Species richness at the guild level: effects of species pool and local environmental conditions on stream macroinvertebrate communities

Authors

  • Mira Grönroos,

    Corresponding author
    1. Finnish Environment Institute, University of Oulu, P.O. Box 413, FI-90014 Oulu, Finland
    2. Department of Biology, University of Oulu, P.O. Box 3000, FI-90014 Oulu, Finland
      Correspondence author. E-mail: mira.gronroos@environment.fi
    Search for more papers by this author
  • Jani Heino

    1. Finnish Environment Institute, University of Oulu, P.O. Box 413, FI-90014 Oulu, Finland
    2. Department of Biology, University of Oulu, P.O. Box 3000, FI-90014 Oulu, Finland
    Search for more papers by this author

Correspondence author. E-mail: mira.gronroos@environment.fi

Summary

1. A fundamental question in ecology is which factors determine species richness. Here, we studied the relative importance of regional species pool and local environmental characteristics in determining local species richness (LSR). Typically, this question has been studied using whole communities or a certain taxonomic group, although including species with widely varying biological traits in the same analysis may hinder the detection of ecologically meaningful patterns.

2. We studied the question above for whole stream macroinvertebrate community and within functional feeding guilds. We defined the local scale as a riffle site and the regional scale (i.e. representing the regional species pool) as a stream. Such intermediate-sized regional scale is rarely studied in this context.

3. We sampled altogether 100 sites, ten riffles (local scale) in each of ten streams (regional scale). We used the local-regional richness regression plots to study the overall effect of regional species pool on LSR. Variation partitioning was used to determine the relative importance of regional species pool and local environmental conditions for species richness.

4. The local-regional richness relationship was mainly linear, suggesting strong species pool effects. Only one guild showed some signs of curvilinearity. However, variation partitioning showed that local environmental characteristics accounted for a larger fraction of variance in LSR than regional species pool. Also, the relative importance of the fractions differed between the whole community and guilds, as well as among guilds.

5. This study indicates that the importance of the local and regional processes may vary depending on feeding guild and trophic level. We conclude that both the size of the regional species pool and local habitat characteristics are important in determining LSR of stream macroinvertebrates. Our results are in agreement with recent large-scale studies conducted in highly different study systems and complement the previous findings by showing that the interplay of regional and local factors is also important at intermediate regional scales.

Introduction

A fundamental question in ecology is which factors determine species richness at regional and local scales. Regional-scale variation in species richness has been shown to depend mainly on biogeographical processes, including speciation, extinctions and productivity (Huston 1994; Rosenzweig 1995). Local-scale variation in species richness is, in turn, often assumed to be determined mainly by local processes, including biotic interactions, abiotic fluctuations and habitat heterogeneity (Tilman & Pacala 1993; Field et al. 2009). However, there are typically interactive effects between processes at different scales, and local species richness (LSR) may depend not only on local but also biogeographical processes (Cornell & Lawton 1992; Ricklefs & Schluter 1993).

A seminal approach to study the relative importance of local and regional processes in determining LSR was introduced by Terborgh & Faaborg (1980). In this commonly used approach, LSR is regressed against regional species richness (RSR) (Hawkins & Compton 1992; Caley & Schluter 1997; Ricklefs 2000; Cornell, Karlson & Hughes 2008). According to the classical interpretation, a linear relationship suggests that local richness depends mainly on the supply of colonists from the regional species pool. On the other hand, LSR may also reach an asymptote with increasing RSR, suggesting that factors operating at a local scale limit the richness of local communities (Cornell & Lawton 1992).

A majority of the LSR–RSR studies have found linear relationships (Hawkins & Compton 1992; Hugueny & Paugy 1995; Caley & Schluter 1997; Griffiths 1997; Witman, Etter & Smith 2004; Cornell, Karlson & Hughes 2008), but also curvilinear relationships have emerged (Ricklefs 2000; Stendera & Johnson 2005; Krasnov et al. 2006). In some studies, the form of the relationship has depended on scale, organism group or successional stage (Angermeier & Winston 1998; Canning-Clode et al. 2009; Soininen et al. 2009). Thus, it may be difficult to draw general conclusions about the form of the relationship because of such context dependency. Several papers have also discussed the problems of the approach and their solutions, concerning, for instance, the discrimination between linear and curvilinear relationship (e.g. Griffiths 1999), effect of under- and overestimation of species richness (e.g. Srivastava 1999) and the underlying processes (He et al. 2005; Hillebrand 2005; Shurin & Srivastava 2005). Originally, the LSR–RSR approach was interpreted to measure the strength of interspecific interactions (Terborgh & Faaborg 1980), but more recent interpretation is that any factor at the local scale may constrain LSR (e.g. He et al. 2005). Nowadays, the LSR–RSR approach is regarded as a good starting point and has recently been used for studying the overall species pool effect (Kristiansen et al. 2011). However, the LSR–RSR approach should be used in association with other approaches because, on its own, it does not provide information on what local factors constrain species richness.

The species pool hypothesis (Zobel 1997) and the concept of environmental filters (Keddy 1992) integrate various processes acting at hierarchical spatial scales into a single framework. Large-scale processes (e.g. speciation and extinction) determine the characteristics of a regional species pool (e.g. RSR), which then translate into variation in species richness at smaller scales through dispersal and subsequent environmental filtering. At each transition, from a larger to a smaller scale, different filters are operating to reduce species richness from that of the species pool above in the hierarchy. At the smallest scales, biotic interactions and abiotic factors constrain species richness. Biotic interactions are extremely difficult to measure in multispecies systems, but abiotic environmental factors can be relatively easily attained and used to explain LSR. Because of the difficulties in using the LSR–RSR approach to assess the relative importance of RSR and local environment (White & Hurlbert 2010), there is a recent trend towards simultaneous investigation of the effects of RSR and local environment on LSR (Harrison & Cornell 2008; White & Hurlbert 2010; Kristiansen et al. 2011).

The studies of the effects of regional and local processes on LSR have typically treated communities as a whole (e.g. benthic macroinvertebrates; Witman, Etter & Smith 2004) or with regard to a certain taxonomic group (e.g. palms; Kristiansen et al. 2011). However, studying the question with organisms that have widely varying biological traits and including them in the same analysis may hinder the detection of ecologically meaningful patterns. Thus, the effects of RSR and local environment on LSR should preferably be examined with both whole communities and sets of species potentially showing strong interspecific interactions. Guilds are defined as a group of species that exploit the same recourses in a similar way and are thus more likely to show strong interspecific interactions (Fauth et al. 1996). Different guilds may also have differing environmental requirements depending on the resource used or feeding habits (Parker, Parker & Vale 2001; Pearman 2002; LeCraw & Mackereth 2010). In consequence, the relative importance of RSR and local factors may differ between guilds. To our knowledge, only one study has investigated the LSR–RSR relationship within groups of species that are likely to show strong interspecific interactions (Russell et al. 2006).

In biogeographical research, regional species pool often refers to the pool of species that have coevolved together or co-occurred during their evolutionary history (Ricklefs 1987, 2008). However, when using LSR–RSR approach, regional species pool is often defined as containing all the species that can colonize each locality (Srivastava 1999; also termed ‘local source pool’, see Guisan & Rahbek 2011). Local scale, on the other hand, is recommended to be such that all the species are able to encounter each other in ecological time (Srivastava 1999). We follow the latter definitions and emphasize that the colonization from regional species pool to localities is considered to happen in a relatively short time period.

Here, we used stream macroinvertebrates as the model group for studying the relative importance of RSR and local environmental variables (ENV) to LSR. We defined the local scale as a riffle site (c. 50 m2) and the regional scale (i.e. the regional species pool) as a stream (extent c. 2 km). Using these scales and given that lotic macroinvertebrates mainly disperse short distances along stream corridors (Petersen et al. 2004; Macneale, Peckarsky & Likens 2005), dispersal should not be substantial across regional species pools (i.e. across streams) in ecological time, while dispersal should not generate strong spatially-structured patterns within a region (i.e. within streams). In general, studies considering intermediate-sized regional scales are demanded when examining regional effects on LSR (Harrison & Cornell 2008). In previous studies of the LSR–RSR relationship in stream macroinvertebrates, a drainage basin has been used as the regional scale and a riffle site as the local scale (Heino, Muotka & Paavola 2003; Marchant, Ryan & Metzeling 2006). However, there is also much variation in species richness at smaller scales (Palmer & Poff 1997; Heino, Louhi & Muotka 2004), and thus studies at regional scales smaller than a drainage basin are also needed.

We studied two main questions: (i) What is the relative importance of regional species pool and local environmental conditions in determining the LSR of stream macroinvertebrates? (ii) Does the pattern differ between guilds? To answer these questions, we used the LSR–RSR approach to examine overall species pool effect (i.e. the effect of RSR) and then studied the relative importance of RSR and local ENV on LSR. Based on previous findings, we first hypothesized that for the whole community, the species pool effect should be significant and thus the relationship between LSR and RSR would be linear. Second, we expected that the species pool effect should be weaker and the relationship between LSR and RSR curvilinear (or less steep) for each guild (due stronger interspecific interaction within guilds). Third, we expected that in addition to RSR, also ENV would account for an important part of variation in LSR. Fourth, we expected that the relative importance of RSR and ENV varies between guilds.

Materials and methods

Study area

The study area is located in the Koutajoki drainage basin in the Oulanka National Park in Finland (centred on 66˚21′N, 29˚26′E). The extent of the present study area is c. 16 × 10 km. The bedrock of the area is highly variable, and there are extensive occurrences of calcareous rocks. Accompanied by considerable altitudinal differences, this geological variability is reflected in highly variable vegetation, ranging from old-growth coniferous forests to mixed-deciduous riparian woodlands, and from nutrient poor bogs to fertile fens. These factors also provide the basis for a high variability of stream habitats within the drainage basin. Headwater streams in the drainage basin are generally near-pristine, and they are characterized by circumneutral to alkaline water, low to high levels of humic substances and low to moderate nutrient concentrations (Heino et al. 2009). A detailed description of the drainage basin can be found elsewhere (Malmqvist et al. 2009).

Ten second- or third-order streams that drain into the River Oulankajoki were sampled in the first half of September 2009 (Fig. S1, Supporting Information). Ten riffles were sampled in each of the 10 streams (altogether 100 sites were surveyed). The distance between the consecutive riffle sites in each stream ranged from c. 50 to 200 m, and the sites were located at maximum two kilometers upstream from their confluence with the River Oulankajoki.

Macroinvertebrate data

At each riffle, a 2-min kick-net (net mesh size 0·3 mm) sample that covered most microhabitats present at c. 50 m2 area was taken. With this sampling effort, more than 70% of species occurring at a site in a given season is typically obtained (Mykrä, Ruokonen & Muotka 2006). Samples were immediately preserved in ethanol in the field, and further processed and identified in the laboratory. All macroinvertebrates, including often-omitted groups, such as mites (Hydracarina), nonbiting midges (Diptera: Chironomidae) and blackflies (Diptera: Simuliidae), were identified to species, species group or genus level.

Macroinvertebrate taxa were assigned to five guilds [functional feeding guilds (FFGs)] that were (i) filterers that obtain fine particulate organic matter from the water column, (ii) gatherers (=detritivores) that consume fine particulate organic matter deposited on stream bottom, (iii) predators that either consume parts of other animals or engulf whole prey individuals, (iv) scrapers (=grazers) that graze algae and biofilm growing on various in-stream surfaces and (v) shredders that feed on coarse leaf material by shredding large fragments. The assignment of species to FFGs was mainly based on the study by Moog (2002). In this system, one to 10 points are given for a species depending on its association with each FFG. Here, a species with ≥5 points for a given FFG was assigned to belong to the respective group. Additional information was derived from the studies by Merritt & Cummins (1996) and Vieira et al. (2006). We used our expert criterion in the few cases this method rendered inconsistent classifications. Eight taxa could not be assigned into any guild, because information on their feeding habits was not available (including mostly poorly known nonbiting midges). These taxa were, however, included in the analyses of the whole community.

Simple divisions of species into guilds may be criticized. For example, there is evidence that the feeding roles of freshwater macroinvertebrates may vary with the larval stage, as well as temporally and geographically, and many taxa may be highly flexible in their feeding habits (Mihuc 1997; Dangles 2002; Moog 2002; Vieira et al. 2006). However, in the context of the present study, such a simple classification was defensible. If a more defined classification had been used (e.g. intermediate classes such as scraper-gatherers), species richness in each more defined FFG would have been too low for meaningful comparisons. Additionally, the use of similar simple guild classification systems is common in modern ecological studies, comprising various organism groups (LeCraw & Mackereth 2010; Both et al. 2011; Kissling, Sekercioglu & Jetz in press).

Environmental data

Environmental characteristics were measured at each riffle site immediately after the macroinvertebrate sampling. Riparian tree species composition (coniferous vs. deciduous) was assessed in a 50-m section on both banks directly upstream of the sampling site. Shading was estimated visually as per cent canopy cover at the whole study section, based on mean of independent estimates by three field workers. Stream width was measured at five cross-stream transects. Depth and current velocity (at 0·6 × depth) were measured at 30 random locations. Visual estimates of bryophyte cover (%) and substratum particle size class cover (%) were made at ten randomly spaced 50 × 50 cm quadrates. Substratum particle size was classified according to a modified Wentworth scale as follows: (i) sand (diameter 0·25–2 mm), (ii) gravel (2–16 mm), (iii) pebble (16–64 mm), (iv) cobble (64–256 mm), (v) small boulder (256–512 mm) and (vi) large boulder (>512 mm). To describe the average particle size, an index was calculated: particle size classes were given a grade from 1 to 6 (1 for sand, 6 for large boulder), and weighted average (based on percentage cover) was then calculated (for a similar approach, see Heino, Muotka & Paavola 2003). In addition, particle size class diversity and the standard deviations of depth, current velocity and bryophyte cover were used as measures of local habitat heterogeneity. Particle size class diversity was calculated following the Shannon equation (for a similar approach, see Townsend, Scarsbrook & Dolédec 1997). In total, 11 local (riffle scale) physical ENV were obtained. These variables were used in the statistical analyses.

We did not measure directly the amount of resources such as leaf litter and algal biomass. However, these variables are likely to correlate with some of the measured variables. For instance, high current velocities are likely to remove detrital matter (Lepori & Malmqvist 2007), while bryophyte cover may help to retain it at a site (Suren & Winterbourn 1992). Shading, on the other hand, may reduce algal biomass (Death & Zimmermann 2005). We thus believe that our set of ENV should capture the most important species richness-environment relationships either directly or indirectly and describe the overall environmental characteristics important for stream macroinvertebrates.

In addition to the riparian and in-stream variables, water samples were collected from the uppermost and the lowermost riffle site in each stream, and they were analysed for pH, conductivity, alkalinity, total phosphorus, total nitrogen, phosphate phosphorus, nitrate-nitrite nitrogen, ammonium nitrogen and colour using Finnish national standards (National Board of Water and the Environment 1981). Chemical measurements were only used to represent variation between the study streams (Table 1). Owing to the close proximity of riffles in a given stream, water chemistry was not assumed to vary meaningfully within the same stream, and thus the chemical variables were not used to explain LSR. The measured physical and chemical ENV have previously been found to be highly influential local factors in explaining variation in macroinvertebrate communities of headwater streams (Sandin & Johnson 2004; Mykrä, Heino & Muotka 2007; LeCraw & Mackereth 2010).

Table 1.   Environmental characteristics of the 10 streams sampled. For physical variables, mean of 10 riffles are given (min and max values in parentheses). For chemical variables, only mean values of the uppermost and the lowermost sampling sites are shown. Also mean, min and max for the observed and Chao-estimated species richness are given
 A-AnsapuroJ-RytipuroJuhtipuroKulmakkapuroMerenojaMetsosuonpuroP-RytipuroPutaanojaUopajanpuroY-Ansapuro
Shading (%)30 (5–75)62 (10–85)40 (25–65)28 (10–45)30 (5–80)30 (5–75)30 (7–55)17 (0–35)35 (5–75)33 (10–60)
Deciduous (%)27 (15–35)44 (15–80)28 (17–35)36 (15–55)44 (10–90)23 (15–35)43 (25–60)31 (10–70)27 (15–38)32 (20–50)
Width (cm)126 (79–164)96 (61–127)127 (100–188)314 (163–530)376 (254–600)115 (88–148)140 (90–262)494 (365–760)273 (240–368)147 (121–181)
Depth (cm)
 Mean12 (9–13)25 (21–29)24 (17–29)28 (20–39)22 (13–30)18 (14–22)21 (17–31)27 (21–42)25 (20–32)12 (9–15)
 SD4 (2–7)8 (4–11)7 (4–11)9 (4–15)9 (4–14)6 (4–9)6 (5–7)9 (5–18)8 (5–11)5 (3–8)
Velocity (m s−1)
 Mean0·38 (0·28–0·48)0·68 (0·50–0·86)0·55 (0·42–0·67)0·79 (0·55–1·11)0·57 (0·38–0·79)0·27 (0·17–0·44)0·41 (0·22–0·60)0·60 (0·44–0·77)0·64 (0·45–0·85)0·29 (0·16–0·39)
 SD0·21 (0·14–0·28)0·37 (0·24–0·50)0·30 (0·22–0·48)0·38 (0·20–0·53)0·34 (0·13–0·60)0·19 (0·11–0·32)0·21 (0·08–0·34)0·26 (0·14–0·48)0·27 (0·15–0·44)0·21 (0·11–0·31)
Bryophyte cover (%)
 Mean22 (0–83)51 (23–87)59 (0–89)19 (0–70)68 (1–99)25 (0–78)5 (0–34)21 (0–86)50 (3–94)42 (0–83)
 SD9 (0–30)20 (10–35)24 (0–36)13 (0–28)13 (3–25)14 (0–36)4 (0–23)9 (0–30)15 (4–31)20 (0–36)
Particle size diversity1·3 (1·1–1·6)1·4 (1·2–1·7)1·2 (0·7–1·6)1·1 (0·8–1·4)1·0 (0·6–1·3)1·0 (0·3–1·4)1·3 (1·0–1·6)1·2 (1·1–1·4)1·3 (1·0–1·8)1·1 (0·7–1·4)
Particle size (see text)3·1 (2·5–4·9)4·0 (3·3–5·0)4·5 (3·5–5·7)4·6 (3·5–5·5)4·5 (2·4–5·7)2·5 (1·4–3·6)3·2 (1·7–5·4)3·3 (2·2–4·8)3·7 (2·2–5·2)4·4 (2·1–5·4)
pH7·27·47·56·67·17·38·17·77·87·3
Conductivity (mS m−1)8·366·236·723·368·157·3718·26517·7520·879·23
Alkalinity (mmol L−1)0·720·640·590·220·660·651·701·251·730·80
Total phosphorus (μg L−1)6·39·610·411·04·73·93·66·67·24·5
Total nitrogen (μg L−1)17615315911814288189229245162
PO4 (μg L−1)2·62·552·72·33·5<2<23·52·72·3
NO2 + NO3 (μg L−1)19·53·78·7<215·54·616·52810·511·3
NH4 (μg L−1)<512·55·25<511<5<5<5<5<5
Colour (Pt mg L−1)6010512521060704012010065
Species richness
 Observed33 (25–40)30 (17–43)23 (7–28)29 (15–42)34 (24–42)27 (16–36)38 (19–50)24 (15–29)41 (29–48)35 (19–41)
 Chao-estimated37 (30–47)38 (23–54)27 (7–42)36 (19–48)38 (29–51)37 (20–59)49 (21–59)33 (15–76)53 (31–81)40 (20–54)

Statistical analyses

In this study, a riffle is considered as the local scale and a stream as the regional scale. Calculation of LSR and RSR depended on the statistical analysis used.

In the LSR–RSR approach, LSR was calculated as mean species richness across all 10 riffles for each stream to avoid pseudoreplication (Srivastava 1999). RSR was calculated as cumulative species richness of all the 10 riffles in a stream. The determination of linear vs. curvilinear relationship between LSR and RSR was conducted according to a decision tree approach for log-transformed data (Griffiths 1999). In the first phase, regression analysis was used for log(x + 1)-transformed data to test whether the regression line passed through origin (intercept not different from 0). If the intercept was significantly higher than 0, the shape was interpreted to be curvilinear. If the intercept did not differ from 0, the second test was conducted, and a new regression was fitted for log-transformed data. One-sample two-tailed t-test was then used to test whether the slope for log-transformed regression differed from 1. Slopes significantly below 1 were interpreted to indicate curvilinearity. If the slope did not differ from 1, the shape of the relationship was interpreted to be linear. Regressions were not forced through origin, because doing so inflates R2 values (Griffiths 1999). Similar log-log regressions have been used in several studies of the LSR–RSR relationship (Krasnov et al. 2006; Cornell, Karlson & Hughes 2008; Canning-Clode et al. 2009; Soininen et al. 2009).

Linear multiple regression analysis and variation partitioning were used to study the relative contribution of RSR and local ENV on LSR. Prior to the regressions, ENV were tested for normality and transformations were used when necessary. To reduce the number of explanatory variables and avoid multicollinearity, factor analysis was used for the eight local physical instream variables (Table 2). Based on varimax-rotated loadings, factor 1 was associated with habitat stability (mean and standard deviation of bryophyte cover and particle size), factor 2 with velocity, factor 3 with depth and factor 4 with particle size diversity. Thus, seven weakly correlated (all r < 0·37) local ENV (shading, deciduous trees, stream width, factors 1, 2, 3 and 4) were subsequently used in the regression analyses, where variance inflation factors (VIF) for final models were also always low (VIF < 2·5). Three kinds of regressions were fitted. First, the significant ENV were selected with stepwise linear regression using Akaike Information Criterion (AIC) for model selection. Second, both the significant ENV and RSR were used simultaneously in explaining variation in LSR. Third, a regression with only RSR in explaining the variation in LSR was fitted. Stepwise regressions were based on Poisson error distribution, as is generally recommended for count-data (e.g. Crawley 2007).

Table 2.   Varimax-rotated loadings for in-stream variables in factorial analysis. Each variable was associated with the factor that gave it the highest loadings (in bold). SD = standard deviation
 FA1FA2FA3FA4
Depth mean−0·020·140·990·04
Depth SD0·250·230·71−0·18
Velocity mean0·080·910·300·06
Velocity SD0·460·810·11−0·04
Bryophyte cover mean0·840·170·13−0·08
Bryophyte cover SD0·680·060·030·25
Particle size0·720·450·10−0·17
Particle size diversity0·040·01−0·080·99

Variation partitioning (see Legendre & Legendre 1998) was conducted using only those ENV that had emerged as significant in the stepwise regression analysis described earlier. The first regression model (LSR ∼ ENV) reveals the variation explained by ENV (b), and the third model (LSR ∼ RSR) reveals the variation explained by RSR (c). The second model (LSR ∼ ENV + RSR) reveals the total variation explained (c). The pure environmental fraction (a) can be calculated by subtraction [(c) − (c)] and pure RSR fraction (c) by [(c) − (b)]. Shared faction (b) can be attained by [(b) + (c) − (c)]. The shared fraction indicates how much the two sets of variables (i.e. RSR and environment) are intercorrelated (Borcard, Gillet & Legendre 2011). Variation partitioning and significance of fractions (1999 permutations) were calculated using functions varpart and rda in the r-package vegan (version 1.17–4; Oksanen et al. 2010), respectively. Negative fractions were assumed to be zero (Borcard, Gillet & Legendre 2011). To reduce the dependence between LSR and RSR in multiple regression analysis and variation partitioning, RSR was calculated for each riffle site separately by pooling all the other sites in the given stream except the given riffle itself. Qualitatively, similar results of the effects of RSR on LSR were obtained when RSR was calculated based on all sites (results not shown). Adjusted coefficients of determination (adj. R2) were used, because they take into account the varying number of independent variables (Peres-Neto et al. 2006).

All analyses were conducted with both observed species richness values and Chao-estimated LSR and RSR (bias-corrected version; Chao 2005). Chao-estimator was selected, because it has been reported to be a stable estimate of LSR (Karlson, Cornell & Hughes 2004), it has been found adequate when compared to other estimation methods (e.g. Walther & Moore 2005) and it is commonly used in community ecology (Witman, Etter & Smith 2004; Cornell, Karlson & Hughes 2008; Soininen et al. 2009). In the present study, Chao-estimated species richness is considered as the upper bound of true species richness and observed species richness as the lower bound of true species richness. The primary purpose in using species richness estimator was to avoid underestimation of LSR, which may lead to pseudosaturation in the LSR–RSR approach (Cornell & Lawton 1992; Srivastava 1999).

All analyses were performed separately for each guild and for the whole-community data. All statistical analyses were conducted using r version 2.12.0 (R Development Core Team 2010). In addition to the base package of r, packages mass (multiple regression analysis; version 7.3–8; Venables & Ripley 2002), QuantPsyc (standardized estimates; version 1.3; Fletcher 2008), car (variance inflation factor; version 1.2–16; Fox & Weisberg 2011) and vegan (version 1.17–4; Oksanen et al. 2010) were used.

Results

We found a total of 167 taxa. Of these, we could assign 15 taxa to filterers, 52 to gatherers, 34 to predators, 36 to scrapers and 22 to shredders. At the riffle scale, the observed LSR for the whole community varied from 7 to 50 and Chao-estimated LSR from 7 to 81 (Table 1). Lowest mean observed LSR was 23 (Juhtipuro; Chao-estimated 27) and highest 41 (Uopajanpuro; Chao-estimated 52). The observed RSR (i.e. cumulative species richness of 10 riffles) for the whole community varied between 48 (P-Rytipuro) and 78 (Putaanoja) and Chao-estimated RSR between 51 (P-Rytipuro) and 109 (Metsosuonpuro).

Shape of the LSR–RSR relationship

The whole macroinvertebrate data and three guilds (i.e. predators, scrapers and shredders) showed linear relationships (Table 3; Fig. 1). The intercepts in the regressions of log(x + 1)-transformed data did not differ significantly from 0, and slopes in the log-transformed regressions did not differ from 1. Only gatherers showed curvilinear relationship, where the slope in the log-transformed regressions differed from 1 (two-tailed t-test, = 0·027). For filterers, both regressions were nonsignificant, and thus the shape of the relationship remained undetermined. In significant models using log-transformed data, coefficients of determination were moderate varying between 0·45 and 0·59, being highest for gatherers. To be able to compare the slopes in significantly linear cases, a regression line was fitted to untransformed data (Fig. 1). The slope was 0·41 for the whole community and between 0·39 and 0·70 for each of the guilds. When the regression procedure was repeated with Chao-estimated species richness, four relationships (whole community, filterers, scrapers and shredders) were undetermined (Table 3; Table S1 and Fig. S2, Supporting Information). Significant relationships were found only for gatherers (curvilinear) and predators (linear).

Table 3.   Results of regression analysis for log-transformed and log(x + 1)-transformed observed local (LSR) and regional (RSR) species richness and determination of linear (LIN) and curvilinear (CUR) relationships. If a model was nonsignificant, the relationship was left undetermined (UNDET). For comparison, the outcome for Chao-estimated species richness data is given (see Supporting Information). Significant models (P < 0·05) are in italics. For all models, d.f. are 1 and 8.
 ObservedChao
EstimateSEtPR2FT-test (H0: b = 1)LIN/CUR
tPLIN/CUR
Whole community
 log(LSR + 1) =b*log(RSR + 1)    0·538·86    
 a−0·061·18−0·050·963      
 b0·830·282·980·018      
 log(LSR) =b*log(RSR)    0·538·91−0·540·607LINUNDET
 a−0·141·20−0·110·912      
 b0·850·282·980·017      
Filterers
 log(LSR + 1) = b*log(RSR + 1)    0·252·70    
 a0·600·581·020·336      
 b0·470·291·640·139      
 log(LSR) =b*log(RSR)     0·232·38−1·460·183UNDETUNDET
 a0·340·620·550·597      
 b0·510·331·540·161      
Gatherers
 log(LSR + 1) = b*log(RSR + 1)    0·6011·93    
 a0·560·451·240·250      
 b0·520·153·450·009      
 log(LSR) = b*log(RSR)    0·5911·54−2·710·027CURCUR
 a0·340·480·690·509      
 b0·560·163·400·009      
Predators
 log(LSR + 1) =b*log(RSR + 1)    0·497·73    
 a−0·080·77−0·100·920      
 b0·760·272·780·024      
 log(LSR) = b*log(RSR)     0·507·95−0·640·537LINLIN
 a−0·330·80−0·410·695      
 b0·810·292·820·022      
Scrapers
 log(LSR + 1) = b*log(RSR + 1)    0·466·92    
 a0·320·680·470·649      
 b0·670·252·630·030      
 log(LSR) = b*log(RSR)     0·456·54−1·110·299LINUNDET
 a0·160·710·230·827      
 b0·700·272·560·034      
Shredders
 log(LSR + 1) = b*log(RSR + 1)    0·518·25    
 a−0·910·97−0·940·374      
 b1·130·402·870·021      
 log(LSR) = b*log(RSR)     0·518·260·520·617LINUNDET
 a−1·181·00−1·180·271      
 b1·220·422·870·021      
Figure 1.

 Average observed local species richness (LSR) vs. observed regional species richness (RSR) of stream macroinvertebrates. Axes are untransformed for clarity, although the determination between linear and curvilinear relationship was conducted using log-transformed data (see text and Table 3). Linear regression equations are based on untransformed data. The curvilinear relationship for gatherers is based on a power equation. The dashed line is the 1:1 line.

Relative importance of RSR and local environmental variables for LSR

In the stepwise multiple regression analyses, 30% of the variation in the observed LSR of the whole community was associated with ENV (Table 4). Based on standardized estimates, stream width had the largest relative importance in explaining variation in the observed LSR of the whole community. The whole-community model included variables that were part of the models of the guilds, resulting in the largest number of influential explanatory variables in the whole-community model. The sets of ENV that appeared important varied to some degree among the guilds. Velocity (FA2) appeared most often in the most parsimonious models, being absent only in the model for scrapers. However, based on standardized estimates, stream width had the largest relative importance in explaining variation in the observed LSR of predators, scrapers and shredders. Percentage of riparian deciduous trees and velocity were most important in explaining the variation in the observed LSR of gatherers and filterers, respectively. Generally, in all regression analyses, percentage of deciduous trees and velocity (FA2) had a negative relationship with LSR. Filterers were the only guild that showed a positive relationship with velocity (FA2). All individual regression models for observed species richness were highly significant (P < 0·012).

Table 4.   Results of multiple regression analysis for testing the relative importance of observed regional species richness (RSR) and environmental variables (ENV) for observed local species richness (LSR). In-stream variables were reduced to four factors using factorial analysis (see table 2). Variables associated with each factor were bryophyte cover and particle size (factor 1, FA1), velocity mean and standard deviation (FA2), depth mean and standard deviation (FA3) and particle size diversity (FA4). Shading did not enter any model and is thus omitted from the table. Standardized estimates were used
Data set/ModelRSRENVFSEAdj. R2d.f.1d.f.2P
DeciduousWidthFactorial Analysis
FA1FA2FA3FA4
Whole community
 LSR ∼ ENV −0·0040·0160·007−0·010 0·0089·46·960·30594<0·001
 LSR ∼ RSR + ENV0·008−0·0010·0120·006−0·008 0·0108·96·820·33693<0·001
 LSR ∼ RSR0·013      16·47·730·13198<0·001
Filterers
 LSR ∼ ENV    0·070−0·057 5·91·270·092970·004
 LSR ∼ RSR + ENV0·068   0·058−0·041 6·81·230·15396<0·001
 LSR ∼ RSR0·087      12·21·260·101980·001
Gatherers
 LSR ∼ ENV −0·028  −0·027  4·62·630·072970·012
 LSR ∼ RSR + ENV0·029−0·021  −0·023  4·82·580·103960·004
 LSR ∼ RSR0·039      8·52·620·071980·004
Predators
 LSR ∼ ENV −0·0220·055 −0·045 0·0449·32·790·25495<0·001
 LSR ∼ RSR + ENV0·057−0·0080·042 −0·048 0·04112·52·560·37594<0·001
 LSR ∼ RSR0·064      23·32·910·18198<0·001
Scrapers
 LSR ∼ ENV  0·0780·031  0·03422·72·070·40396<0·001
 LSR ∼ RSR + ENV0·059 0·0400·028  0·04922·41·950·46495<0·001
 LSR ∼ RSR0·074      38·62·270·28198<0·001
Shredders
 LSR ∼ ENV −0·0300·0770·038−0·059  8·51·900·23495<0·001
 LSR ∼ RSR + ENV0·062−0·0260·0610·018−0·033  9·91·800·31594<0·001
 LSR ∼ RSR0·091      27·41·930·21198<0·001

When similar regressions were conducted with Chao-estimated species richness, results regarding the specific ENV remained largely same as with observed species richness (Table S2, Supporting Information). However, for predators, the most important variable in explaining the variation in LSR was velocity (FA2). Depth (FA3) appeared more often, and bryophyte cover and particle size (FA1) less often in the models with Chao-estimated than with observed species richness. For Chao-estimated species richness, almost all models were also significant, excluding the models of RSR in explaining variation in the LSR of filterers and scrapers.

Variation partitioning results revealed relatively high pure environmental and shared fractions. When the impact of RSR was partitioned out, the ENV explained 19% of the variation in LSR of the whole community, and the shared fraction exceeded the importance of RSR (Fig. 2). The relative importance of the fractions differed highly between guilds. The total explained variation in observed LSR was highest for scrapers (46%) and lowest for gatherers (10%). Among guilds, the pure environmental fraction was most important for scrapers (19%) and almost as important for predators (18%). Pure RSR fraction was most important for predators (12%). Shared fraction was most important for scrapers (21%). For filterers and gatherers, the amounts of the three fractions were also almost equal, and the total variation explained was low. All pure environmental and RSR fractions were significant for observed species richness, excluding the nearly significant pure environmental fraction for gatherers (P = 0·064; Table 5).

Figure 2.

 The proportions of variation in observed local species richness partitioned into fractions explained purely by environmental (ENV) variables and purely by observed regional species richness (RSR). Also shared and unexplained fractions are shown.

Table 5.   The significance of pure fractions of observed regional species richness (RSR) and environment (ENV) in variation partitioning when explaining observed local species richness
 RSRENV
Whole community0·0260·001
Filterers0·0070·030
Gatherers0·0310·064
Predators0·001<0·001
Scrapers0·002<0·001
Shredders0·0020·002

When the variation partitioning procedure was repeated with Chao-estimated species richness, the total variation explained was lower in all cases compared to results with the observed species richness (Fig. S3, Supporting Information). The total explained variation was between 4% and 32%. The relative importance of environmental fraction increased for the whole community and for two guilds. For filterers and scrapers, all explained variation was associated with ENV when using Chao-estimated species richness. However, for shredders, the RSR fraction increased compared to the results with observed species richness. The increase in the amount of unexplained fraction diminished the significance of pure fractions, thus making RSR or environmental fraction nonsignificant in four cases (Table S3, Supporting Information).

Discussion

We found that both the species richness of the regional species pool (at the stream scale) and the local ENV (at the riffle scale) were important in determining local macroinvertebrate species richness. However, the relative importance of these factors differed between guilds, and different analytical approaches led to slightly different inferences. The classical LSR–RSR regression approach showed mainly linear relationships, suggesting that the overall species pool effect was important. By contrast, the variation partitioning results suggested a relatively larger role of local ENV. First, we will discuss the discrepancies between the results obtained with observed and Chao-estimated species richness.

Chao-estimator vs. observed species richness

We used Chao-estimator to avoid the underestimation of species richness, because the underestimation of LSR may lead to pseudosaturation in the LSR–RSR plots (Cornell & Lawton 1992; Srivastava 1999). In previous studies, Chao-estimator has sometimes affected the detection of linearity vs. curvilinearity (Soininen et al. 2009). In our study, when both observed and Chao-estimated data gave significant relationships, the conclusions between linear vs. curvilinear relationship did not differ. In variation partitioning, the relative importance of the fractions differed. The most distinct difference was that the results were clearly more inconsistent when using the Chao-estimated compared to the observed species richness data; in both types of analysis (i.e. the LSR–RSR approach and variation partitioning), some significant results changed to nonsignificant. This finding was somewhat surprising, given that estimated species richness is usually better predicted by environmental factors (Hortal, Garcia-Pereira & García-Barros 2004) or no difference has been found (Borges et al. 2009) compared to observed species richness.

Here, differences between the results obtained with the observed species richness and Chao-estimated data are likely to stem from the fact that Chao-estimator reflects strongly to the number of rare species. There are direct (Belmaker 2009) and indirect (White & Hurlbert 2010) observations that the number of rare species may be more limited by RSR, because of the higher dependence on random colonisations from the regional species pool (Ulrich & Ollik 2004; Ulrich & Zalewski 2006) and that the number of core species may be more limited by local environmental conditions. If rare and common species respond contrastingly to local and regional processes, the use of species richness estimators, which emphasize rare species, may generate unexpected variation and lead to unnatural results. Because of the reason mentioned earlier, and because the LSR–RSR plots for observed species richness were not more curvilinear than when using Chao-estimated data (and thus the underestimation of LSR when using observed species richness should not have affected the patterns profoundly), we will here emphasize the results based on observed species richness.

Relative roles of regional species pool and local environment

The relationships between observed local and RSR supported our first hypothesis that the relationship should be linear for the whole community. This finding was in line with previous studies (Griffiths 1999; Heino, Muotka & Paavola 2003; Cornell, Karlson & Hughes 2008), supporting a strong impact of regional species pool on LSR. By contrast, we did not find support for our second hypothesis that the LSR–RSR relationships should be more curvilinear within the guilds. Only one guild, gatherers, showed a curvilinear relationship according to the decision tree approach (Griffiths 1999), although curvilinearity was highly ambiguous based on visual interpretation.

Clear curvilinearity has traditionally been interpreted to imply an effect of interespecific interactions in limiting LSR (Cornell & Lawton 1992). It is possible that the riffle scale was too large for interspecific interactions to operate, as experimental studies on freshwater macroinvertebrate communities have found effective species interactions at very small spatial scales (Hemphill 1988; Kohler 1992; Hertonsson, Åbjörnsson & Brönmark 2008). Russell et al. (2006) studied mostly sessile rocky intertidal communities at a lot smaller local scale (0·25 m2; locality at their smallest scale) than in our study (c. 50 m2). They found that the saturation potential increased when using trophic groups that are more likely to interact compared with kingdom-level groupings. However, a local scale that small for stream macroinvertebrates would not have been reasonable given that lower abundances would possibly have profoundly limited the species richness in a sample, and for such mobile organisms, the scale of interactions in natural multispecies assemblages is likely to be wider than for sessile organisms. Perhaps a more likely explanation for the lack of saturation in our study system is the high seasonality of boreal headwater streams. If there is no time for competitive exclusion between frequent disturbances (e.g. freezing), then local communities would be structured mainly by successful colonization events. In this case, LSR should depend on both habitat characteristics and the size of the regional species pool.

We used variation partitioning to provide more detailed information on the importance of local habitat characteristics and to strengthen the interpretation of the LSR–RSR plots. Findings from these two approaches were, in fact, partly conflicting. Our third hypothesis was supported, because the environmental fraction was important. However, we expected that the relative importance of RSR would be higher compared to the importance of local environmental characteristics, because the LSR–RSR plots suggested a relatively large importance of regional species pool. Surprisingly, the importance of RSR was very low in some cases, and it never exceeded 50% of the total variation explained. One reason for the differing results may be that the avoidance of pseudoreplication by using mean LSR in the LSR–RSR approach simultaneously eliminates the within-region variability in LSR, which may lead to overestimation of the importance of RSR (White & Hurlbert 2010).

A more likely reason might be seen in the large shared fraction in variation partitioning that is translated into variation of RSR in the LSR–RSR approach. The shared fraction can be interpreted such that variation in the environmental conditions is associated with a certain region (e.g. basin characteristics). Shared fraction was especially large for scrapers. High topographical variability has emerged as important variable for species richness of scrapers (LeCraw & Mackereth 2010), and this characteristic varies considerably between our study streams (M. Grönroos and J. Heino, pers. obs.). Also, stream diatoms and terrestrial birds have shown linear LSR–RSR relationships, although variation partitioning has revealed large environmental and shared fractions (Passy 2009; White & Hurlbert 2010). By studying the regional environmental characteristics, Passy (2009) found that a large proportion of variance in LSR that was explained by RSR originated actually from the regional-scale environmental characteristics, explaining the large shared fraction.

Our hypothesis that the importance of regional species pool would be weaker for guilds than for the whole community was rejected, because the fraction explained by RSR was especially small for the whole community. One possible reason for this finding is that some regionally (across streams) varying (unmeasured) factor is correlated to the feeding circumstances, thus increasing the amount of variance explained by RSR in the analysis of each guild. For example, percentage of wetland in the catchment could indirectly influence the LSR of filterers (LeCraw & Mackereth 2010) or riparian tree species richness could influence the LSR of shredders (Muto, Kreutzweiser & Sibley 2011).

There were also considerable differences in the importance of pure RSR and pure environmental fractions between guilds, which supported our fourth hypothesis. The most distinct result was the high relative importance of RSR for predators compared with the other guilds. It seems that at the studied scale, the species richness of predators depends more on the supply of individuals from the regional species pool than the species richness of other feeding guilds. This finding might be explained by the generally lower population densities of predator species compared with the species of lower trophic levels (Odum 1971). Because a low population density increases extinction rates, predator species richness must rely more on dispersal from other nearby sites. However, more studies are needed to find out if this pattern persists in other systems and organism groups.

It can be expected that, as size of the region decreases in relation to the local scale, the amount of variation in LSR explained by RSR would increase. Using large regional scales for freshwater macroinvertebrates, the importance of RSR has sometimes been larger (Heino, Muotka & Paavola 2003) and sometimes smaller (Stendera & Johnson 2005) in relation to local ENV. Contrary to our expectation earlier, we found that even with relatively small regional scales, the importance of local environmental conditions is high and may exceed that of RSR in explaining LSR. Species sorting (Leibold et al. 2004) could be the primary mechanism behind this pattern, as has also been suggested for the community structure of headwater stream macroinvertebrates by recent studies (Brown & Swan 2010; Heino et al. in press).

Conclusions

We emphasize the results of variation partitioning and conclude that both the species pool effect and the local habitat characteristics are important in determining LSR, with the latter being slightly more important. The relative importance of local and regional processes may vary depending, as shown in our study, on feeding guild and trophic level. The novel findings of our study relate to the intermediate regional scales used to determine the relative importance of regional species pool and local environmental conditions to LSR: even within such small regions (i) regional species pool effects may be important, (ii) local environmental variation may be profound and (iii) local environmental factors mainly structure biotic communities. These findings add to the growing body of literature that species richness at a local scale is determined jointly by local and regional processes (Harrison & Cornell 2008; White & Hurlbert 2010; Kristiansen et al. 2011).

Acknowledgements

We thank Allen Hurlbert and one anonymous referee for constructive comments on the manuscript. An earlier draft of the paper was improved by comments from Mari Annala, Kaisa-Leena Huttunen, Musawenkosi Mlambo, Timo Muotka and Mikko Tolkkinen. We also thank Jari Oksanen for advising with statistical analyses, and Jari Ilmonen, Tommi Karhu, Maija Niva and Lauri Paasivirta for identifying the macroinvertebrate samples. Tommi Karhu and Päivi Rusi assisted in the field, and Oulanka Research Station offered logistical support during the field work. This paper is part of the project ‘Spatial scaling, metacommunity structure and patterns in stream communities’ funded by the Academy of Finland.

Ancillary