. Relatedness may be associated with dispersal in several ways. First, elevated relatedness to potential mates may indicate an inbreeding risk, thereby favouring dispersal (Lehmann & Perrin 2003). The variable we used to test for this effect was the minimum relatedness of each male to the females in his natal patch (high values mean an absence of unrelated females: Table 1). Secondly, relatedness may be a consequence of dispersal patterns. Dispersal patterns can be highly consistent between individuals from the same site (Penteriani & Delgado 2011). We assumed there would be a negative relationship between the probability of emigration from a particular site and the relatedness between individuals in that site, so we included a variable representing the mean relatedness among individuals sampled at that site. This variable was not of intrinsic biological interest, but we wished to account for this non-independence of relatedness and dispersal before testing for other effects such as inbreeding avoidance.
Environmental variables that we evaluated included patch size (unfragmented sites were allocated the same area as the largest patch: 95 ha) and two descriptors of vegetation that are known to be important descriptors of habitat quality and resource availability for the agile antechinus in this ecosystem. These were basal stand area of trees and three combinations of forest type and topography (Banks et al. 2005b). We also evaluated several patch isolation metrics based on distance to the nearest occupied patch, abundance of antechinus within a buffer radius, or a combination of both (distance-weighted abundance within a buffer radius: Table 1). We used two different buffer radius thresholds to investigate the spatial scale over which patch isolation affects dispersal (Table 1). We also tested for effects of demographic variables such as population size and the pre-dispersal sex ratio estimated from sampled pouch young.
We commenced our analyses by fitting generalized linear models with the logit link function fitted in R 2·15 (R Development Core Team 2012) to our data. We did not evaluate all combinations of variables, as some sets of variables (indicated in Table 1) included several alternatives for testing the same hypotheses (e.g. the patch isolation metrics), one of which was only ever fitted in a single model, and some were always fitted together (e.g. the relatedness metrics). Given these restrictions, we worked through candidate models and ranked them by the Bayesian information criterion (BIC) (Schwarz 1978). We chose BIC over AIC as it imposes a heavier penalty on adding terms to a model and identifies more parsimonious models. Since our aim was to interpret the effects of our explanatory variables (not to maximize model predictive power), we used BIC. Following the identification of the most parsimonious model, we tested for random effects of site and year using generalized linear mixed models with the logit link function fitted using lme4 (Bates, Maechler & Bolker 2012) in R 2.15 (R Development Core Team 2012).
Statistical analysis of settlement choices
We used multinomial logistic regression in the mlogit package (Croissant 2011) in R 2.15 (R Development Core Team 2012) to investigate factors associated with the dispersal destination choices made by the sampled males from a set of alternative patches (including the natal patch). Candidate explanatory variables in the modelling are shown in Table 1. These included the relatedness variables used to estimate inbreeding risk (rmin(F) and rmean) and demographic and environmental variables measured for each patch. Instead of testing for patch isolation (as used in the analysis of emigration), we tested a set of interpatch connectivity metrics (Table 1) representing the resistance distance between patches when each landscape element (eucalypt forest, pine plantation, cleared grazing land, roads) was allocated a different resistance to dispersal (McRae 2006; McRae et al. 2008). Briefly, Eucalyptus forest had a resistance (to movement) value of 1, and pine plantation, cleared farmland and roads had all possible combinations of resistances of 1, 2, or 10, respectively. We fitted the multinomial models to test for generic effects of explanatory variables across all alternatives (i.e. without alternative-specific coefficients). We used the BIC model building and selection protocol described in the previous section, using Euclidean distance between sites (DEuclidean: Table 1) as the only interpatch connectivity metric. To evaluate the alternative landscape resistance scenarios, we then built on the most parsimonious model to test 36 further models in which DEuclidean was replaced with resistance distances from each of the 36 landscape resistance scenarios. These 36 models were ranked by BIC and the sum of BIC model weights for each resistance value of pine plantation, farmland and road relative to Eucalyptus forest was calculated. We did not consider unoccupied patches as potential settlement choices in our multinomial models because key genetic variables could not be estimated for those sites.
We found that 55·4% of sampled males (77 of 139) had dispersed from their natal patch. We identified interpatch dispersal events up to a distance of six kilometres (Appendix S1), although the mean distance moved by all males was 1390 m (SD 2070 m) and by dispersing males was 2830 m (SD 2040 m).
Statistical analysis of apparent emigration
The most parsimonious models of the probability of a sampled male having emigrated from its natal patch featured effects of patch isolation and genetic relatedness (Table 2). Males were more likely to have emigrated if their minimum relatedness to the females in their natal patch was high (i.e. unrelated females were not present) (Table 3, Appendix S2). The models also accounted for the expected negative relationship between emigration probability and mean relatedness among the individuals in the patch. Of the patch isolation metrics, the best-supported models featured the distance-weighted ‘local population size’ metrics (Summed patch population sizes within a buffer distance of the focal patch, weighted by the inverse of the distance to each patch). These metrics represented demographic isolation of the focal patch, and the best-supported model featured the more distance-restricted metric where the buffer was set to 1392 m, the mean dispersal distance of all males (Table 3, Appendix S2). The fourth best model (Table 2) showed an effect of natal patch area, with males less likely to emigrate from large patches (coefficient = −0·016, SE = 0·006, Z = −2·47, P =0·013). However, patch area and the ‘demographic isolation’ metrics were strongly negatively correlated (−0·67). This was most likely because small patches were usually not occupied unless they were adjacent to a large patch population (and therefore had low demographic isolation scores). We did not find support for a role of sex ratio at the juvenile stage in apparent emigration decisions.
Table 2. Explanatory variables included in logistic regression models of emigration choices by male Antechinus agilis in fragmented habitat (intercept not listed). Models are sorted by Bayesian information criterion (BIC) scores, with the lowest BIC score indicating the most parsimonious model and all models with ΔBIC <10 [approx. 2*ln(n)] are shown
∑n/d1392 + rmean + rmin(F)
∑n/d2830 + rmean + rmin(F)
∑n1392 + rmean + rmin(F)
Area + rmean + rmin(F)
∑n/d1392 + n + rmean + rmin(F)
Area + ∑n/d1392 + rmean + rmin(F)
∑n/d1392 + nF + rmin(F) + rmean
∑n/d1392 + n + rmean + rmin(F)
∑n/d1392 + n + rmean + rmin(F)
Area + n + rmin(F) + rmean
∑n/d1392 + nF + rmean + rmin(F)
∑n/d1392 + nF + rmean + rmin(F)
∑n/d1392 + Area + rmean
∑n2830 + rmean + rmin(F)
dOcc + rmean + rmin(F)
Table 3. Coefficients and significance tests for the most parsimonious BIC model of the probability of a male having emigrated from its natal patch. The model presented is a generalized linear model with logit link function
The best models of settlement choices featured an effect of minimum relatedness to females in that patch, the number of females per patch, interpatch distance, patch area (larger patches were preferred), habitat quality (represented by forest type) and an effect of overall mean relatedness between the source and destination patches (Table 4). Males were most likely to select dispersal destinations where the minimum relatedness to females in that patch was low (Tables 4 and 5). Although female abundance (n(F)) was a significant predictor of male settlement choices, the second and third best models in Table 4 indicate that female relatedness (rmin(F)) was a better predictor of male settlement choices than female abundance.
Table 4. Bayesian information criterion-ranked multinomial regression models of dispersal destination choices by 139 male agile antechinus
Following the initial model fitting, we repeated the analysis by fitting the same multinomial model to a data set of dispersing males only, to check that the results were not solely driven by emigration choices. From these models, we found strongest support for models with the same set of variables.
When we replaced Euclidean distance in the best model with each of the landscape resistance metrics, the three best models featured a common resistance score of 10 (the maximum value) for pine plantation and farmland (Appendix S3). Further, the difference in BIC between the top model and the ‘null’ landscape resistance model (where all landscape elements had a resistance value of 1) was 16·1 (Appendix S3), which suggests that heterogeneity in resistance of the different landscape elements had an important effect on dispersal. Summing the BIC model weights for all models featuring each resistance value of pine plantation, farmland and roads revealed strong support for the highest relative resistance of pine plantation and cleared grazing land compared with eucalypt forest but no clear resolution of the relative resistance to dispersal across roads compared with eucalypt forest (Table 6).
Table 6. Summed BIC model weights for each resistance value (relative to eucalypt forest) of pine plantation, cleared grazing land and roads
Resistance value (relative to eucalypt forest)
Reproductive and environmental influences on dispersal decisions by males
We tested hypotheses about the effects of environmental and reproductive factors on apparent emigration and settlement decisions by male agile antechinus and found that two common factors, inbreeding risk and landscape connectivity, influenced these different components of dispersal behaviour. The decision to emigrate from the natal patch was best explained by a model featuring patch isolation and inbreeding risk, and the choice of settlement location was best explained by a model featuring inbreeding risk, interpatch distance (accounting for heterogeneous landscape permeability) and patch metrics such as patch size and habitat quality (represented by vegetation). We discuss below the major novel conclusions arising from our analyses of these data.
Inbreeding avoidance influences emigration decisions and dispersal destination choices
Our results suggest a surprisingly pervasive effect of inbreeding risk on dispersal, indicating that it affected not only the decision to emigrate, but also the choice of dispersal destination. Males were most likely to settle in patches where at least some females were highly unrelated. Inbreeding has well-documented negative fitness consequences (Frankham 1995; Amos et al. 2001), and inbreeding avoidance can be an important component of mate choice decisions (Tregenza & Wedell 2000). Where relatedness to local females is high, female preference for unrelated males as sires can influence the decision by males to emigrate to increase their chances of successful reproduction (Lehmann & Perrin 2003). Many studies have provided empirical links between inbreeding risk and emigration propensity. Indeed, the strong male-bias in dispersal in Antechinus has been attributed to inbreeding avoidance (Cockburn, Scott & Scotts 1985). However, inbreeding avoidance has typically been considered as a driver of emigration, not subsequent settlement decisions (Bowler & Benton 2005).
Although our finding of the relationship between inbreeding avoidance and settlement decisions is novel, it is consistent with research that shows mate availability to be a key influence on dispersal patterns (Fisher 2005; Gilroy & Lockwood 2012). We found the raw number of females in a patch to be a significant predictor of male settlement choices (although not as strong as female relatedness). However, female relatedness to males may mediate their availability as mates through the ‘glass effect’ (Tainaka & Itoh 1996), whereby females may be present but inaccessible as mates due to inbreeding avoidance. Males may avoid settling in such patches, as their future reproductive success may be low.
Two important requirements for the glass effect to occur are for females to prefer mating with unrelated males and to be able to distinguish among unfamiliar males on the basis of genetic relatedness. Both of these phenomena have been demonstrated in a number of species (Pusey & Wolf 1996; Mateo & Johnston 2000; Parrott, Ward & Temple-Smith 2006; Bonadonna & Sanz-Aguilar 2012). Indeed, female agile antechinuses are able to distinguish between related and unrelated males with olfactory cues (Parrott, Ward & Temple-Smith 2007) and preferentially select less-related males as sires (Parrott, Ward & Temple-Smith 2006). Thus, variation between potential destination patches in the relatedness of dispersing males to resident females may add an important consideration to mate choice decisions relating to potential future mating success.
We expect that inbreeding avoidance might exert an influence on post-dispersal settlement decisions when inbreeding confers a fitness risk (Charlesworth & Willis 2009), and the scale of positive genetic structure is greater than the scale of most dispersal events, such that leaving the natal patch does not necessarily equate to leaving the local genetic neighbourhood. The frequency of these conditions is unknown, and there are few studies that have directly measured individual dispersal distances and spatial patterns of relatedness between potential mates (Lowe & Allendorf 2010). In our study system, the genetic similarity of potential mates (opposite-sex pairs) in fragmented habitat did not decrease appreciably over the scale of most dispersal events (Fig. 2). Therefore, an inbreeding risk may still exist even when individuals move out of the natal patch. This may be a particular characteristic of populations in patchy habitats, because the individuals that disperse do so in a targeted manner between small patch populations (Gustafson & Gardner 1996; Kuefler et al. 2010), thus elevating the average relatedness between them. Essentially, the spatial scale of genetic neighbourhoods can increase due to changed dispersal behaviour in fragmented systems (Dick, Etchelecu & Austerlitz 2003), potentially imposing a novel constraint on dispersal patterns: inbreeding avoidance as a post-emigration influence on settlement location.
Landscape connectivity influences on dispersal
We found that the geographic isolation of patches and heterogeneity in the permeability of the intervening matrix affected dispersal patterns in fragmented habitat. In congruence with previous simulation studies (Tischendorf, Bender & Fahrig 2003), we found support for area-based patch isolation metrics, and an inverse distance-weighted area metric with a low cut-off distance (1392 m) was best supported by our data. In a simulation study, Jaquiery et al. (2011) also found evidence that associations between dispersal and environmental variation were strongest at a local (neighbouring patch) scale. In our study (Appendix S1) and others (Kuussaari, Nieminen & Hanski 1996), most dispersal events occur over short distances, suggesting that the measurement of environmental variables relevant to population connectivity should focus on a local spatial scale.
Patterns of animal movement across landscapes can be strongly influenced by habitat heterogeneity (Gustafson & Gardner 1996; Revilla et al. 2004). In fragmented landscapes, contrasts in habitat suitability and permeability between different landscape elements can be particularly high, influencing decisions to disperse or not, the subsequent dispersal pathways used and distances moved (Revilla et al. 2004; Long et al. 2005). A review of empirical studies suggests that structural similarity of the matrix to the patch habitat favours increased permeability (Prevedello & Vieira 2010). Our results contrasted with this in documenting strong support for increased resistance to dispersal through pine plantation (which is superficially structurally similar to eucalypt forest), slightly weaker support for increased resistance of cleared farmland (very different to eucalypt forest) and no resolution of the relative resistance to dispersal across roads (Table 6).
We suggest several reasons for the differences between our observed data and predictions of increased dispersal with increasing structural similarity of the matrix to patch habitat. First, although pine plantation is superficially structurally similar to eucalypt forest, it does not contain key resources to sustain populations of the agile antechinus (Lindenmayer, Cunningham & Pope 1999), and its unsuitability as habitat appears to correlate with a relative unsuitability for dispersal. Secondly, matrix permeability has an important influence on dispersal (Prevedello & Vieira 2010), but its effects can be mediated by matrix scale (Powney et al. 2011). The perceptual range of a species represents an important threshold over which responses to matrix type change considerably (Prevedello, Forero-Medina & Vieira 2010). Thus, although roads can be important dispersal barriers (Holderegger & Di Giulio 2010) many species may be insensitive to even a highly structurally modified matrix, such as roads, over the short distances required to cross them (typically 5–20 m in this landscape). Thirdly, sampling will affect the power to detect effects of matrix permeability on dispersal, and, compared with pine plantation, grazing land is poorly represented in the landscape that we sampled (Fig. 1), so it is likely that we had considerably less power to resolve its relative resistance than we did for pine plantation.
Insights from individual dispersal data
Data on individual dispersal events enable the inclusion of dispersal direction and individual variables into analyses and can thus be used to answer a broader range of questions than indirect dispersal estimation methods such as landscape genetics or population synchrony (Lowe & Allendorf 2010). Further, such data can be used to directly quantify migration probabilities and feed into demographic models to predict population viability in fragmented landscapes (Beissinger et al. 2006; Heinz, Wissel & Frank 2006). However, such data do have limitations that need to be understood. Indeed, samples of post-dispersal survivors will underestimate emigration rates, as dispersal mortality is likely to be high (Pita, Beja & Mira 2007). Individual movement data also lead to underestimation of dispersal distances due to the finite areas that can be sampled (Koenig, Van Vuren & Hooge 1996). In our study, we sampled individuals before and after dispersal. Thus, we did not know the precise demographic conditions at the time of dispersal and could not estimate or model the density of competing males in alternative destination patches at the time of dispersal (we only had estimates of post-dispersal male abundance).
Despite the limitations of individual dispersal data, they provide detailed insights into the non-randomness of dispersal patterns. Penteriani & Delgado (2011) proposed the term ‘birthplace-dependent dispersal’ to describe the propensity of individuals from a given natal location to disperse in a non-random manner with regard to direction and distance (i.e. to similar locations). Many studies, particularly in the field of landscape genetics, have identified geographic correlates of inferred population connectivity (Storfer et al. 2007). While environmental variation is a key influence on non-randomness of dispersal, individual-focussed studies have identified important physiological, genetic and demographic influences on dispersal that are not easily predicted from landscape patterns alone (Gundersen et al. 1999; Conradt, Roper & Thomas 2001; Revilla et al. 2004; Fisher 2005; Shafer et al. 2011; Teichroeb, Wikberg & Sicotte 2011). Thus, approaches that measure individual dispersal events contribute importantly to our understanding of dispersal by integrating behavioural, physiological, genetic and landscape processes.
A. Taylor, S. Ward, G. Finlayson and S. Lawson contributed to the project and W. Blanchard advised on statistics. Funding was provided by the Australian Research Council to A. Taylor and S. Ward. Fieldwork was conducted under Monash University ethics permit BSCI/2001/10 and NSW research permit B2293.
Input files for the modelling of apparent emigration and settlement decisions are provided in CSV format. Multilocus microsatellite genotypes are provided in GenAlEx/Microsoft Excel format. All data are deposited in the Dryad repository: doi:10.5061/dryad.jq878