Determinants of macroinvertebrate diversity in headwater streams: regional and local influences

Authors


Jani Heino, Department of Biological and Environmental Sciences, University of Jyväskylä, POB 35, 40351 Jyväskylä, Finland. E-mail: jheino@cc.jyu.fi

Summary

  • 1Multiscale determinants of diversity and the relationship between regional (RSR) and local richness (LSR) have recently attracted increased attention, yet such studies on stream organisms remain scarce. We studied the relationships among RSR, β-diversity, LSR and local environmental variables in 120 headwater streams in Finland. Approximately similar-sized areas of eight drainage systems were defined as regions, and 15 stream riffles (= locality) per region were sampled.
  • 2RSR showed a strong positive relationship with mean LSR (R2 = 0·686), and there was no sign of curvilinearity within the observed range of RSR. RSR was also positively, although non-significantly, related to β-diversity (r = 0·662).
  • 3In stepwise regression, RSR was the first variable to enter the model, and a model incorporating RSR and stream width explained 32·5% of variation in LSR. If RSR was omitted from the model, then stream width emerged as the most important variable, followed by water pH, which together accounted for 20·6% of variation in LSR.
  • 4At the within-region scale, different variables were important in accounting for variation in LSR. Factors correlated with LSR reflected either stream size, spatial heterogeneity, adverse water chemistry conditions (pH), or a limiting resource base for macroinvertebrates (nutrient concentrations).
  • 5The major role of RSR in setting the upper limit to LSR suggests that macroinvertebrate assemblages of boreal headwater streams are unsaturated. This finding is supported by evidence from experimental studies, where it has been shown that competitive interactions among stream macroinvertebrates are effective only at very small spatial scales, and competitive exclusion is prevented typically by frequent disturbances. However, although RSR was generally the most influential variable contributing to LSR, it is far from clear whether RSR consistently sets the limits to LSR, or vice versa. For instance, uniformly adverse water chemistry conditions across a region may lead to low levels of local richness and low species turnover among sites, leading eventually to an impoverished regional species pool.
  • 6These findings do not deny the importance of local factors, but emphasize that understanding the organization of stream benthic communities requires simultaneous examination of factors prevailing at multiple spatial scales.

Introduction

An emerging paradigm in community ecology emphasizes the role of regional factors and history in regulating the organization of biotic communities (Menge & Olson 1990; Ricklefs & Schluter 1993). Empirical support for strong regional effects has been obtained from a number of studies reporting a linear relationship between regional species richness (RSR) and local species richness (LSR) for various taxonomic groups (e.g. Cornell & Karlson 1996; Caley & Schluter 1997; Griffiths 1997; Shurin et al. 2000). Theoretically, if LSR increases linearly with RSR, local communities are considered unsaturated and mainly free from local control. By contrast, if LSR reaches an asymptote at high levels of RSR, then communities are regarded as saturated and controlled by interspecific competition and other local processes (Cornell & Lawton 1992; Cornell 1993). As a tentative generalization emerging from studies conducted thus far, local communities seem rarely saturated, biotic interactions are relatively weak and regional processes have a large impact on setting the upper limits to local species richness.

The degree of interdependence among RSR (γ-diversity) and LSR (α-diversity) is connected closely with patterns of species turnover among sites, i.e. β-diversity (Ricklefs & Schluter 1993; Srivastava 1999). Regional β-diversity is determined by (i) variation in environmental characteristics among local habitats, and (ii) the degree of habitat specialization of the biota. If local habitat conditions determine local diversity, and there is a high degree of species turnover among sites, then it is in fact RSR that mirrors variation in LSR across a region. Thus, regional differences in diversity may stem from smaller-scale processes, rather than from among-region differences in dispersal and taxa diversification (Ricklefs & Schluter 1993). Clearly, this generates a chicken-and-egg problem (Cornell & Lawton 1992; Huston 1999): is it really RSR that determines LSR, or vice versa? Understanding the determinants of diversity across multiple scales thus necessitates a simultaneous examination of β-diversity and the RSR–LSR relationship.

Stream communities are structured by processes prevailing at multiple spatial scales (Minshall 1988; Poff 1997), yet surprisingly few studies have addressed regional–local species richness relationships among running water biota. Studies on fish communities suggest either strong regional control of local diversity (Hugueny & Paugy 1995; Oberdorff et al. 1998) or the joint influence of regional and local factors (Angermeier & Winston 1998) in determining the number of locally co-occurring species. Similar studies on stream invertebrates are virtually lacking, with the exception of a study reporting a linear relationship among RSR and LSR for river-dwelling mussels (Vaughn 1997). As stream invertebrates experience frequent and unpredictable disturbances and show a high capability for dispersal, one may predict that their species richness should be primarily under regional control (Palmer, Allan & Butman 1996). Alternatively, because even neighbouring streams may differ widely in environmental conditions, local factors may modify macroinvertebrate diversity considerably (Townsend, Hildrew & Francis 1983; Malmqvist & Mäki 1994; Paavola, Muotka & Tikkanen 2000).

In broad-scale surveys, the number of macroinvertebrate taxa has been observed to increase with stream size, substratum heterogeneity and the amount of macrophytes, suggesting a trend of increasing richness with higher environmental heterogeneity (e.g. Vinson & Hawkins 1998). By contrast, low water quality (e.g. low pH) has been associated with low-diversity macroinvertebrate communities (Townsend et al. 1983; Hildrew & Giller 1994). However, the relative importance of local physicochemical factors and the regional species pool as determinants of macroinvertebrate richness in streams remains largely unstudied (see Vinson & Hawkins 1998).

In boreal areas, stream macroinvertebrate assemblage structure varies considerably at both regional and local scales (Malmqvist & Hoffsten 2000; Sandin & Johnson 2000; Heino et al. 2002), making them amenable systems for assessing large-scale determinants of benthic biodiversity. In this study, we analysed data from 120 streams in eight regions (i.e. drainage systems) in Finland. We specifically examined (1) the relationship between RSR and LSR, (2) the relationship between RSR and β-diversity and (3) the relative contribution of RSR and local environmental factors to LSR. Furthermore, at the within-region scale, we examined (4) which environmental factors are correlated most strongly with variation in LSR.

Materials and methods

stream surveys

The majority of stream surveys were conducted in 1998, but additional data from 1992, 1994, 1997 and 2000 were included if field sampling and laboratory methods were identical to those used in 1998. All material was collected, processed and analysed by the same personnel. We limited our consideration to near-pristine streams with base flow < 0·6 m3 s−1 and catchment area < 60 km2, to delineate our analysis to a single habitat type, i.e. headwater streams. Therefore, we excluded spring-fed streams, lake outlets and streams disturbed by recent human activities. Otherwise, streams for each region were selected randomly, with the restriction that they had to be within 2 kilometres from the nearest road. We sampled 15 headwater streams in each of eight regions (Fig. 1).

Figure 1.

Geographical locations of the study regions. River systems were as follows: upper Kymijoki (A), Kyrönjoki (B), Kiiminkijoki (C), upper Oulujoki (D), Koutajoki (E), Muonionjoki (F), Kemijoki (G) and Tenojoki (H).

We measured several riparian and in-stream variables at each site. The integrity (% riparian zone without obvious human impact) and tree species composition of the riparian zone were assessed in a 50-m section along both banks directly upstream of the sampling site. Shading by overhanging vegetation was measured at 20 locations in evenly spaced cross-channel transects. Depth and current velocity (at 0·4 × depth) were measured at 40 random locations in cross-channel transects. Moss cover and substratum particle size were assessed in 10 50 cm × 50 cm quadrats placed randomly in each riffle. We used the following classification of particle sizes (modified Wentworth scale): (0) organic matter; (1) sand (diameter 0·25 mm–2 mm); (2) fine gravel (2 mm–6 mm); (3) coarse gravel (6 mm–16 mm); (4) small pebble (16 mm–32 mm); (5) large pebble (32 mm–64 mm); (6) small cobble (62 mm–128 mm); (7) large cobble (128 mm–256 mm); (8) small boulder (256 mm–400 mm); and (9) large boulder and bedrock (> 400 mm). The proportion of each size class was estimated for each quadrat, and these estimates were subsequently averaged to give the mean substratum particle size for a site. Mean stream width was also measured at each sampling site. Water samples were collected simultaneously with benthic sampling, and they were subsequently analysed for pH, alkalinity, conductivity, total nitrogen [TN], total phosphorus [TP], colour and iron [Fe] following Finnish national standards. Physical and chemical conditions of the study streams by region are summarized in Table 1.

Table 1.  Regional richness (RSR), β-diversity (calculated according to Harrison et al. 1992) and means and ranges of local richness (LSR) and major environmental variables for each region. For the measurement of substratum particle size, see text
VariableRegion
ABCDEFGH
RSR9253774396738162
β29·7110·889·526·8212·459·689·878·13
LSR29·915·919·516·323·622·821·620·5
(22–39)(5–21)(9–33)(7–22)(11–35)(12–31)(12–34)(11–29)
Deciduous trees (%)6642533269687798
(10–100)(20–75)(20–85)(5–75)(30–95)(30–95)(50–100)(90–100)
Current velocity (cm/s)3635392948412838
(28–49)(20–67)(20–97)(20–45)(23–71)(21–63)(17–57)(25–53)
Stream width (m)3·82·02·72·83·22·63·43·6
(1·5–8·4)(0·8–4·5)(0·5–7·0)(0·7–5·2)(1·0–6·5)(0·6–6·0)(0·9–9·0)(0·8–8·5)
Moss cover (%)3310214223503322
(0–37)(3–79)(0–57)(0–93)(0–67)(1–83)(1–86)(0–86)
Particle size5·94·65·87·46·24·85·56·8
(3·5–7·0)(0·8–8·8)(0–8·2)(5·0–9·0)(4·7–7·1)(1·5–6·9)(2·5–8·3)(4·3–7·5)
pH6·55·85·65·57·86·87·37·3
(5·7–7·0)(4·7–6·6)(4·6–6·6)(4·8–6·2)(7·2–8·4)(6·3–7·3)(6·3–7·9)(6·6–7·5)
Total N (µg/L)4331097593257343165281114
(280–710)(298–3700)(400–1100)(180–340)(160–487)(63–320)(123–710)(43–270)
Colour (mg Pt/L)9527225516091597213
(40–200)(100–600)(140–400)(100–200)(20–286)(10–150)(13–150)(5–35)

All invertebrate samples were collected in late autumn (early September–late October). At each site, we took a 2-min kick-net sample (net mesh size 0·3 mm), aiming to cover most benthic microhabitats in a riffle section of approximately 100 m2. Invertebrates and associated material were preserved immediately in 70% alcohol, and they were subsequently sorted and identified in the laboratory. Invertebrates were mainly identified to species or genus level. Chironomids and simuliids were not identified beyond the family level, and they were thus omitted from all analyses.

regional delineations and richness estimates

Defining ‘region’ and ‘locality’ is critical for studies of regional vs. local determinants of species richness. An ideal region should be environmentally homogeneous, it should have ecologically meaningful boundaries, and all localities within it should be easily accessible to all species in the regional pool (Cornell & Karlson 1996; Angermeier & Winston 1998). Furthermore, instead of spanning multiple habitat types, the analysis should focus on comparing a specific habitat type across regions (Srivastava 1999). Streams are organized as natural spatio-temporal hierarchies, where drainage network forms the highest level. Therefore, drainage systems represent a natural, objectively defined region for examining the regional–local richness relationship in stream biota (see also Vaughn 1997; Angermeier & Winston 1998). A locality, in turn, should be an environmentally homogeneous spatial unit within which organisms could encounter each other during ecological time, thereby forming a potentially interactive community (Cornell & Karlson 1996; Srivastava 1999). For most stream organisms, an individual riffle site clearly qualifies for an appropriately scaled sampling unit.

In our study, the streams (n = 15) within each region belong to the same drainage system. The regions comprised the following drainage systems (Fig. 1): Kymijoki (A), Kyrönjoki (B), Kiiminkijoki (C), upper Oulujoki (D), Koutajoki (E), Muonionjoki (F), Kemijoki (G) and Tenojoki (H).

Variation in region size may modify the shape of the regional vs. local richness plot, by increasing the probability of detecting a curvilinear relationship (Caley & Schluter 1997; Srivastava 1999). We estimated the spatial extent of our regions using a simple rectangle method (Shurin et al. 2000). Thus, we drew a rectangle connecting the southern- and northernmost sites, and the western- and easternmost sites for each region. The size of the rectangle was then measured with a ruler and converted to square kilometres. The region size varied between 756 km2 (region E) and 7742 km2 (region G). However, because region size and RSR were non-significantly correlated (P = 0·216) and because spatial variation among our regions was low (all < 8000 km2), we did not attempt to correct for variation in regional extent.

We defined local richness (LSR) as the number of taxa found at a single stream riffle, and regional richness (RSR) as the cumulative number of taxa across all sampling locations within a drainage system. This technique introduces an element of interdependence in our data, which could lead to spurious correlations (Cresswell, Vidal-Martinez & Crichton 1995; Zobel 1997). A potential way to solve this problem is to estimate regional richness from distribution maps or regional checklists (Srivastava 1999). In our case, however, this was inconceivable, because comprehensive species lists of stream macroinvertebrates are lacking in our study area. Nevertheless, this would constitute only a partial solution to the problem: even if such lists were available, regional and local richness estimates would still be spatially autocorrelated. Furthermore, the reliability of data compiled from various secondary sources is questionable (Gaston & Blackburn 1999). For example, incomplete knowledge of taxonomy, inadequate sampling and inconsistent sampling effort across regions and localities may introduce unknown bias to richness estimates. However, because we used strictly standardized sampling methods and the same sampling effort (15 streams/region) throughout the study, our estimates of regional and local richness should not be biased. It is still possible that 15 streams might constitute an inadequate sample size for estimating regional richness. However, our preliminary data (n = 55 streams) from the most taxon-rich region (E) suggest that 15 streams sample approximately 80% of taxa present in headwater streams within a region. Therefore, we consider our sampling effort sufficient for a reliable estimate of the size of the regional pool.

statistical methods

We used the decision tree approach of Griffiths (1999) to determine whether the relationship between LSR and RSR was linear (implying proportional sampling from the regional pool) or curvilinear (implying local saturation). Thus, we first fitted a second-order polynomial regression to our data, and if the second order term was non-significant, we used linear regression to test for a linear relationship among RSR and LSR. Then, we tested for deviation of the intercept a from zero in an unconstrained model; if this was insignificant, we used unconstrained regression to test for the fit of the linear model, i.e. significance of the regression coefficient b. Many previous workers have used constrained regression (forced through the origin) based on the premise that ‘when regional diversity is zero, so too is local diversity’ (Caley & Schluter 1997). This may be inadvisable, however, because it involves extrapolation beyond the range of the data, thereby inflating R2 values (Griffiths 1999; Srivastava 1999).

We ran separate regression analyses on four data sets, testing for two slightly different questions: (i) the influence of regional richness on local richness (RSR vs. mean LSR; RSR vs. maximum LSR; RSR vs. minimum LSR, n = 8 in each analysis), and (ii) the influence of region on local richness (RSR vs. LSR, n = 120) (see Srivastava 1999). The latter approach was dismissed by Srivastava (1999) because it constitutes pseudoreplication and thus confounds site and regional effects on LSR. However, as pointed out by Karlson & Cornell (2002), use of multiple estimates of LSR within a region is necessary if the relative contribution of RSR and local factors to LSR is being assessed. For this purpose, we used stepwise linear regression, where the variable that first enters the model explains most of the variation in LSR, the second one most of the remaining variation, and so on. Data on environmental variables were first screened for intervariable correlations and a set of eight variables representing the physicochemical habitat templet, and which were not strongly intercorrelated, were used as proxies for local environmental conditions in subsequent analyses. The variables included were: stream width, moss cover, particle size, current velocity, proportion of deciduous trees in the riparian zone, pH, total nitrogen and water colour. If necessary, variables were transformed (log or arcsine-square root) to improve normality and stabilize variances. We also ran another stepwise regression, this time without RSR, to reveal whether this variable masked the influence of any of the environmental variables. Finally, we used stepwise regressions at the within-region level to examine whether different local factors were influential in explaining variation in taxon richness in different regions.

We assessed the relationship between RSR and turnover diversity using Pearson correlations. We calculated a beta-diversity index for each region using the formula of Harrison, Ross & Lawton (1992):

β2 = (S/αmax − 1)/(N − 1) × 100,

where S is the total number of taxa in a region, αmax is the maximum number of taxa recorded at a single site, and N is the number of sites. Thus, β2 ranges from 0 to 100, and measures the degree by which regional richness exceeds the maximum local richness. However, β2 is directly dependent on, and can thus generate spurious correlations with, RSR. Therefore, we calculated the average among-site dissimilarity in taxon composition for each region to obtain an index of turnover that was independent of RSR. We used Sørensen's coefficient for calculating pairwise dissimilarities among all sites, using the formula:

βSor = 1 − 2W/(A + B),

where W is the sum of shared taxon occurrences and A and B are the sums of taxon occurrences in individual sample units. Each pairwise dissimilarity and the average among-site dissimilarity for each region thus range from 0 to 1, high values indicating high turnover among sites. All analyses were performed using SPSS (SPSS Inc. 1999), except dissimilarities which were calculated using PC-Ord (McCune & Mefford 1999).

Results

local vs. regional richness

RSR ranged from 43 (region D) to 96 (region E), and mean LSR varied from 15·9 (region B) to 29·9 (region A) (Table 1). RSR showed a positive linear relationship with mean LSR (Fig. 2a). The quadratic term in RSR–LSR regression was insignificant. Furthermore, the intercept a of the linear model did not differ significantly from zero and the slope coefficient b was significant (Table 2). Thus, within the observed range of RSR, there was no ceiling to LSR, suggesting lack of local saturation. RSR was strongly related to maximum LSR (Fig. 2b), whereas RSR showed insignificant, positive relation to minimum LSR (Fig. 2c).

Figure 2.

Relationships between four measures of local (LSR) and regional richness (RSR) of stream macroinvertebrates. LSR is represented by the mean (a), maximum (b) and minimum (c) values for each region, as well as by including all estimates of local richness for a region (d). Solid line represents the associated regression equation, whereas dashed line indicates the theoretical upper limit for LSR.

Table 2.  Summary of the coefficients of the linear and second-order regression models for the relationship among mean LSR and RSR
 CoefficientSEtP
Y = a + bx
 Constant 6·805 4·1031·6590·148
 RSR 0·200 0·0553·6230·011
Y = a + bx + bx2
 Constant11·26718·0540·6240·560
 RSR 0·064 0·5370·1200·909
 RSR2 0·001 0·0040·2550·809

factors explaining variation in lsr: across-regions analysis

In stepwise regression analysis, RSR was the first variable to enter the model, accounting for 24·8% of variation in LSR (Fig. 2d). The final model, explaining 32·5% of variation in LSR, incorporated RSR and stream width (Table 3). In regression analysis omitting RSR, stream width emerged as the most important local variable, followed by water pH. This model, however, accounted for only 20·6% of variation in LSR (Table 4).

Table 3.  Results of stepwise multiple regression analysis of the relationships between local richness, regional richness, and local environmental variables. Cumulative R2 and F include the contribution of the variable named at each step and the one preceding it. The overall model was significant at P < 0·001
Source of variationd.f.SSMSFR2
Model  21872·550936·275  
Regional richness  11430·275 38·9310·248
Stream width  1 442·275 28·1390·325
Residual1173892·917 33·273  
Total1195765·467   
Table 4.  Results of stepwise multiple regression analysis of the relationships between local richness and local environmental variables. Regional richness was omitted from this analysis. Cumulative R2 and F include the contribution of the variable named at each step and the one preceding it. The model was significant at P < 0·001
Source of variationd.f.SSMSFR2
Model  21188·492594·246  
Stream width  1 713·689 16·6700·124
pH  1 474·803 15·1910·206
Residual1174576·975 39·119  
Total1195765·467   

factors explaining variation in lsr: within-region analyses

Different factors appeared important in explaining variation in LSR in different regions. In regions A, B and G, no variable was related significantly to LSR, due probably to the small number of sites. In-stream variables were the best correlates of LSR in regions C (moss cover) and D (current velocity). Percent deciduous trees in the riparian zone accounted for 42% of variation in LSR in region E, and stream width for 69% in region F. In the northernmost region (H), LSR was strongly positively related to potential stream productivity, i.e. [TN] (Table 5).

Table 5.  Final models of stepwise multiple regression on the relationships between taxon richness and environmental variables at the within-region level. No variables were significantly related to taxon richness in regions A, B and G
RegionIndependentsR2FP
CMoss cover†0·267 4·738  0·049
DCurrent velocity†0·407 8·928  0·010
EDeciduous trees†0·418 9·331  0·009
FStream width†0·69129·040< 0·001
HTotal N†0·63422·508< 0·001

relationship between beta diversity and rsr

Beta diversity varied relatively little among the regions, but was overall positively, although non-significantly, correlated with RSR, for both β2 (r = 0·662, P = 0·074) and βSor (r = 0·494, P = 0·213) (Fig. 3a,b). These two measures of turnover diversity were not strongly correlated (r = 0·411, P = 0·311), implying that they emphasized slightly different aspects of taxon turnover. Using both β2 and βSor, turnover diversity was highest in region E, whereas the identity of the region with the lowest turnover diversity varied with the index used. Turnover diversity was lowest in region D when measured by β2, whereas βSor recorded lowest turnover for regions A, B, F and H.

Figure 3.

Relationship between β-diversity, measured as Harrison et al.'s (1992) beta-2 (a), or average Sorensen's dissimilarity beta-Sor (b), and regional richness (RSR) of stream macroinvertebrates.

Discussion

Circumstantial evidence suggesting that regional processes exert strong control over the number of locally co-occurring species has been provided by numerous studies reporting a linear relationship among LSR and RSR for various groups of organisms (e.g. Cornell 1999; Shurin et al. 2000; Stevens & Willig 2002). However, some studies on highly interactive fish communities refer to the predominance of local processes in determining local richness (Jackson & Harvey 1989; Tonn et al. 1990), although RSR–LSR relationships suggesting the primacy of regional control of local diversity have also been reported for both lentic (Griffiths 1997) and lotic fish (Hugueny & Paugy 1995; Oberdorff et al. 1998). In our study, stream macroinvertebrate richness showed a linear relationship to RSR, suggesting that regional-scale processes set the upper limit to LSR. For stream invertebrates, such a relationship between RSR and LSR should not be surprising, because they live in a frequently and unpredictably disturbed environment and exhibit high rates of dispersal, both of which should enhance the regional control of local communities (Palmer et al. 1996). Streams are considered flashy environments where flow-related disturbances move substratum particles and thereby regulate the structure of biotic communities strongly (Resh et al. 1988; Lake 2000). Furthermore, many stream-dwelling invertebrate taxa possess considerable dispersal capacity, by either flying or drifting, facilitating rapid colonization of denuded stream areas after disturbance (e.g. Giller, Sangpradub & Twomey 1991; Malmqvist et al. 1991). Although strong competitive interactions do occur among stream macroinvertebrates, these are typically effective only at very small spatial scales (up to a few centimetres) and spatial exclusion, even in the presence of a strict competitive hierarchy, is prevented by flow-related disturbances (e.g. Hemphill & Cooper 1983; McAuliffe 1984; Kohler 1992). There is thus little indication of competition leading to species exclusion at a scale pertinent to local macroinvertebrate assemblages (for a possible exception, see Kohler & Wiley 1997). Furthermore, streams are extremely heterogeneous environments and organisms typically exhibit strongly aggregated distributions across this spatially variable landscape. Although not directly shown for any stream organism, this heterogeneity may contribute to species coexistence by allowing inferior competitors to persist in the community through ‘probability refuges’, i.e. patches left unoccupied by superior competitors (the aggregation model of coexistence, Atkinson & Shorrocks 1984; see also Murphy, Giller & Horan 1998). All these factors suggest that stream macroinvertebrate communities rarely attain saturation, even in high-diversity regions. Lack of a richness ceiling in our data, however, may also reflect the fact that benthic communities in boreal streams are relatively impoverished, and curvilinearity would be observed only by including more species-rich regions. Finally, it is also possible that some rarely disturbed and productive lotic habitats (e.g. lake outlets, Malmqvist & Eriksson 1995) might support interactive, locally saturated invertebrate communities.

Including LSR estimates drawn from different habitat types (e.g. headwater streams vs. large river sites) in the same plot could create false asymptote (‘pseudosaturation’, e.g. Griffiths 1997), and it would also invalidate the logic of RSR–LSR comparisons (Srivastava 1999). Because several macroinvertebrate taxa are largely restricted to either headwater streams or large rivers (Malmqvist & Mäki 1994; Furse 2000), species pools for different habitat types incorporate organisms that are never expected to occur or interact at the same local arena. Nevertheless, even if sampling were stratified strictly by habitat type, curvilinear RSR–LSR relationships may emerge if beta-diversity is related positively to RSR, i.e. assemblage turnover is higher in high-diversity than in low-diversity regions (Griffiths 1997). The few studies that have tested this assumption directly have concluded that turnover diversity contributes relatively little to regional richness (Harrison et al. 1992; Griffiths 1997; but see Stevens & Willig 2002). For lotic macroinvertebrates, within-region variation in assemblage composition often follows a nested pattern, i.e. low-diversity sites support a subset of taxa from more diverse sites (e.g. Malmqvist & Hoffsten 2000; J. Heino et al. unpublished), suggesting that turnover diversity in species-rich regions is probably not high enough to render the RSR–LSR relationship curvilinear.

Despite strong regional influences, environmental filters at the within-region scale eventually determine how large a portion of the regional species pool prevails in each local community (Tonn 1990; Poff 1997). In our study, LSR showed wide among-stream variation within each region, suggesting that even in diverse regions some localities provide conditions that limit the kind (and, ultimately, number) of taxa able to persist under such circumstances. This finding concurs with former studies showing that streams in close proximity to each other, but with contrasting water chemistry, may harbour markedly differing macroinvertebrate assemblages (Townsend et al. 1983; Paavola et al. 2000). Nevertheless, our results suggest that stream environmental factors are secondary to RSR in accounting for variation in LSR of stream macroinvertebrate assemblages.

Among-region differences in LSR and RSR may also arise through regional differences in stream environmental conditions. For instance, streams in regions B and D mostly drain peatland landscapes, being acidic and brown-coloured (Heino et al. 2002). Such uniformly adverse environmental conditions across a region may eventually affect the size of the regional species pool. Thus, the control of macroinvertebrate richness is not necessarily unidirectional, i.e. from regional to local, but stream environmental factors may also have a feedback effect on RSR (see also Vinson & Hawkins 1998).

Environmental factors related to variation in macroinvertebrate richness varied among the regions. Factors correlated with richness reflected either spatial heterogeneity, adverse environmental conditions or a limiting resource base. Macroinvertebrate richness showed a generally positive relationship to stream width (area effect) and moss cover (heterogeneity effect), both often cited as important correlates of lotic macroinvertebrate diversity (Brönmark et al. 1984; Malmqvist & Mäki 1994; Vinson & Hawkins 1998). Conversely, low pH and high humic content of stream water are known to have strong negative effects on macroinvertebrate diversity (Otto & Svensson 1983; Townsend et al. 1983; Hildrew & Giller 1994). An interesting finding was the positive relationship of nitrogen content with macroinvertebrate richness in the northernmost study region H (70°N). Nutrient concentrations in these subarctic streams ranged from very low (P < 2 µg/l, N < 50 µg/L) to moderate (P = 5 µg/l, N = 270 µg/L), probably limiting ecosystem productivity and biotic communities. Even slight increases in nutrient concentrations could increase algal and macroinvertebrate productivity (e.g. Peterson et al. 1993), with potential ‘bottom-up’ effects on macroinvertebrate diversity. By contrast, no significant relationships between stream water nutrients and taxa richness were detected in regions further south, implying that nutrients did not generally limit stream biodiversity across the study area.

Because RSR appeared as the primary factor related to variation in LSR, future studies should also address the determinants of RSR for stream macroinvertebrates. Latitudinal gradients in climate, environmental productivity and glaciation history have often been proposed as potential explanations to large-scale variation in species richness (Currie 1991; Ricklefs & Schluter 1993; Huston 1994). However, studies on stream organisms have produced somewhat equivocal conclusions regarding regional differences (e.g. latitudinal gradients) in species richness (Patrick 1975; Hildrew & Giller 1994; Jacobsen, Schultz & Encalada 1997). Regional differences in stream macroinvertebrate diversity do exist, however, but they do not necessarily follow any consistent latitudinal trends. For instance, the regional diversity of stoneflies increases with increasing latitude, whereas that of caddisflies exhibits an opposite pattern, and a parallel trend is also seen weakly at the local scale of stream riffles in our study area (J. Heino et al. unpublished). The mixing of taxa with differing biogeographical patterns may be a reason why latitudinal diversity gradients for stream macroinvertebrates are somewhat obscured, although regional patterns at the among-river system scale are obvious. In addition to historical factors and climate, possible correlates for variation in RSR include regional differences in landscape heterogeneity (e.g. altitude, geology and vegetation) which could in turn affect stream riffle area and the variety of stream types available within a region. These factors might affect regional-scale extinction probabilities, metacommunity dynamics and the degree of species turnover among sites (Cornell & Lawton 1992). Within-region habitat variability is an important component of turnover diversity (e.g. Harrison et al. 1992), and although we did not find significant relationship among RSR and β-diversity, differences in within-region environmental heterogeneity still remain a potential determinant of variation in RSR.

To conclude, our results suggest that the upper limit of local macroinvertebrate diversity is controlled largely by regional factors. This finding, however, does not decline the importance of local and basin-scale factors, which produce considerable variation in LSR within regions, and may even have feedback effects on RSR. A potentially fruitful avenue for future studies is to focus on the balance between α- and β-diversity (see Loreau 2000). Because dispersal is clearly a key process determining shifts in the relationship between α and β components of regional diversity (Loreau & Mounquet 1999), a comparison between organisms with widely differing dispersal abilities (e.g. mayflies vs. blackflies; stream insects vs. bryophytes) might prove especially rewarding. That regional and local diversities are linked strongly in stream communities (Vaughn 1997; Oberdorff et al. 1998) clearly indicates that stream ecologists should abandon their traditional adherence to local, in-stream processes as the sole or even primary regulators of local diversity. Studies on the determinants of stream biodiversity are likely to be ineffective unless regional aspects, especially the size and composition of the regional species pool, are considered explicitly. Such studies are indeed needed urgently, because effective conservation of stream biodiversity requires a proper understanding of the determinants of diversity across multiple scales and taxonomic groups.

Acknowledgements

We thank the environmental centres of Central Finland, Lapland, Northern Ostrobothnia and West Finland, and Oulanka Biological Station for analysing the water samples. R. Salvain provided guidance and inspiration during data collection. We also thank B. Malmqvist and an anonymous referee for constructive comments on the manuscript. This paper is part of the Finnish Biodiversity Programme (FIBRE), funded by the Academy of Finland. The study was also supported by a grant from the Maj and Tor Nessling Foundation.

Ancillary