The biotic and abiotic contexts of ecological selection mediate the dominance of distinct dispersal strategies in competitive metacommunities

Abstract

While the influence of dispersal on ecological selection is the subject of intense research, we still lack a thorough understanding of how ecological selection operates to favour distinct dispersal strategies in metacommunities. To address this issue, we developed a model framework in which species with distinct quantitative dispersal traits that govern the three stages of dispersal—departure, movement and settlement—compete under different ecological contexts. The model identified three primary dispersal strategies (referred to as nomadic, homebody and habitat-sorting) that consistently dominated metacommunities owing to the interplay of spatiotemporal environmental variation and different types of competitive interactions. We outlined the key characteristics of each strategy and formulated theoretical predictions regarding the abiotic and biotic conditions under which each strategy is more likely to prevail in metacommunities. By presenting our results as relationships between dispersal traits and well-known ecological gradients (e.g. seasonality), we were able to contrast our theoretical findings with previous empirical research. Our model demonstrates how landscape environmental characteristics and competitive interactions at the intra- and interspecific levels can interact to favour distinct multivariate and context-dependent dispersal strategies in metacommunities.

This article is part of the theme issue ‘Diversity-dependence of dispersal: interspecific interactions determine spatial dynamics’.

1. Introduction

The well-established concept of evolutionary selection, a major driver of changes in population genetics, finds a parallel under an ecological perspective [1]. Ecological selection emerges from the interactions of species’ traits with their biotic and abiotic environments, ultimately determining the subset of species in the regional pool that will persist or be excluded from local communities [2] and regional metacommunities. Much as gene flow modulates the effects of evolutionary selection on population genetics, dispersal—a multistage process in which individuals depart from their natal patches, traverse the landscape and settle in breeding patches (sensu [3])—modulates the strength of ecological selection. For instance, dispersal can buffer ecological selection by rescuing populations in unsuitable habitat conditions [4]. Similarly, emigration can reduce the intensity of ecological selection on community dynamics when it decreases the sizes of populations, making them prone to local extinctions owing to demographic stochasticity [5,6]. Despite extensive research about dispersal’s impact on ecological selection (reviewed by [7]), it is not widely acknowledged that dispersal itself is regulated by ecological selection operating at the metacommunity scale. Yet, this knowledge is critical for understanding metacommunity dynamics and the effects of global change on biodiversity (e.g. [8]).

Ecological selection affects dispersal when species’ persistence or exclusion in metacommunities depends on their dispersal strategies. This filtering process is known to vary according to the abiotic and biotic contexts where ecological selection occurs, generating non-random distributions of dispersal traits along ecological gradients (e.g. [9,10]). Nevertheless, we currently lack a comprehensive understanding of how ecological selection shapes dispersal. Achieving this understanding necessitates the integration of conceptual refinements in two interconnected yet distinct fields within ecology: movement ecology and metacommunity ecology.

Movement ecology teaches us that dispersal is a multistage process involving a series of decisions involving when to leave the natal patch (emigration), the distance travelled (displacement) and the selection of a new suitable location for settlement [3,11]. Each of these decisions is typically informed and context-dependent, meaning that organisms adjust them based on information about the surrounding biotic (e.g. predation, kin competition and interspecific competition) and local abiotic (e.g. resource availability and spatiotemporal environmental variation) environments [12,13]. The diverse potential combinations of these informed, discrete decisions and traits, along with their quantitative manifestations, determine a variety of multivariate dispersal strategies upon which ecological selection can operate.

Metacommunity ecology teaches us that the interplay between competitive dynamics and landscape characteristics can lead to both synergistic and opposing ecological constraints on the success of distinct dispersal strategies. For instance, in landscapes where environmental conditions change abruptly over time and show weak spatial autocorrelation, density-independent ecological selection tends to favour dispersal strategies that include high emigration rates, long displacement distances and effective tracking of suitable conditions [9,14–16]. Nonetheless, the success of any dispersal decision in response to spatiotemporal environmental variation is influenced by both heterospecific and conspecific competition occurring before, during and after dispersal. For instance, consider a scenario where competitive dynamics impede local coexistence between immigrants and heterospecific residents, such as when heterospecific competition is stronger than conspecific competition (destabilizing competition, sensu [2]). Then, species with a dispersal strategy that reduces the rates of heterospecific encounters in extant patches (i.e. low emigration rates and a preference for colonizing patches dominated by conspecifics) are likely to dominate the metacommunity. These examples illustrate how the complex interplay between abiotic and biotic contexts of ecological selection within metacommunities may influence the success of various dispersal strategies.

Investigating the effects of ecological selection on dispersal remains challenging for multiple reasons. Broad-scale observational data on multi-species dispersal strategies are limited, and experimental studies often face constraints in terms of the number of species and environmental factors that can be manipulated (but see [13,17]). Furthermore, these studies often involve a restricted range of experimental conditions, which hinders their capacity to encompass the diverse array of competitive dynamics and landscape characteristics that favour distinct dispersal strategies in metacommunities.

In this context, the present study is the first to investigate the influence of ecological selection on multivariate and context-dependent dispersal strategies across multiple abiotic and biotic contexts. To do so, we employed process-based metacommunity models to assess how ecological selection determines the permanence or exclusion of species with distinct combinations of traits associated with each dispersal stage (hereafter dispersal strategies; figure 1). We simulated metacommunity dynamics considering a broad range of ecological contexts given by the cross-factorial combination of landscape attributes and competitive dynamics that have been previously suggested to underpin ecological selection on dispersal [9,18,19]. Our modelling framework can be thought of as the in-silico replication of experimental studies that investigate the effects of competition on dispersal across broad-scale gradients on landscape features (see §2 and electronic supplementary material, figure S1). By assessing and contrasting the dispersal traits of species that persisted and dominated metacommunities across different simulation scenarios, we were able to derive theoretical predictions about how biotic and abiotic contexts of ecological selection interact to determine the dominant dispersal strategies in metacommunities.

Figure 1.

In this study, dispersal strategies are multivariate, i.e. they are determined by species-specific and random combinations of three independent traits (ep, hs and dp; see §2b). These dispersal traits determine the shape of functions regulating context-dependent emigration probability, habitat selection and displacement probability. Species-specific combinations of traits shown here are for illustrative purposes only.

The predictions derived from our framework were intentionally reported as ‘dispersal strategies–ecological gradients’ relationships’ to enhance the alignment of our theoretical findings with the types of trait-based results commonly reported in observational studies investigating the effects of ecological selection on community assembly (reviewed in [20]). While our models were not parametrized to reproduce the dynamics of any specific empirical metacommunity, they successfully mirrored well-known empirical relationships between dispersal strategies and ecological gradients observed in studies carried out at the local and biogeographic scales [9,15,21]. This congruence not only validates our model as a suitable theoretical framework for examining the effects of ecological selection on dispersal but also facilitates the generation of novel predictions that can be empirically tested in future studies.

2. Methods

(a). Simulated landscapes

We provide only a brief description of our landscape simulation here. A more detailed explanation and R code can be found in the electronic supplementary material. We created 25 landscape types using a cross-factorial design that combines five levels of temporal periodic oscillations in habitat conditions with five levels of spatial autocorrelation in environmental conditions. The periodic oscillations in habitat conditions considered here were meant to reproduce in-silico the effects of seasonality on the abiotic conditions experienced by communities. Seasonality is widely recognized as a significant ecological predictor of species dispersal strategies at biogeographic scales [9,22]. Spatial autocorrelations in environmental conditions determine the size of habitat clusters and, therefore, the costs and risks of dispersal events [19]. By considering combinations of these two landscape characteristics, we could assess the most successful dispersal strategies under a broad spectrum of abiotic conditions, ranging from those with low to high temporal variability and spatial unpredictability in habitat conditions. Each landscape, characterized by unique combinations of seasonality and spatial autocorrelation, consisted of 50 habitat patches with x and y spatial coordinates randomly taken from a uniform distribution within a range of 0–60.

(b). Parametrizing metacommunity dynamics

Metacommunity dynamics were dictated by a Beverton–Holt model with generalized Lotka–Volterra competition:

$${\displaystyle N_{i,j,t+1}=Poisson\left(N_{i,j,t}\times P_{i,j,t}\right)-E_{i,j,t}+I_{i,j,t}}$$ (2.1)

where N _i,j,t is the abundance of species i in site j at time t, $P_{i,j,t}$ , is the local performance of species i (i.e. growth rate) when conditioned by competition and habitat selection as:

$${\displaystyle {\displaystyle P_{i,j,t}=R.max\times \mathrm{e}\mathrm{x}\mathrm{p}\left(\frac{-{\left(En{v}_{j,t}-{\mu }_i\right)}^2}{2{\sigma }^2}\right)\times \frac{1}{\left(1+{\alpha }_{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{a}}{N}_{i,j,t}+{\alpha }_{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}}\sum _{k\ne i}^S{N}_{k,j,t}\right)}}}$$ (2.2)

where $R.max$ (fixed at 3 for all species) represents the species’ maximum intrinsic growth rate when habitat conditions are optimum and competition at the intra- and interspecific levels are negligible. As such, $P_{i,j,t}$ ranges from 0 to $R.max$ .

The second term of equation (2.2) sets density-independent species-environment sorting and is given by the match between local (within patches) environmental conditions ${(Env}_{jt})$ and species’ environmental optima ( ${\mu }_i),$ considering their environmental tolerance $\left(\sigma \right)$ . This term (hereafter EnvSuit) ranges between 0 (i.e. the most extreme mismatch between ${Env}_{jt}$ and ${\mu }_i$ ) and 1 (i.e. perfect match between ${Env}_{jt}$ and ${\mu }_i$ ). Environmental tolerance $\left(\sigma \right),$ set at 1 for all species, ensured sensitivity to environmental variation while minimizing local extinction risks under suboptimal conditions; this value was confirmed by preliminary simulations.

The third term of equation 2.2 generates density-dependent competition on population dynamics, incorporating per capita intraspecific ${\displaystyle \left({\alpha }_{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{a}}\right)}$ and interspecific effects ${\displaystyle \left({\alpha }_{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}}\right);}$ this approach models three competition forms influencing local and regional coexistence (sensu [2] and see diagnostic plots in electronic supplementary material, figures S2 and S3). Stabilizing competition ( ${\displaystyle {\alpha }_{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{a}}>{\alpha }_{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}}}$ ) promotes local stable coexistence by allowing rare species populations to grow when populations of dominant competitors are in equilibrium. Equalizing competition ( ${\displaystyle {\alpha }_{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{a}}}$ = ${\displaystyle {\alpha }_{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}}}$ ) implies neutrality among individuals, requiring distinct niches or dispersal strategies for coexistence. Localized destabilizing competition ( ${\displaystyle {\alpha }_{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{r}\mathrm{a}}<{\alpha }_{\mathrm{i}\mathrm{n}\mathrm{t}\mathrm{e}\mathrm{r}}}$ ) reduces the chances of local coexistence, as locally dominant species can drive better-adapted species to extinction.

To incorporate the influence of demographic stochasticity on local birth and survival, we draw the final local species abundances from a Poisson distribution whose mean was determined by the estimated population size after density-independent species-environment sorting and density-dependent competition [2].

Individuals persisting in a local community post-selection and demographic stochasticity at time t could disperse. Context dependence in dispersal strategies was introduced by making species’ emigration probability $\left(EP\right)$ , displacement probability $\left(DP\right)$ and habitat selection probability ( ${\displaystyle HS}$ ) vary as a function of local performance ( $P_{i,j,t}$ ), geographic distance and environmental suitability ( ${\displaystyle EnvSuit}$ ), respectively. The shape of these relationships was made species-specific by randomly assigning different values of parameters ep, dp and hs to species in the regional pool at the beginning of each simulation iteration. Each parameter represents species-specific dispersal traits that are proxies of species dispersal abilities (e.g. hand-wing index and eye sizes in birds are a proxy of displacement capacity and habitat selection, respectively [21]). Assigning random combinations of values for ep, dp and hs to each species implies the assumption of no integration (genetic and/or phenotypic correlations) between dispersal traits at the species level, thereby creating a wide array of dispersal strategies upon which ecological selection could operate.

Emigration probability ( $EP$ ) determines the likelihood of an individual leaving its natal patch based on its current local performance. While the relationship between $EP$ and local performance can take various shapes in theoretical and empirical studies [23,24], the common underlying assumption is that individuals are more likely to emigrate as local performance declines. As such, the probability that species i will emigrate from patch j at time t was set to decrease with local performance $P_{i,j,t}$ (given by equation 2.3):

$${\displaystyle E{P}_{i,j,t}={\left(1-\frac{P_{i,j,t}}{R_{max}}\right)}^{\mathrm{e}\mathrm{x}\mathrm{p}\left(e{p}_i\right)}}$$ (2.3)

where ${ep}_i$ is species-specific (equally spaced values within the interval ranging from −2 to 2 [−2,0) ∪ (0,2] that were then randomly assigned to each species) and determines the rate of change in ${EP}_{i,j,t}$ as a function of $P_{i,j,t}$ . When ${ep}_i$ > 0, emigration probability steeply decreases even with incipient increases in local performance (i.e. as observed in species 2, 4 and 6 in figure 1). In contrast, species for which ${ep}_i$ < 0 were prone to emigrate from patches even if those provided high local performance (as observed in species 1, 3 and 5 in figure 1). ${EP}_{i,j,t}$ sets the probability of success in binomial trials determining the number of emigrants of species i departing from site j at time t ( $E_{i,j,t}$ ).

Random sampling process with replacement and unequal probabilities determined the total number of emigrants of species i that will immigrate to patch j at time t ( $I_{i,j,t}$ ) (see electronic supplementary material for a detailed description of our random sampling process). Immigration probabilities (IP) represent a compromise between the suitability of extant patches and their geographic distance from the natal patch. As such, the probability that species i will immigrate to patch j at time t ( ${IP}_{i,j,t}$ ) was estimated as ${\displaystyle H{S}_{i,j,t}\times D{P}_{i,kj}}$ .

${HS}_{i,j,t}$ represents the probability of species i choosing to immigrate to patch j at time t based on its environmental suitability. As such, we assumed habitat-matching at immigration (sensu [25]), i.e. species are more likely to immigrate to patches with suitable environments during dispersal. In nature, habitat-matching is common and observed when individuals integrate environmental cues at the local and regional scales to guide their choices during dispersal [26]. To incorporate habitat-matching into immigration probabilities, we assume ${HS}_{i,j,t}$ to increase with patch suitability as:

$${\displaystyle H{S}_{i,j,t}={\left(EnvSui{t}_{i,j,t}\right)}^{\mathrm{e}\mathrm{x}\mathrm{p}\left(h{s}_i\right)}}$$ (2.4)

where ${hs}_i$ is species-specific (equally spaced values within the interval [–2,0) ∪ (0,2] that were then randomly assigned to each species) and determined the rate of change in ${HS}_{i,j,t}$ with $EnvSuit$ . Species with ${hs}_i$ < 0 are less selective and can choose to immigrate to patches with suboptimal habitat conditions at time t (e.g. species 1, 2 and 3 in figure 1). Conversely, species having ${hs}_i$ > 0 are highly selective and tend to choose to immigrate only to patches with optimal or close to optimal environmental conditions (e.g. species 4, 5 and 6 in figure 1).

$DP$ _{i ,kj} is the probability that species i will displace from patch k to j according to the geographic distance between them. $DP$ _{i ,kj} decayed with the (Euclidean) distance between k and j ( ${dist}_{kj})$ ) as follows:

$${\displaystyle D{P}_{i,kj}=\mathrm{e}\mathrm{x}\mathrm{p}\left(d{p}_i\times dis{t}_{jk}\right)}$$ (2.5)

where ${dp}_i$ is species-specific (equally spaced values within the interval [−0.99, −0.01] that were then randomly assigned to each species) and determines the rate at which ${DP}_{i,kj}$ decayed with ${dist}_{kj}$ . The smaller the species’ ${dp}_i$ , the shorter its displacement capacity, and the more limited the number of nearby patches a species could reach in a single dispersal event.

(c). Simulation iterations

Metacommunity dynamics were set to run for 1200 time-steps (100 seasonal cycles of 12 time-steps; see electronic supplementary material, figure S1). Each scenario was replicated 30 times, totalizing 2250 simulations (30 replicates × 5 seasonality levels × 5 spatial structure levels × 3 competitive structures). In the first time-step (t ₁), we seeded each patch with abundances of 300 species randomly drawn from a Poisson distribution with λ = 0.5. Then, during the first 10 seasonal cycles of each iteration (i.e. the first 120 time-steps), we kept each patch seeded with species abundances randomly drawn from a Poisson distribution with λ = 0.1. This seeding procedure ensured equal chances for all species to be initially present in all patches and allowed patches with similar environmental conditions to host diverse communities over time, owing to priority effects [2]. Metacommunity dynamics, without seeding, proceeded for 1080 time-steps (90 seasonal cycles), with analyses focusing on the 100th seasonal cycle (final 12 steps) to ensure stable equilibria (when regional extinctions near zero, as shown in diagnostic electronic supplementary material, figures S2 and S3).

To assess the effects of ecological selection, the most successful dispersal strategy at the metacommunity level was estimated as the combination of metacommunity-weighted mean (MWM) values of ep, hs and dp, where the weights were given by a species’ regional abundance × occupancy (i.e. the relative number of patches occupied in the landscapesee more in electronic supplementary material, figure SI). Note that estimates of MWM ep, hs and dp correlate with other common metrics typically used to categorize dispersal strategies in empirical and theoretical studies (e.g. average dispersal distance and proportion of emigrants in the metacommunity; see electronic supplementary material, figure S4).

(d). Understanding how landscape features and competition dynamics select for dispersal strategies in metacommunities

We used unsupervised machine learning (k-means) to group metacommunities based on similarities in dominant dispersal traits (assessed through estimates of MWM ep, hs and dp). The optimal number of metacommunity clusters was identified using the general consensus of multiple algorithms in the NbClust package [27]. Concurrently, a principal component analysis ordinated metacommunities in a multivariate trait space, revealing associations of MWM traits in each cluster. These analyses together identified groups of dispersal strategies promoting species dominance under varying abiotic and biotic contexts.

Random forest (RF) models characterized how landscape attributes and competition types interact to determine: (i) the dominant dispersal strategies observed in metacommunities (categorical response determined by the approach described above) and (ii) changes in the MWM ep, hs and dp values individually (continuous response). The first RF model enabled the generation of broad predictions about the abiotic and biotic conditions influencing metacommunity dominance by the broad dispersal strategies identified via k-means. The second RF model provided deeper insights into how each trait’s dominance varies with landscape attributes and competition types. In both models, we used the same predictors: seasonality (ordinal, five levels), spatial autocorrelation (ordinal, five levels), competition types (categorical, three levels) and remaining regional richness at the end of each simulation iteration as a covariate. RF models are suited to identify general patterns in cross-factorial data because they automatically model the effects of multilevel interactions among predictors on the response.

Partial dependence plots revealed the direction of predictor–response relationships while controlling for other model predictors. These plots characterize predictions about how dominant dispersal strategies in our simulated metacommunities change along gradients of abiotic and biotic conditions, and their interactions. We employed bootstrapping to estimate 95% confidence intervals for the predicted responses shown in the partial dependence plots.

RF models were performed using the package randomForestSRC [28]. All simulations and statistical analyses were conducted using R (v. 4.2.0) [29].

3. Results and discussion

While we observed that landscape characteristics and competitive dynamics interact in intricate ways to influence the success of distinct dispersal strategies (figures 2–4), three primary dispersal strategies consistently emerged as dominant under different abiotic and biotic conditions (figure 3a,b ). We called the first dominant strategy ‘nomadic’ because of its inherent high emigration rates even at high-performing patches (i.e. lower MWM ep), high habitat selectivity (i.e. higher MWM hs) and long-distance displacement (i.e. higher MWM dp). The nomadic strategy was more likely to dominate metacommunities under stabilizing competition and in landscapes where environmental conditions were temporally variable (seasonal) and weakly spatially autocorrelated (figure 3c ). The second strategy was coined ‘homebody’, describing a dispersal strategy that minimizes exposure to heterospecific interactions in extant patches. More specifically, ‘homebody’ species displayed a lower emigration rate in high-performing patches (high MWM ep) and a preference to immigrate to safer, intraspecific clusters near their natal patch (lower MWM dp), independent of the suitability of extant patches (low MWM hs). The homebody strategy tended to dominate in metacommunities where dispersal risks were increased owing to the low probability of settlement in patches dominated by heterospecific competitors (i.e. destabilizing competition). Finally, we named the third strategy ‘habitat sorting’ because it tended to dominate under conditions where a species’ persistence in the metacommunity largely hinged on its capacity to locate and settle in optimal habitat patches. This was particularly the case when metacommunities were under equalizing competition and in landscapes with temporally stable (aseasonal) environmental conditions and weak spatial autocorrelation (figure 3c ). This strategy is characterized by low emigration rates at high local performance (higher MWM ep), high environmental selectivity (higher MWM hs) and long-distance displacement (higher MWM dp).

Figure 2.

Metacommunity-dominant dispersal traits observed across simulation scenarios. Species-specific dispersal traits (ep, hs and dp) determine the shape of the functions defining context-dependent changes in emigration probability (a), habitat selection (c) and displacement capacity (e), respectively. The functions shown here represent a small subset of the diverse range of dispersal traits seeded into metacommunities at each simulation iteration. The colour scales indicate the range of dispersal traits that dominated metacommunities at the end of each simulation (estimated as the metacommunity-weighted mean [MWM] values for ep, hs and dp). For illustrative purposes, we coloured the curves corresponding to some of these metacommunity dominant traits. The colour scales also serve as a reference for the heatmaps (b, d and f), illustrating the changes in MWM dispersal traits across levels of seasonality, spatial autocorrelation and types of competition. Each entry (square) in the heatmap represents the average MWM traits obtained across the 30 replicates for a given simulation scenario.

Figure 3.

(a) Biplot showing the results of k-means that grouped metacommunities (points) based on similarities in their dominant multivariate dispersal strategies (metacommunity-weighted mean [MWM] values for traits ep, hs and dp). Three broad dispersal strategies (colours) were identified, and their main differences are summarized in (b). See extended discussion about each strategy in the main text (§3a, b, and c). (c) Partial dependence plots showing the predicted probability of each strategy to dominate a metacommunity under different levels of seasonality, spatial autocorrelation in habitat conditions and competition forms. Out-of-bag misclassification rates for the random forest model reported in C was 34%.

Figure 4.

Partial dependence plots showing the predicted dominant values for the dispersal traits (i.e. estimated as metacommunity-weighted mean values for ep, hs and dp) across levels of seasonality (colours), spatial autocorrelation in habitat conditions and different types of competitive dynamics. R² of random forest models fitted for each trait: ep = 0.90, hs = 0.81 and dp = 0.95.

(a). When the nomadic dispersal strategy dominates

The primary dispersal characteristic of the nomadic strategy is its high emigration rates even from high-performing patches (relatively lower MWM ep). Although it seems at first glance to be a maladaptive emigration response, our models indicate that ecological selection favours species with such a dispersal strategy when two conditions are met (figure 4): (i) environmental conditions are temporally variable (here, seasonal) and (ii) competitive dynamics at the intraspecific level are stronger than at the interspecific level (i.e. stabilizing competition).

Nomadic species adjust their spatial distribution to respond to fast changes in local performance caused by temporal variability in habitat conditions [9,16,22]. These species are particularly successful in dominating the metacommunity by colonizing multiple patches at each dispersal event (i.e. high MWM dp; figure 4) and were highly selective to the environment of extant patches (high MWM hs). Empirical studies investigating latitudinal clines in species dispersal have observed a similar influence of seasonality, which selects for higher emigration rates and displacement capacity. For instance, the relationship between seasonality and dispersal observed in our model serves as theoretical evidence that the conditions for the emergence of well-known latitudinal patterns in species range sizes and dispersal capacity (e.g. [9,30]) can emerge by solely considering metacommunity dynamics at fine spatiotemporal scales (i.e. no need to consider trait evolution and speciation, but see §4).

Nomadic species exhibiting high emigration rates and long dispersal distances also tended to dominate metacommunities when intraspecific competition had a more detrimental impact on local performance than interspecific competition (i.e. stabilizing competition, figures 3 and 4). This strategy proves effective because it alleviates the negative effects of intraspecific competition in natal patches by promoting high emigration rates, even when local performance is high. Additionally, it decreases the likelihood of encounters with conspecifics by enabling species to travel farther away from intraspecific clusters in neighbouring patches. These trends are consistent with empirical and theoretical studies exploring how population density—an indicator of intraspecific competition—influences dispersal. Typically, such studies illustrate that high population densities lead to increased emigration rates, especially toward suitable patches located farther from their natal patch (see [31] and references therein).

(b). When the homebody dispersal strategy dominates

The homebody strategy essentially results in species emigrating less (i.e. higher MWM ep), keeping them close to the natal and neighbouring patches dominated by conspecifics (i.e. higher MWM dp) while avoiding patches dominated by heterospecifics with similar habitat requirements (i.e. lower MWM hs). This strategy is particularly advantageous when local coexistence among heterospecific competitors is unlikely (i.e. destabilizing competition) because homebody species can evade the negative effects of competition by reducing their contact with heterospecific competitors before and after dispersal.

We also observed that the homebody strategy was favoured in aseasonal landscapes. If the chances of settlement in extant patches are decreased owing to destabilizing competitive dynamics, the optimal response is to avoid constant emigration from temporally stable and high-performing patches [32]. By doing so, species can ensure local dominance in suitable patches by maintaining large local populations, making them less susceptible to the adverse effects of demographic stochasticity and reducing the likelihood of being outcompeted by other species. Note that strong spatial autocorrelation in habitat conditions was an essential condition for the success of homebody species. The benefits of strong habitat selectivity and high displacement capacity diminish in landscapes where habitats are highly correlated because there is a lower risk of dispersing to unsuitable neighbouring patches from suitable natal patches. Furthermore, strong autocorrelation is essential for the formation of ‘safe’ clusters of neighbouring patches dominated by conspecifics, thereby decreasing the frequency of heterospecific encounters during dispersal. Our results are consistent with previous theoretical models showing that strong spatial autocorrelation favours species with reduced displacement capacity, while weak autocorrelation favours species with strong displacement capacity [14]. Additionally, our findings support empirical evidence showing that species tend to adopt passive dispersal, a strategy characterized by weak habitat selection, when landscape structure is characterized by large clusters of patches with suitable habitat conditions [33].

(c). When the habitat-sorting dispersal strategy dominates

The habitat-sorting dispersal strategy allows species to effectively track the limited number of patches where adequate niche-environment matching outweighs the negative effects of competition. Species exhibiting such strategy can achieve regional coexistence by partitioning habitats based on their preferences and tolerances, resembling the expectations within species-sorting metacommunities (sensu [34]). This type of regional coexistence, linked to spatial and temporal environmental heterogeneity, is especially relevant under neutral interactions within and between species (i.e. equalizing competition).

Even though seasonal landscapes favoured species with high selectivity to patch environmental conditions, we observed that the nomadic strategy is more likely to dominate seasonal landscapes than the habitat-sorting strategy (figures 3c and 4). Differences in emigration responses may explain why the nomadic strategy dominates metacommunities in seasonal landscapes, while the habitat-sorting strategy is superior in aseasonal landscapes, despite both strategies being characterized by high habitat selectivity (high MWM hs). The nomadic strategy, characterized by high emigration rates (low MWM ep) and high habitat selectivity, ensured the rescue of populations in extant patches following abrupt changes in habitat conditions caused by seasonality. Conversely, the low emigration rates associated with the habitat-sorting strategy allowed the maintenance of large populations in suitable and temporarily stable (aseasonal) habitat conditions, ensuring the local dominance of the dispersal habitat-sorting strategy. This dominance is further enhanced by the high displacement capacity inherent to this strategy, which ensures that species can reach distant suitable patches before competitors with similar habitat requirements.

4. Conclusions, assumptions and future directions

Here, we used models to demonstrate that different dispersal strategies not only shape the structure of metacommunities but also emerge from metacommunity dynamics. Previous theoretical and empirical studies sharing some of our goals focused on investigating the ecological drivers of fixed behaviours associated with one or two stages of dispersal (e.g. [19]). Our study is the first to use metacommunity theory to generate predictions regarding the selective effects of landscape features and competition dynamics on species-specific and context-dependent dispersal behaviours involved in all three dispersal stages (i.e. departure, transience and settlement). In doing so, our models successfully recreated well-known variations in dispersal strategies across spatial scales, encompassing changes induced by different forms of intraspecific and interspecific competition at local spatial scales, as well as shifts in dispersal patterns along broad-scale ecological gradients. Therefore, our study enhances the understanding of the factors influencing the success and diversity of dispersal strategies in a wide range of ecological contexts.

Note that our simulation models did not include the full complexity of species dispersal and its intricate relationships with metacommunity dynamics. For example, we did not consider cost-related trade-offs that can cause covariance between dispersal, morphological and behavioural traits [35]. By doing so, we could understand whether colonization-competition and ecological specialization–dispersal trade-offs can emerge as eco-evolutionary consequences of community assembly in landscapes with varying levels of environmental stability and habitat heterogeneity.

Despite its intermediate complexity, our models provide deeper insights into metacommunity dynamics, demonstrating that the relative strength of interspecific compared to intraspecific competition can heavily dictate the success of distinct dispersal strategies at the metacommunity level. However, we acknowledge that other biotic interactions can also drive the selection of optimal dispersal strategies. For instance, predation risk can drive emigration [36], while parasitism can have dual effects on host dispersal: it can promote movement if the host perceives a threat and relocates, or it can inhibit movement if the host remains and becomes infected [37]. Incorporating these and other biotic interactions into our framework could be a logical progression for future research.

Finally, in our model, while we explored how biotic and abiotic factors were selected for a wide range of predefined traits (ep, hs and dp) that govern species-specific dispersal strategies, we did not account for trait evolution. Trait evolution is a significant component of metacommunity theory [38] and the literature contains examples where the role of dispersal in maintaining biodiversity in changing environments is offset by the evolution of traits determining species’ habitat requirements [39]. Moreover, we did not consider the entire spectrum of potential nonlinear changes in dispersal strategies, which are commonly observed in nature. For instance, we assumed that emigration propensity decreased monotonically and continuously with local performance. However, Allee effects or other density-dependent behaviours could result in various nonlinear relationships between population density and, consequently, local performance and dispersal (e.g. u-shaped [18] or threshold functions [24]). Future studies could expand upon our modelling framework to explore how traits underlying a wider array of dispersal strategies evolve under selection within even more diverse ecological and evolutionary context.

Ethics

This work did not require ethical approval from a human subject or animal welfare committee.

Data accessibility

The code and data required to run simulations and analyses reported in this study are available as electronic supplementary material [40].

Declaration of AI use

We have not used AI-assisted technologies in creating this article.

Authors’ contributions

G.K.: conceptualization, formal analysis, investigation, methodology, writing—original draft, writing—review and editing; P.S.: conceptualization, formal analysis, investigation, methodology, writing—original draft, writing—review and editing; P.R.P.-N.: conceptualization, investigation, methodology, supervision, writing—original draft, writing—review and editing.

All authors gave final approval for publication and agreed to be held accountable for the work performed therein.

Conflict of interest declaration

We declare we have no competing interests.

Funding

This research was funded by the Canada Research Chair in Spatial Ecology and Biodiversity (chair holder P.R.P.-N.) and P.S. was assisted by the Concordia Horizon Postdoctoral program.