Metacommunity structure in a small boreal stream network


Corresponding author. E-mail:


  1. Current ecological frameworks emphasize the relative importance of local and regional drivers for structuring species communities. However, most research has been carried out in systems with discrete habitat boundaries and a clear insular structure. Stream networks deviate from the insular structure and can serve as excellent model systems for studying hierarchical community dynamics over different temporal and spatial extents.
  2. We used benthic invertebrate data from streams in a small northern Swedish catchment to test whether metacommunity dynamics change between seasons, across spatial hierarchies (i.e. at the whole catchment scale vs. the scales of first-order and second/third-order sites within the catchment) and between stream-order groups.
  3. We assessed metacommunity structure as a function of three relevant dispersal dimensions (directional downstream processes, along-stream dispersal and overland dispersal). These dispersal dimensions were related to species groups with relevant dispersal traits (flying capacity, drift propensity) and dispersal capacities (weak vs. strong) to elucidate whether the observed spatial signals were due to dispersal limitation or mass effects.
  4. Results showed complex community organization that varied between seasons, with the scale of observation, and with stream order. The importance of spatial factors and specific dispersal dimensions was highly dependent on the time of sampling and the scale of observation. The importance of environmental factors was more consistent in our analyses, but their effect on species community structure peaked at first-order sites. Our analyses of species dispersal traits were not unequivocal, but indicated that both mass effects and dispersal limitation could simultaneously contribute to the spatial signal at the scale of the whole catchment through different dispersal pathways.
  5. We conclude that the study of hierarchically organized ecosystems uncovers complex patterns of metacommunity organization that may deviate substantially from those of systems with insular structure and discrete habitat boundaries. Moreover, we show that dispersal constraints imposed by the dendritic structure of stream networks and distinct dispersal mechanisms (e.g. dispersal limitation) may be evident also at very small spatial extents. Thus, even at this small scale, a landscape management approach that takes the dendritic nature of stream networks into account is needed to effectively conserve stream biodiversity.


Ecologists have traditionally studied biological diversity in both local and regional contexts. However, these perspectives are not mutually exclusive, and contemporary ecology reconciles these opposing views in theoretical meta-ecological frameworks that span various levels of biological hierarchy, from metapopulations (Hanski 1999) to metacommunities (Leibold et al. 2004; Holyoak et al. 2005) to metaecosystems (Loreau, Mouquet & Holt 2003; Massol et al. 2011). These frameworks consider local entities (abiotic and biotic) to be spatially connected at the landscape scale, with dispersal perhaps serving as an important linking agent (Holyoak et al. 2005).

Metacommunity ecology theory centres around four main paradigms that weight local environmental factors and dispersal differently in structuring biotic communities (Leibold et al. 2004; Holyoak et al. 2005). The patch-dynamic model (PD) assumes that patches are identical (no environmental filtering) and that a competition–colonization trade-off, where poor competitors are good dispersers and vice versa, drives community dynamics. Similar to the PD model, the neutral model (NE) also assumes that patches are identical, but, in contrast to PD, it assumes that all species are ecologically identical (no competition–colonization trade-off) and that stochastic processes, including extinctions, immigrations and evolution, regulate species composition. The species sorting (SS) and mass-effect (ME) paradigms both assume that there is an environmental gradient and that species are ecologically different. The difference between the two is the relative strength or importance of dispersal. The SS model assumes that dispersal is moderate, allowing species to sort along environmental gradients and is therefore highly related to niche theory (Hutchinson 1959). ME, on the other hand, becomes apparent when dispersal (immigration/emigration) is strong enough to alter local species composition (equivalent to SS + high dispersal in Ng, Carr & Cottenie 2009), allowing the survival of maladapted or extinction-prone populations via source-sink dynamics (Mouquet & Loreau 2002, 2003). A fifth situation (dispersal limitation) may arise if all species cannot reach all sites within a region, thereby preventing species from being sorted to their preferred environment (equivalent to SS + low dispersal in Ng, Carr & Cottenie 2009).

Two reviews on metacommunity structure in terrestrial and aquatic environments revealed that SS and ME are dominant in natural systems (Cottenie 2005; Logue et al. 2011). However, Logue et al. (2011) found that empirical metacommunity ecology is biased towards systems with discrete habitat boundaries or a clear insular metacommunity structure (e.g. lakes, ponds, islands and moss patches). Thus, current results mainly reflect the dominant type of metacommunity paradigms tested, rather than the dominant type of metacommunities that exist in nature. In contrast to the study systems given most attention in metacommunity research (see examples above), many ecosystems are hierarchically structured, comprise habitats with unclear boundaries that can, but must not, be highly variable in space and time (e.g. streams, floodplains, grasslands, forests, coral reefs). Further research is therefore needed on ecosystem types that deviate from insular metacommunity structure to assess community assembly under the constraints set by local and regional factors.

Research has begun to address metacommunity structure in response to disturbances (Urban 2004; Bloch, Higgins & Willig 2007), varying connectivity patterns and regimes (Cadotte 2006; Angeler et al. 2010), hierarchical landscape extent (Ng, Carr & Cottenie 2009; Declerck et al. 2011) and temporal variability (Ellis, Lounibos & Holyoak 2006; Vanschoenwinkel et al. 2010). Studying the collective influences of these factors in a single study system holds potential to generate ecologically realistic patterns and can provide insight into the organization of biodiversity as a function of natural spatial and environmental constraints. These patterns can then be used to refine biodiversity management and conservation.

Stream networks can serve as excellent model systems for studying hierarchical community dynamics over different temporal and spatial extents. They are highly dynamic, showing predictable seasonal variation in hydraulic conditions (stream flow, discharge), water chemistry (Laudon & Bishop 1999; Laudon, Köhler & Buffam 2004), temperature and availability of food resources (Giller & Malmqvist 1998). They are highly heterogeneous systems both within and across the network, due to differences in in-stream environmental and riparian conditions between upstream and downstream sites (e.g. Vannote et al. 1980; Richardson & Danehy 2007) and due to the large environmental (non-longitudinal) heterogeneity across stream networks, known to be especially pronounced in headwater catchments (Temnerud & Bishop 2005; Buffam et al. 2007; Clarke et al. 2008). Also, the dendritic structure of stream networks allows for analysing community assembly in linear and hierarchically branched system units, along habitat gradients, wherein habitat patches have no clearly defined boundaries, and at different dispersal dimensions (Grant, Lowe & Fagan 2007). That is, dispersal can occur both along the watercourses in stream networks (‘within/along-stream dispersal’) and across watersheds (‘overland dispersal’). However, the dispersal dimensions can be species specific and depend on whether or not life cycles include a winged adult stage (Grant, Lowe & Fagan 2007; Chaput-Bardy et al. 2009; Landeiro et al. 2011). Thus, there is a need to consider dispersal dimensions and space explicitly in analyses of stream metacommunity dynamics.

Benthic invertebrates are key organisms in stream ecosystems. They comprise a diverse species group with different life cycles, life histories and functional and trophic traits. They also differ in their dispersal ability (Poff et al. 2006) and include organisms that can disperse both overland (adult flying insects) (e.g. Wilcock, Nichols & Hildrew 2003; Macneale, Peckarsky & Likens 2005), along the watercourses (adult flying insects) (e.g. Petersen et al. 2004; Macneale, Peckarsky & Likens 2005) and within the watercourses (drifting insect larvae; Giller & Malmqvist 1998). In addition, the organisms with a winged adult life stage usually have a phenologically well-defined (seasonal) period of emergence (e.g. Petersen et al. 1999) triggered by extrinsic physical factors (e.g. Cowell, Remley & Lynch 2004; Bogan & Lytle 2007). Invertebrates are therefore ideal organisms for assessing both seasonal signals and the relevance of different dispersal dimensions (across, along and within stream networks) in community assembly. Researchers have started to assess the importance of different dispersal dimensions (Cadotte 2006; Brown & Swan 2010; Landeiro et al. 2011; Carrara et al. 2012). However, despite the known importance of downstream dispersal (i.e. drift) within stream networks, directional spatial processes have never been explicitly assessed in previous metacommunity studies. This is probably because of the trade-off between the effort involved in collecting large-scale data sets and the possibility to include several sampling points along each stream segment. In any case, it has most likely led to an underestimation of the effect of spatial factors in stream metacommunities.

In this study, we analysed benthic invertebrate metacommunity structure in a very small northern Swedish boreal headwater catchment (total investigated area c. 20 km2), comprised of first- to third-order streams, all with sub-catchment areas <10 km2, and where the Euclidian distance between the two farthest sampling points measures c. 4·7 km. Previous studies, covering much larger spatial extents (150–2173 km2), have found that metacommunities in small (first-third order) streams are not spatially structured, but rather the result of environmental control (Heino & Mykrä 2008; Heino et al. 2012). At the same time, the dendritic structure of stream networks causes different metacommunity dynamics in upstream vs. downstream sites (Brown & Swan 2010). However, the generality of these findings has so far not been exhaustively assessed in empirical studies.

To our knowledge, this is the first study to investigate changes in invertebrate metacommunity structure by sampling the communities during two periods (seasonal snapshots) during which community assembly processes differ (e.g. Petersen et al. 1999; Cowell, Remley & Lynch 2004; Bogan & Lytle 2007) and with a focus on a very small spatial extent. The small spatial extent was selected for several reasons. First, it allowed us to investigate whether dispersal constraints caused by the dendritic structure (Brown & Swan 2010) are strong enough to be evident also at small spatial scales. Second, we were able to assess whether spatial signals are evident in very small catchments and whether the underlying dispersal mechanism is different from what is expected at larger spatial extents (i.e. ME vs. dispersal limitation, respectively). Finally, the small spatial extent allowed us to make detailed measurements across a stream network and thereby assess the importance of independent spatial processes related to dispersal through flight (across and along the stream network) and through drift (associated with water flow).

We specifically test the following hypotheses:

  1. Metacommunity dynamics differ between seasonal snapshots (i.e. over time) due to predictable seasonal variation in environmental conditions (e.g. Laudon & Bishop 1999; Laudon, Köhler & Buffam 2004), and differences in the timing of invertebrate metamorphosis, emergence and aerial dispersal (life cycle differences) (e.g. Petersen et al. 1999).
  2. Metacommunity dynamics differ between hierarchal spatial extents, that is, when community structure is examined either at the whole catchment scale or in upstream/downstream sections in the stream network in isolation. More specifically, we expect both environmental and spatial signals to be evident at the whole catchment scale and hypothesize that the underlying dispersal mechanism at this small spatial extent is due to ME as previous studies (including first- to third-order streams) suggest at least moderate dispersal over much larger spatial extents (Heino & Mykrä 2008; Heino et al. 2012).
  3. Metacommunity dynamics differ between upstream and downstream sites (i.e. at lower hierarchal extents) as the dendritic structure will make upstream sites more isolated from each other and from the regional species pool compared with downstream sites (Brown & Swan 2010). This will be evident as a larger effect of spatial factors at downstream sites (due to high dispersal, i.e. ME) while upstream sites are only structured by environmental factors (moderate dispersal, i.e. SS).

To elucidate the dispersal mechanism behind the spatial signals in our study (with reference to hypotheses 1–3), we related different dispersal traits (adult flight, female dispersal and drift) to relevant dispersal dimensions and tested whether the relative importance of species contributing to the spatial signal varies with dispersal capabilities (i.e. strong vs. weak). Following the same reasoning as Ng, Carr & Cottenie (2009), dispersal limitation will be evident if a spatial signal is significant in low dispersal metacommunities (weak dispersers), but only environmental effects are significant in high dispersal metacommunities (strong dispersers), while the opposite is true for high dispersal effects (ME).

Materials and methods

Study site

The stream network selected for this study is a part of the Krycklan river catchment, situated in the northern part of Sweden (county of Västerbotten) (Fig. 1). The catchment is dominated by mixed coniferous forests and wetlands (Buffam et al. 2007), and the underlying bedrock and soil mainly consists of gneiss and moraine, respectively. Annual mean temperature is 1 °C, and annual mean precipitation is c. 600 mm (one-third of the precipitation falls as snow; Ottosson Löfvenius, Kluge & Lundmark 2003).

Figure 1.

Map showing the location of the study area (Krycklan catchment) in Sweden (top right corner) and the location of the sampling sites within the study area.

We selected 30 first-order sites (upstream sites) and 22 second- and third-order sites (downstream sites) for macroinvertebrate sampling and habitat characterization. Thus, all upstream sites were situated upstream of the most headward node (stream confluence) while all downstream sites were situated downstream of at least one node. The most headward sites were located as far up in the stream network as possible (i.e. where stream flow was sufficient for a Surber sampler to work properly) (Fig. 1). At each site, a 30-m representative stretch was selected for the sampling and habitat characterization, which was carried out once during spring (18–24th May) and once during autumn (4–8th October) 2009.

Benthic invertebrate sampling

The macroinvertebrate sampling was performed with a Surber sampler (frame size: 14 × 14 cm, mesh size: 500 μm). In total, three subsamples were collected along the 30-m stretch. One subsample consisted of three Surber samples (i.e. nine Surber samples, and a total sample area of c. 0·18 m2 was collected at each site). The samples were preserved in 70% ethanol and brought back to the laboratory for sorting and identification. The benthic invertebrates were identified to the lowest possible taxonomical level, in most cases to species or genus, but some groups were identified to a higher taxonomical level, for example, Simuliidae (family), Chironomidae (subfamily) and Coleoptera (family).

Habitat characterization and water chemistry sampling

Stream width, depth, flow and canopy cover were measured every 5 m along the same 30-m stretch as the benthic invertebrate sampling was performed. Width and canopy cover were measured once at each transect, while depth and flow were measured at three points along each transect (from each channel edge and from the middle of the stream). Canopy cover was estimated using digital photographs of the canopy, taken from the middle of the stream (at the stream surface) pointing upwards. The images were manipulated in computer software Image-Tools (Health Science Center, University of Texas, USA) so that black pixels represented the canopy and white pixels represented open areas. The percentage of black pixels in each picture was then used to calculate the mean canopy cover (percentage cover) at each site. Substratum composition was measured using a pebble count method (Wolman 1954) where 100 stones were randomly picked up and measured along the 30-m stretch and then divided into eight different substratum classes based on the size of the particles. The percentage of each substratum class could then be calculated from the pebble count data. The number of items of dead wood (branches and logs > 1 cm in diameter) in the stream channel was counted at each site, and the moss cover was estimated by noting the presence/absence of moss on each of the 100 substrate particles that were picked up during the pebble count procedure. Water chemistry samples were taken in conjunction with the invertebrate sampling and habitat characterization. Samples were analysed for major anions and cations, metals (Fe, Al), pH, nutrients (Total-N, Total-P, NO2+NO3), water colour (absorbance at 420 nm), total organic carbon (TOC) and carbon dioxide (CO2).

One characteristic of the Krycklan catchment is that portions of the stream network (upstream sites) are intermittent, that is, they dry out during the summer months. Therefore, an inventory was performed on two occasions during the warmest period of the summer (late June and early July 2009). Coordinates were used to mark the point above which the stream was dry or only consisted of scattered pools without any permanent water flow and sites were classified as affected/not affected by drought. Land use was assessed from digital maps (shape files acquired from the Swedish forestry agency and Lantmäteriet) and calculated for each site in ArcGIS, version 9.3.1 (ESRI, Redlands, California, USA).

Data preparation

Distance matrices

Because dispersal in stream networks can take place at different dimensions, we assessed spatial signals based on (i) overland distances (i.e. the shortest distance between each pair of sampling points), (ii) watercourse distances (i.e. the distance between each pair of sampling points following the stream channel) and (iii) directional downstream distances.

Distances between each pair of sampling sites were calculated in ArcGIS (version 9.3.1) using Hawth Tools (overland distances) and Network Analyst/OD Cost Matrix tool (watercourse distances), after which two triangular distance matrices were constructed. Principal Coordinates of Neighbourhood Matrix (PCNM) analysis, based on both overland and watercourse distance matrices, was then performed using the pcnm function in r package vegan (Oksanen et al. 2011). This was carried out to create the spatial variables (eigenvectors) for further analysis. The spatial eigenvectors based on the overland and watercourse distance matrices will hereafter be called DO and DW, respectively.

To calculate directional downstream distances, a binary connexion diagram was constructed based on the stream network links (stream reaches), denoting the presence/absence of links between each site and all other sites in a downstream direction. Links were also given a weight, based on the geographic distance of each link (reach) (Blanchet, Legendre & Borcard 2008). The connection diagram and the weights were then used to model asymmetric eigenvector maps (AEM) using the aem function in r package AEM (Blanchet 2010). Similarly to the PCNM analysis, we obtained spatial variables (eigenvectors) from the AEM analysis that were used for further analysis. The directional spatial eigenvectors based on downstream flow will hereafter be called AEMD.

Species and environmental matrices

The species abundance matrices consisted of data from different seasons (spring vs. autumn) as well as from sites with different positions within the network, that is, upstream (stream-order I) vs. downstream (stream-order II–III) sites. In addition, smaller species matrices were constructed by dividing invertebrates into groups depending on their dispersal ability. We used three relevant measures of dispersal ability to be related to our modelled spatial variables, following Poff et al. (2006): drifting propensity (DP) (related to AEMD), adult flying strength (AFS) (related to DW and DO) and female dispersal (FD) (related to DW and DO) and assigned taxa to one of three dispersal ability groups: (i) no (only DP), (ii) low and (iii) high dispersal ability. No DP are genera that rarely occur in drift samples (i.e. mainly found during catastrophic drift), low DP are genera that commonly occur in drift samples and high DP are genera that are very abundant (dominant) in drift samples. Low FD are genera that fly < 1 km before laying eggs, and high FD are genera that can fly > 1 km before laying eggs. Low AFS are genera that cannot fly into light breeze (weak flyers), and high AFS are genera that can fly into light breeze (strong flyers). Genera that were not included in Poff et al. (2006) were excluded from the analysis. The species abundance matrices were always transformed using Hellinger transformation prior to statistical analysis (Legendre & Gallagher 2001).

The environmental matrix (E) consisted of the collected water chemistry, hydromorphological and land-use data. All environmental variables were checked for normality and log transformed or square-root transformed if necessary. A centred log ratio transformation (Aitchison 1986; Wang, Meng & Tenenhaus 2010) was performed on the substratum composition and the land-use data (compositional data that sum to one). The centred log ratio transformation is defined as:

display math

where t is the transformed variable value, xj is the original value, xi is the number of parts (portions) in the unit one and p is the total number of parts. This transformation creates linearity in the data and solves the problem of compositional data that sum to unit one (Aitchison 1986; Wang, Meng & Tenenhaus 2010).

Statistical analysis

Effect of environment vs. spatial factors

All statistical analyses were performed in statistical software r (R Development Core Team 2011). First, we selected spatial and environmental variables with a forward selection procedure using the function forward.sel in r package packfor (Dray 2009). To separate the relative effect of environmental and spatial factors on the structure of species communities, we used the function varpart (variance partitioning) in r package vegan (Oksanen et al. 2011). This function uses partial redundancy analysis (pRDA) to calculate how much of the variance in species community structure that can be explained uniquely by each explanatory matrix (here E, DO, DW and AEMD) as well as the shared variance explained by the explanatory matrices. The significance of each testable fraction (pRDA) in the variance partitioning analysis was obtained by using function rda (r package vegan). These analyses were performed separately and in the exact same way for each data set divided by seasonal snapshot, hierarchical level and stream-order group. If our variance partitioning analysis based on overall community structure showed that DO, DW or AEMD could explain a significant amount of the variation in species community structure, we ran the same analysis for that particular season, hierarchical level and stream-order group using our predefined dispersal ability groups (see above) and correspondent dispersal dimension. This allowed us to draw conclusions regarding the mechanisms behind the spatial signals (ME vs. dispersal limitation) (see 'Introduction').

Our variance partitioning approach allows for assessing ecologically meaningful spatial patterns for organisms that disperse along different dimensions (flight, drift). That is, it allows for discriminating between spatial signals and assessing the relevance of each individual fraction (DO, DW or AEMD) by accounting for the confounding influence of the other fractions. For instance, as directional patterns of water flow (AEMD) are not a priori important for flying invertebrates (i.e. aerial flight is uncoupled from the downstream transport of water), we can account for AEMD in the analyses of pure effects of DO and DW. This allows for the exclusive assessment of factors relevant for dispersal through flight without being masked by factors associated with hydrological features that are known to be important for in-stream processes. By contrast, accounting for DO and DW when assessing pure effects of AEMD allows for an assessment of the exclusive importance of water flow for drifting invertebrates without being confounded by other spatial processes that are not relevant or even realistic.


At the largest spatial extent (whole network scale), both local environmental factors (E) and spatial factors (DO, DW and AEMD) explained a significant portion of the species community structure. In spring, the variance explained by E, DO, DW and AEMD was 13%, 11%, 7% and 7%, respectively, while in autumn only E (9%) and DO (10%) were significantly related to species community structure. At upstream sites, E explained a significant and relatively large portion of the species community structure (spring: 22%, autumn: 17%) while the effect of spatial factors was less evident (except for AEMD during autumn, which explained 5% of species community structure). In downstream sites, E also explained a significant portion of the species community structure (spring: 7% [significant at α = 0·1], autumn: 16%), but significant spatial structures were only detected in autumn (DO: 5%, AEMD: 6%) (Fig. 2). It was also notable that (i) the effects of dispersal dimensions (DO, DW and AEMD) were highly contingent on the scale of observation and time of sampling, (ii) in general, DO and AEMD explained a larger portion of the species community structure than DW and (iii) more variation could be explained in spring compared with autumn, in particular at the whole catchment scale (total explained fraction in spring: 57% and autumn: 35%) (Fig. 2).

Figure 2.

Results of the variance partitioning analyses between explanatory variables and the species data. The figure shows the amount of variation (%) in the species data that is explained by local environment (E), overland distance eigenvectors (DO), watercourse distance eigenvectors (DW), directional eigenvectors (AEMD), unexplained variation and the total shared variance (i.e. the sum of all pairwise shared components + the shared variance explained between all four explanatory variables). The tests are divided by season (spring vs. autumn) and hierarchical levels (i.e. whole network scale vs. upstream (order I) vs. downstream (order II–III) sections). The level of significance is indicated next to the bars (**significance at α 0·01, *significance at α 0·05, _significance at α 0·10).

Our analyses including species dispersal traits (only analysed for the significant spatial structures in our overall test shown in Fig. 2) showed varying changes in the spatial signal between low and high dispersal communities. The analyses of spring data at the whole network scale showed somewhat consistent patterns. Along two dispersal dimensions (DO and DW), there was a change from a significant spatial structure for low AFS and FD to a non-significant structure (only environmental effects) for high AFS and FD (except FD vs. DW which showed non-significant results for both dispersal-strength groups). This indicates a change from limited dispersal in our low dispersal metacommunity (weak dispersers) to moderate dispersal in our high dispersal metacommunity (strong dispersers). Along the third dispersal dimension (AEMD), the pattern was different. Here, significant spatial structures were found for all dispersal groups with an increased variance explained by high DP compared with no and low DP. This indicates an increased effect of dispersal with increasing dispersal capacity and thus possibly ME. The analyses of autumn data at all hierarchal levels showed less consistent patterns, indicative of more than one dispersal mechanism (as well as non-significant results between dispersal capacity groups) even along the same dispersal dimension (Table 1a,b).

Table 1. Results from the variance partitioning analyses between explanatory matrices (local environment [E], overland distance eigenvectors [DO], watercourse distance eigenvectors [DW] and directional eigenvectors [AEMD]) and species abundance matrices in (a) spring and (b) autumn. Subsets of the species data set based on species dispersal traits and capacities were analysed separately: no drifting propensity, low and high drifting propensity, female dispersal capacity and adult flying strength. All results in the table are corrected for the variance explained by E and DO and/or DW and/or AEMD. Thus, the table only shows the variance (adjusted R2) uniquely explained by the spatial variable of interest (DO, DW and AEMD). Environmental variables were significant in all tests except where they were not retained in forward selection (indicated as nsf)
Stream order/hierarchal levelDispersal dimensionDispersal traitNoLowHigh
  1. a

    P < 0·10.

  2. b

    P < 0·05.

  3. c

    P < 0·01.

  4. ns, non-significant (P > 0·10).

  5. d

    Only E, DO and DW retained in forward selection.

  6. e

    Only E, DO and AEMD retained in forward selection.

  7. f

    No explanatory variables retained in forward selection.

  8. g

    Only E and AEMD retained in forward selection.

Network scaleDOFemale dispersal0·07bns
Network scaleDOAdult flying strength0·04ansd
Network scaleDWFemale dispersalnsns
Network scaleDWAdult flying strength0·03bnsd
Network scaleAEMDDrifting propensity0·08b0·05be0·11c
Network scaleDOFemale dispersal0·10cns
Network scaleDOAdult flying strengthns0·05b
IAEMDDrifting propensitynsens0·10c
II–IIIDOFemale dispersalnsens
II–IIIDOAdult flying strength0·05aensf
II–IIIAEMDDrifting propensity0·22c gnsdnsd


Using streams as a model of hierarchical ecosystems with diffuse habitat boundaries, our results demonstrate complex community organization at very small spatial extents. Differences in metacommunity dynamics were observed between our seasonal snapshots, thereby providing support for our first hypothesis. For example, at the entire network scale, DW and AEMD were significant in spring but not in autumn, which could reflect seasonal differences in the dispersal pathways used by abundant species in our study due to, for example, seasonal differences in the timing of adult aerial dispersal and drifting magnitude (see more detailed discussion about mass effects below). In addition, the total variance explained was consistently higher in spring compared with autumn, indicating generally stronger responses of the invertebrates not only to spatial gradients, but also to environmental gradients. This is not surprising as the spring flood is known to cause environmental extremes, not only in terms of hydrology, which may influence, for example, dispersal magnitude (e.g. Waringer 1992), but also in terms of water chemistry (e.g. decrease in pH and associated changes in other water chemistry variables; e.g. Buffam et al. 2007), which invertebrates are known to respond to (Ormerod et al. 1987; Giller & Malmqvist 1998).

Clear differences were also observed between hierarchical scales (whole catchment vs. stream-order groups), and in line with our predictions, metacommunity structure at the whole catchment scale was structured by a combination of environmental and spatial factors. We also investigated the underlying dispersal mechanism behind our spatial structures. That is, whether the sorting of individuals along environmental gradients was limited by low dispersal (SS + low dispersal) or subsidized through successful colonizations (ME) (Leibold et al. 2004; Ng, Carr & Cottenie 2009). Our analyses of the dispersal-strength groups were not entirely conclusive, but they gave some indications of the dispersal mechanisms at play at the whole catchment scale. For example, during spring, the spatial signal at the whole catchment scale changed from significant for low AFS and FD to non-significant for high AFS and FD along two dispersal dimensions (DO and DW), which indicates a change from limited to moderate dispersal effects for invertebrates flying across and along the watercourses, respectively. A different pattern was found for DP along the third dispersal dimension (AEMD), indicating a change from moderate to high dispersal effects for drifting invertebrates. This suggests that during spring, dispersal limitation is occurring for flying invertebrates through DO and DW, while ME may be evident for invertebrates drifting downstream.

Although it is tempting to suggest that ME is the most likely mechanism behind any spatial signal at this small spatial scale as the likelihood of species to reach all sites is high (both through overland and watercourse pathways), our analyses of different dispersal-strength groups suggested that this may not be the case along all dispersal dimensions. This was in contrast to our predictions made in the second hypothesis. However, in agreement with our results, dispersal limitation has been shown to be important in very small catchments, and in some cases, the recruitment of larvae at the reach scale has been found to be the results of just a few mating individuals (Bunn & Hughes 1997). That overland dispersal may be limited at our largest spatial extent is also in line with the findings of Petersen et al. (1999) showing that many common species in our study [e.g. Leuctra nigra (Olivier, 1811) and Nemurella pictetii (Klapálek, 1900)] do not travel far from their native habitat (lateral dispersal <51 m for 90% of the individuals). In contrast, Macneale, Peckarsky & Likens (2005) showed that rare, long dispersing events can guarantee colonization of nearby streams in small catchments. However, a prerequisite for a successful colonization and establishment to occur through rare dispersal events is a sufficiently long time period. It is therefore possible that apparent patterns of dispersal limitation can be observed in highly disturbed and sometimes even temporary habitats similar to the ones in our study, as species are constantly ‘forced’ to recolonize disturbed/open patches. Thus, species with low dispersal capacities may never have enough time to establish at all sites. What we are observing at this small spatial scale is therefore more likely to be dispersal limitation acting on shorter time scales. This might also explain why studies covering larger spatial extents, but probably looking at hydrologically more stable stream types (i.e. not including intermittent streams), have seen weak or no evidence of dispersal limitation (Heino & Mykrä 2008; Shurin, Cottenie & Hillebrand 2009; Heino et al. 2012). On the contrary, Mykrä, Heino & Muotka (2007) and Soininen (2004), who analysed invertebrate and diatom species distributions in streams over different spatial scales, respectively, showed that the variance explained by spatial variables do not necessarily decrease with a decreasing spatial extent, which could partly be in support of our findings. However, note that none of these studies (and very few metacommunity studies in general, but see Ng, Carr & Cottenie 2009) have attempted to elucidate whether their spatial signals origin from dispersal limitation or mass effects, which makes it hard to draw any firm conclusions about where the breakpoint is (in terms of spatial extent) between low, moderate and high dispersal communities in natural stream ecosystems. However, our study shows that dispersal limitation may occur on a much smaller spatial extent than previously expected and that it is likely to be dependent on the stream type investigated.

Mass effects are also likely to occur in stream networks, and especially at small spatial scales such as in our study where connectivity between streams and thus dispersal is assumed to be high (Leibold et al. 2004). For example, small- and/or large-scale disturbances, which are continuously and naturally occurring in streams and known to be a strong structuring force for species communities (Townsend 1989), may cause source-sink dynamics (Pulliam 1988; Mouquet & Loreau 2003) through processes such as colonization of disturbed sites from nearby undisturbed sites (Giller & Malmqvist 1998). Townsend (1989) also pointed out that dispersal events such as downstream drift (of abundant upstream species better adapted to more acid conditions) and upstream flight (of abundant downstream species better adapted to circumneutral conditions) may cause source-sink (ME) dynamics (Mouquet & Loreau 2002, 2003), leading to species existing in habitats where they would not otherwise be expected if it was not for high dispersal rates (e.g. Pulliam 1988). Upstream/downstream movements are likely to occur continuously in our system (except during winter), but perhaps especially during spring when (i) many stoneflies (which on average made up 30% of the individuals per site and season in this study) emerge and female flight is mainly occurring along the watercourses (Petersen et al. 1999; Macneale, Peckarsky & Likens 2005) and/or (ii) the rate of downstream movement (drift) may be mediated by high flows (e.g. Waringer 1992). Altogether, theory, and to some extent our results, suggests that in continuous and highly dynamic systems such as stream networks, there can be several dispersal mechanisms at play at the same time, occurring through different dispersal pathways (e.g. Townsend 1989; Macneale, Peckarsky & Likens 2005). That is, both dispersal limitation (through overland dispersal and flight along the watercourses) and ME (through downstream drift) could simultaneously contribute to the spatial signals observed at the largest spatial extent.

A clear finding from our comparison between upstream and downstream sites was that our upstream sites were consistently and strongly related to environmental variables in both seasons. Further, we found no effect of spatial variables in upstream sites during spring. We did, however, find a significant effect of AEMD during autumn. This is not surprising, as drift is the type of dispersal that is most likely to show an effect in our upstream sites considering that we had at least two sampling sites along each upstream stream segment. However, when removing the effect of AEMD during autumn, upstream sites were completely unaffected by spatial variables associated with overland dispersal (DO) and flight along the watercourses (DW), indicating mainly SS dynamics. Previous analyses of our abiotic data suggested that the variability in water chemistry was higher between our upstream sites compared with downstream sites (permutation test of multivariate homogeneity of groups dispersions, autumn: F1,50 = 20·21, P < 0·001, spring: F1,50 = 20·19, P < 0·001; E. Göthe, unpublished results), indicating a strong environmental gradient. Also, the recurrent low flows in some of the most headward sites may represent a strong filter for species communities. Such strong environmental gradients, known to be important for invertebrate community structure (e.g. Poff 1997), together with sufficient dispersal allowing species to sort along those gradients could partly explain why SS dominates at the top of Krycklan catchment.

In downstream sites, results were less consistent between seasons. We found weak environmental effects coupled with no spatial effects during spring, while both environmental and spatial factors (DO and AEMD) were significant during autumn. Also, the spatial signals detected at lower hierarchal scales during autumn were more difficult to interpret as our analyses of dispersal-strength groups indicated more than one possible dispersal mechanisms at play through downstream drift (AEMD) and weak or non-significant patterns for overland pathways. However, theory strongly suggests that any spatial signal in downstream sites is more likely to originate from high dispersal as they have a more ‘beneficial’ position within the network in terms of dispersal in comparison with upstream sites. That is, colonization can occur through different routes, including (i) downstream drift from upstream sites (Waringer 1992), (ii) colonization from upstream and downstream sites through flight along the watercourses (Macneale, Peckarsky & Likens 2005) and (iii) sites can also more easily be reached overland due to shorter flight distances (Clarke et al. 2008), which creates prerequisites for ME to occur (Brown & Swan 2010). Considering this, it is also possible that the absence of a spatial signal in downstream sites during spring is an indication of that unlimited drift has occurred at small spatial extents, swamping both environmental (significant at α 0·10) and spatial signals (non-significant). Moreover, it is very unlikely that any spatial effect in downstream sites is due to dispersal limitation, especially as our results mainly indicated moderate dispersal/sufficient dispersal, allowing species to track changes in environmental conditions, that is, SS (Leibold et al. 2004), in the more spatially isolated upstream sites (Ng, Carr & Cottenie 2009). However, our analyses of dispersal-strength groups did not support this assumption. Despite this limitation, our results partly confirm our second hypothesis and are also in agreement with previous studies performed over larger spatial extents, where upstream sites were shown to be structured only by environmental factors, that is, SS (Brown & Swan 2010).

The use of dispersal-strength groups certainly has great potential for elucidating underlying dispersal mechanisms in metacommunity analyses. However, we acknowledge that very little is still known about actual dispersal rates and capacities of benthic invertebrates as it is very hard to measure directly (Bohonak & Jenkins 2003). Therefore, to separate dispersal mechanisms by only using benthic invertebrate dispersal traits could be difficult at present time, and thus, interpretation should be made cautiously. This uncertainty may also explain some of the inconsistency in our results from the analyses of our dispersal-strength groups. Thus, further research is warranted to increase the knowledge of species-specific dispersal rates and distances, and their consequences for metacommunity assembly. However, despite the current limitations to inferences of realistic patterns of dispersal using available dispersal traits, our results suggest that dispersal is highly context dependent and spatially contingent.

To conclude, our main findings were that (i) the timing of sampling (i.e. seasonal snapshots) was important for detecting differences in metacommunity structure, most likely reflecting seasonal differences in dispersal magnitude along specific dispersal pathways as well as differences in the strength of environmental gradients, (ii) the importance of spatial factors was highly dependent on the scale of observation and on specific dispersal dimensions and (iii) upstream sites were mainly structured by environmental factors and consistently unaffected by spatial signals associated with overland dispersal and flight along the watercourses. These findings emphasize the complexity of natural systems, suggesting that it may be inadequate to generalize results based on analyses performed on (i) single scales, (ii) snapshots in time and (iii) simple estimates of space, which are not specific to the system and organism studied. Moreover, our results indicate that dispersal constraints imposed by the dendritic structure of stream networks (Brown & Swan 2010; Carrara et al. 2012) and community assembly processes that have previously been expected to take place at broader landscape scales (e.g. dispersal limitation) (Heino & Mykrä 2008; Heino et al. 2012) may also be evident at very small spatial extents. More generally, we show that the study of hierarchically organized ecosystem uncovers complex patterns of metacommunity organization that may deviate substantially from those of systems with insular structure and discrete habitat boundaries (Carrara et al. 2012), for example, results indicated that more than one dispersal mechanism can occur at the same time, but through different dispersal dimensions. These findings are not only insightful from a scientific point of view but also from an applied perspective, as dispersal-driven communities (independent of dispersal mechanism) would require more focus on the management of landscapes in addition to local environmental conditions (Bengtsson 2010). This means that, even at the small scale of boreal headwater catchments, a landscape management approach that takes the dendritic nature of stream networks into account may be needed to effectively conserve stream biodiversity (Carrara et al. 2012). However, more studies are needed over small spatial scales to elucidate the generality of our results.


We thank Lars Eriksson and Dan Evander who helped with the taxonomy, Peter Carlson who helped with field sampling and Leif Göthe who provided both GIS support and assistance in the field. We are also very grateful to the crew at Svartberget research station for helping us with various practical issues, especially Hjalmar Laudon, Peder Blomqvist and Viktor Sjöblom, for their assistance during field sampling. We also thank two reviewers and Richard K. Johnson for helpful comments that improved the manuscript.