A truce with neutral theory: local deterministic factors, species traits and dispersal limitation together determine patterns of diversity in stream invertebrates

Summary

• 1
  Studies seeking to explain local patterns of diversity have typically relied on niche explanations, reflected in correlations with local environmental conditions, or neutral theory, invoking dispersal processes and speciation.
• 2
  We used macroinvertebrate community data from 10 streams that varied independently in local ecological conditions and spatial proximity. Neutral theory predicts that similarity in communities will be negatively associated with distance between sites, while niche theory suggests that community similarity will be positively associated with similarity in local ecological conditions.
• 3
  Similarity in total invertebrate, grazer and predator assemblages showed negative relationships with distance and, for grazers and predators, positive relationships with local ecological conditions. However, the best model predicting community similarity in all three cases included aspects of both local ecological conditions and distance between sites.
• 4
  When assemblages were analysed according to dispersal ability, high-dispersal species were shown to be freely accessing all sites and community similarity was not well predicted by either local ecology or spatial separation. Assemblages of species with low and moderate dispersal ability were best predicted by combined models, including distance between sites and local ecological factors.
• 5
  The results suggest that the perceived dichotomy between neutral and local environmental processes in determining local patterns of diversity may not be useful. Neutral and niche processes structured these communities differentially depending on trophic level and species traits.
• 6
  We emphasize the potential for both dispersal processes and local environmental conditions to explain local patterns of diversity.

Introduction

Stephen Hubbell's unified neutral theory of biodiversity (Hubbell 2001) provoked intense debate among ecologists (e.g. McGill 2003; Volkov et al. 2003; Gilbert & Lechowicz 2004; Wootton 2005). Hubbell's theory considers communities to consist of ecologically equivalent individuals distributed across a fixed number of species derived from a regional species pool. Population dynamics is modelled at regional and local scales. Replacement individuals at a local scale immigrate from the regional species pool, while at regional scales new individuals result from speciation. The dynamics of a community can therefore be modelled with a minimum of parameters: regional population size, speciation rate, migration rate and death rate. Such an approach can successfully generate patterns consistent with species–area (MacArthur & Wilson 1967) and abundance–frequency relationships (Tokeshi 1999). Neutral theory explicitly ignores differences between individuals in response to local ecological conditions. In contrast niche theory suggests that patterns of biodiversity should closely relate to underlying variability in ecological parameters such as physico-chemistry, disturbance regime, productivity and competition with other species (e.g. Tilman 1982; Tokeshi 1999).

Neutral theory is difficult to test in practice (Harte 2003; Gilbert & Lechowicz 2004; Wootton 2005) particularly as key population parameters have rarely been measured (but see Wootton 2005). Attempts to fit species-abundance curves generated by neutral theory and other models to real data have been made (McGill 2003; Volkov et al. 2003; Adler 2004; Alonso & McKane 2004; Chisholm & Burgman 2004), but differences in model fit are often negligible (Harte 2003; Hubbell & Borda-de-Agua 2004). Even when the model fits (such as the log-normal curve fitted to Panamanian tree diversity by McGill 2003) it may not inform us about the underlying biological processes. There have been attempts in recent years to test empirically patterns expected to emerge from neutral processes (e.g. Condit et al. 2002; Gilbert & Lechowicz 2004; Wootton 2005). One of these emergent patterns is that of ‘distance decay’ (Hubbell 2001). Because dispersal limitation underlies differences between sites in a neutral world, it is expected that widely separated points will harbour different communities (Harte 2003). Differences in local species richness between sites can be explained by random extinctions and replacements of species through time, a process Hubbell (2001) calls ‘ecological drift’. This process has a direct analogy with ‘neutral allele theory’ (Kimura 1983), the process whereby changes in genomes are accumulated passively through time via ‘genetic drift’. As in genetics, where the relative importance of genetic drift vs. natural selection has been widely debated (e.g. Mayr 1991; Ridley 2002), so too has Hubbell's assertion of the dominance of ecological drift over local determinism (Nee & Stone 2003).

Neutral theory can be tested by comparing the fit of community data to local ecological conditions, vs. their fit to distance decay expectations. Under niche theory, similarity between species-abundance matrices will be positively correlated with similarity in local ecological conditions. Neutral theory predicts a negative correlation with distance between sites. Such a test is made more difficult because distance between sites is often positively correlated with differences in local ecological factors (Gilbert & Lechowicz 2004). Such an approach is only valid when local conditions and spatial separation are independent of one another. Stream systems provide a useful test system because they generate a spatially constrained set of local conditions with intervening inhospitable habitat. Aerial dispersal of adults allows movement of individuals between patches on realistic scales.

We used macroinvertebrate communities from a set of stream sites in a large river catchment in New Zealand to test neutral and niche predictions. Neutral theory predicts decreased similarity between invertebrate communities that are spatially distant. Niche theory predicts decreased community similarity with lower similarity in ecological conditions. Alternatively, we might expect both neutral and niche processes to contribute to patterns of local diversity, leading to a pattern of decreased similarity in communities as spatial distance increases and niche similarity decreases. Because dispersal limitation underlies neutral expectations, we also analyse separately communities with varying dispersal ability. Community similarity for species with poor dispersal should be strongly negatively associated with spatial separation of sites, while species with good dispersal should overcome dispersal limitation and be strongly positively associated with local ecological conditions.

Materials and methods

spatial organization of study sites

Ten streams were sampled as a part of a study of differences in food web structure (see Townsend et al. 1998). Each sampling location was 30 m long and included at least one pool and one riffle. All sites were in grassland catchments and were sampled once during the austral summer between 5 and 16 January 1995. The study streams were separate third or fourth order tributaries of the Taieri River in New Zealand (Townsend et al. 1998). The sites were distributed over a total area of 4385 km². Intersite distances were calculated by digitizing a map of the area and using image measurement software (bersoft 4·01). Direct distances among sites were calculated by measuring vector lengths between all site pairs.

measurement of local ecological factors

A range of ecological factors were measured at each site (see Townsend et al. 1998 for details). Water chemistry was measured by taking duplicate water samples from each site and analysing for pH and nitrate plus nitrite (NO₃ + NO₂) (American Public Health 1992). These measurements were used because they provide good proxies of variation in land use intensity (Townsend et al. 1998).

Physical parameters were summarized by measuring width, maximum depth and flow at the swiftest point (using a Marsh–McBirney ‘Flo-mate’ meter), at 1-m intervals at each location (30 measurements for each variable in total). One hundred random substrate particles were collected in each location and their length along the longest axis measured. Averages and standard deviations for width, depth, flow and substrate size were calculated.

Disturbance was measured using the methods reported in Townsend, Scarsbrook & Dolédec (1997a). At each location bed particles were taken corresponding to the 50th, 75th and 90th percentiles of the substratum size distribution. These were painted and arranged on the stream bed in regular arrays (five rows of three particles in each of the three size classes randomly assigned to transects placed 1 m apart). The movement of these particles was monitored on five occasions from September 1993 to June 1994. Intensity of bed disturbance at a location was calculated as the average of the percentage of painted particles that moved between consecutive sampling occasions (see Townsend et al. 1997a).

Organic matter standing crop was measured as described in Townsend et al. (1998). Ten Surber samples (area 0·06 m², mesh size 250 µm) were placed at random locations along the stream during a period of base flow. Organic matter derived from these benthic samples was weighed, ashed (550 °C for 12 h) and reweighed to ascertain the amount of organic matter on the stream bed. Two 500-mL water samples were filtered through preweighed Whatman GF-C filters, dried, weighed, then ashed (550 °C for 3 h) and reweighed to ascertain availability of organic matter as seston (water-borne organic matter).

Biofilm at each site was measured as ash free dry weight (AFDW) per m² of substrate. Ten cobbles were gathered at random along the stream reach and scrubbed clean of biofilm into a known volume of distilled water. The sample was homogenized and three 15-mL samples were filtered on to preweighed Whatman GF-F filters, dried, weighed, ashed (550 °C for 3 h) and reweighed to measure the amount of biofilm on each rock. Each scrubbed rock was wrapped in aluminium foil, then the foil was weighed and a foil weight/area regression used to estimate the surface area of each rock.

Total potential algal production was estimated using a radioactive carbon isotope [7 mL of ¹⁴C-NaHCO₃ (185 MBq ml⁻¹)] within portable chambers (Fuller & Bucher 1991; Thompson & Townsend 1999). Twenty rock samples (each approximately 10 cm²) were collected at random from along the study reach and placed in two recirculating perspex containers. The chambers were placed in the study reach and the isotope was added. After a 2-h incubation period rocks were removed and the amount of carbon taken up was calculated by extraction in dimethyl sulphoxide followed by counting in a scintillation counter. To provide an estimate of the amount of carbon fixed per m² of substrate, the surface area of each rock was measured. This was calculated by wrapping the rock in aluminium foil, weighing the foil, and then applying a weight/area regression to estimate foil area.

sampling of stream biota

Macroinvertebrate samples were taken from 10 random locations at each location using a benthic sampler (area 0·06 m², mesh size 250 µm, taking the top 5 cm of substratum). Analyses of a subset of these streams have shown that this effort is sufficient accurately represent the invertebrate community (Thompson & Townsend 1999). Samples were preserved in 5% formalin for return to the laboratory. Samples were sorted for aquatic invertebrate larvae, excluding those less than 1·5 mm. Such a procedure excludes meiofaunal species – but these are rare in these systems (Thompson and Townsend 2004). All invertebrates were identified and counted. Invertebrate identification was carried out at 10–40× magnification, identifying animals to the highest degree of taxonomic resolution possible (Thompson & Townsend 1999).

Species were attributed functional feeding groups based on similar feeding characteristics. Species were divided into: grazers (predominantly consume algae), predators (feed only on other invertebrate species), and others, based on dietary information published from these streams (Townsend et al. 1998). Dispersal ability of invertebrate species was attributed using an independent data set gathered for an unrelated study (unpublished data, Ngaire Phillips, NIWA, Hamilton, New Zealand). Dissemination potential was described as low (10 m), medium (1 km) or high (10 km) based on a combination of published and unpublished literature, and augmented where required with expert knowledge (sensuTownsend, Doledec & Scarsbrook 1997b). All dispersal classes were represented in all trophic groups.

statistical analyses

Three potential predictor matrices were extracted from the data. Ecological data (Table 1) were standardized and reduced to a matrix of Euclidean dissimilarities between sites (primer, Plymouth Marine Laboratories). A spatial matrix was constructed of pairwise spatial distances between sites (Table 2). A combined ecological and spatial matrix was constructed by incorporating pairwise distances between sites as variables in the ecological matrix, before standardization and reduction to Euclidean dissimilarities. To ensure that ecological and spatial data were not correlated, the correlation between these matrices was tested using a Mantel test (Mantel 1967; Bonnet & de Peer 2002). Abundance data for invertebrates were summarized into a matrix of pairwise Bray–Curtis similarities between sites, using the program primer (Plymouth Marine Laboratories). To test for an effect of trophic group, separate matrices including grazers and predators were extracted. To test for an effect of dispersal ability, matrices were extracted of low, intermediate and high dispersal species separately. Associations between the invertebrate matrices and the three predictor matrices were assessed using Mantel tests (Bonnet & de Peer 2002).

Table 1.  Local ecological conditions in the 10 streams. Disturbance is an index based on the percentage of bed particles moving in a series of discharge events (see Materials and methods)

	Stream	Blackrock	Broad	Canton	Dempsters	German	Healy	Kye Burn	Little Kye	Stony	Sutton
Water	pH	6·60	6·32	6·68	6·90	7·35	6·94	6·60	6·81	6·81	7·01
chemistry	Nitrate/nitrite (µg L^−1)	15·73	8·79	16·80	21·18	4·11	7·28	5·06	5·30	4·50	1·45
Algae	(mg C m^−2 h^−1)	20·74	27·35	23·04	85·38	88·29	61·75	15·56	11·14	22·45	9·26
	Biomass (g m^−2)	1·45	1·20	1·95	1·32	2·29	3·61	1·24	2·06	11·99	0·95
Organic	Seston (mg L^−1)	7·12	0·77	4·59	0·63	1·07	0·34	0·47	0·76	0·88	0·94
matter	Benthic (g m^−2)	11·30	4·85	11·45	3·43	0·75	2·56	0·62	13·26	1·10	4·14
Channel	Average depth (cm)	13·50	15·90	20·30	22·00	7·30	15·80	6·50	21·50	14·10	11·60
	SD	0·03	0·11	0·10	0·14	0·02	0·15	0·03	0·15	0·05	0·06
	Average width (m)	1·20	1·18	1·25	3·72	3·28	3·03	1·97	6·65	2·85	2·40
	SD	0·71	0·34	0·36	0·69	0·56	0·59	0·38	1·66	0·37	0·26
	Average flow (cm s^−1)	161·70	118·50	110·40	66·30	83·70	96·10	43·90	87·90	117·80	70·45
	SD	52·76	111·92	48·12	77·11	34·83	133·53	41·84	60·20	46·47	44·61
	Average substrate (mm)	34·75	57·43	81·19	118·93	59·61	96·75	122·02	34·19	182·63	106·22
	SD	35·99	70·06	96·44	99·03	62·13	83·35	113·90	21·25	120·57	112·88
	Gradient (m m^−1)	0·02	0·02	0·01	0·03	0·03	0·03	0·05	0·01	0·02	0·02
	Disturbance (%)	46·57	43·46	26·25	28·55	62·21	39·76	42·20	84·57	28·36	45·94

Table 2.  Direct distances between the 10 sampling locations in kilometres

	Blackrock	Broad	Canton	Dempsters	German	Healy	Kye Burn	Little Kye	Stony
Broad	2·56
Canton	2·89	3·36
Dempsters	38·74	41·08	40·41
German	92·26	94·32	91·85	71·78
Healy	96·11	98·57	95·88	73·91	6·14
Kye Burn	96·00	98·64	95·95	73·28	7·42	1·62
Little Kye	87·55	89·78	87·08	67·89	5·25	11·12	12·00
Stony	25·07	26·57	23·08	41·45	72·50	77·49	77·83	67·40
Sutton	22·61	23·77	20·07	40·74	74·77	79·50	79·83	69·57	2·85

Linear regression was used to identify the best predictors of similarities between invertebrate communities (SPSS 11.0.12001). Bray–Curtis similarities between communities were used as the dependent variable and regressed separately against all ecological factors, spatial distance between sites, and ecological factors and spatial distance combined. Because of nonindependence of data, significance was tested using randomization tests. The best model was chosen based on the requirement of minimizing the Akaike Information Criterion (AIC) (Akaike 1973; Anderson, Burnham & Thompson 2000). AIC values are minimized for models with high regression coefficients and incorporate a minimum of predictors. Plots were generated between Bray–Curtis similarities and each of: the best ecological model; spatial distance; and the best combined ecological and spatial distance model.

Results

The 10 streams contained a total of 89 taxa, 67·4% of which we identified to species (Table 3). We identified 64·2% of individuals to species level and 98·7% to genus. Seventy-two of the taxa could be attributed a dispersal ability value.

Table 3.  Summary of invertebrate data from the 10 locations showing number of individuals identified, total number of taxa and number of taxa in different trophic and dispersal ability groupings

Stream	No. of individuals	No. of taxa	Trophic grouping		Dispersal ability
			Grazers	Predators	Low	Moderate	High
Blackrock	3206	40	11	9	5	25	7
Broad	2845	37	12	9	7	22	5
Canton	3447	48	11	11	6	26	11
Dempsters	2959	46	14	13	5	28	8
German	3041	32	5	6	3	22	5
Healy	3510	45	10	13	4	27	9
Kye Burn	2778	35	9	6	4	22	7
Little Kye	3305	29	7	7	2	16	6
Stony	3178	45	11	12	5	27	12
Sutton	2787	35	9	12	4	20	8

There was no statistically significant correlation between spatial location of sites and their local ecological conditions (Pearson's Correlation = 0·108, P = 0·165) (Fig. 1). Therefore our analyses of effects of spatial location and ecological conditions were independent. There were high correlation coefficients between the spatial matrix and invertebrate matrices for all taxa, grazers and taxa with low dispersal ability (Table 4). However, these invertebrate groups, as well as predatory taxa and taxa with moderate dispersal abilities, also showed moderate to strong correlations with local ecological conditions (Table 4). For all invertebrate taxa, grazers, and taxa with low and moderate dispersal abilities, Mantel test's found the highest correlation with the predictor matrix combining spatial separation and local ecological conditions (Table 4).

Figure 1. Relationship between local ecological factors and the spatial distance between sites (in kilometers).

Table 4.  Pearson's correlation coefficients from Mantel tests between invertebrate abundance matrices and the three predictor matrices

	Spatial	Local ecology	Local ecology + spatial
All invertebrates	0·457	0·200	0·519
Trophic group
Grazers	0·415	0·375	0·462
Predators	0·148	0·407	0·384
Dispersal ability
Low	0·381	0·540	0·594
Mid	0·054	0·323	0·369
High	0·087	0·191	0·150

Distance between sites was negatively related to similarity in patterns of abundance for all invertebrate taxa and for grazers and predators separately (Fig. 2a–c). Grazing community similarity had a negative relationship with variability in depth, and a positive relationship with disturbance and average current (Fig. 2d). In contrast, similarity between predator communities was negatively associated with current and positively associated with channel slope and width (Fig. 2e). Similarities in the overall invertebrate communities were not well predicted by any of the ecological variables (Fig. 2f). Linear models incorporating spatial distance and ecological factors were the best predictors of similarities in invertebrate community structure (Fig. 2g–i), although support for the predator community model was relatively weak. The models for all groups included average current speed, with the model for grazers also including negative associations with pH and variability in current (Fig. 2g).

Figure 2. Relationships between community similarity (Bray–Curtis) and best linear models (shown on x-axis), for grazers, predators and all invertebrates from the 10 streams. Regression lines are shown where the slope is significant (P < 0·05) and r² > 0·10. sd = standard deviation, av = average, disturb = disturbance, curr = current, dist = distance. For units see Tables 1 and 2.

When the invertebrate community was differentiated according to dispersal ability, a negative relationship was found between spatial distance between sites and community similarity for low dispersal and moderate dispersal groups (Fig. 3a,b), but no relationship was found for the high dispersal group (Fig. 3c). The slope of this relationship was higher for the low dispersal taxa than the moderate dispersal taxa (ancova; F_1,86 = 4·343, P = 0·040). A model including disturbance, primary production, seston, algal biofilm biomass, pH and average depth was the best predictor of similarity between moderate dispersal groups (Fig. 3e), but low and high dispersal groups were not well predicted by ecological variables (Fig. 3d,f). Models incorporating spatial distance together with a variety of ecological factors were reasonable predictors of community similarity for low and moderate dispersal groups (Fig. 3g,h), but not for their high dispersal counterpart (Fig. 3i). Models combining spatial distance and ecological factors were the best predictors of similarity in all community combinations (Table 4, Figs 3 and 4).

Figure 3. Relationships between community similarity (Bray–Curtis) and best linear models (shown on x-axis), for taxa with low, moderate and high dispersal ability. Regression lines are shown where the slope is significant (P < 0·05) and r² > 0·10. sd = standard deviation, av = average. For units see Tables 1 and 2.

Figure 4. Akaike Information Criterion (AIC) values for models predicting community similarity for grazers, predators, all taxa, and taxa with low, moderate and high dispersal abilities. Lower AIC values indicate a better fitting model. The results for the best models incorporating distance only, ecological factors only and both sets of predictors are shown (see Fig. 3 for model parameters).

Discussion

Our results are consistent with predictions of neutral theory and a role for local ecological conditions. Patterns of species occurrence and abundance were correlated with the spatial arrangement of sites and the conditions that were present at each. Grazer communities (the numerically dominant component of these streams) were most strongly correlated with a combined model of local ecological conditions and spatial distance, followed by local ecology alone then spatial distance alone. In contrast, the predatory component of the community was most strongly correlated with ecological factors alone followed by the combined model. There was also evidence for an interaction with species’ dispersal ability. While the strongest correlation for taxa with moderate dispersal ability was with the combined model, taxa with low dispersal ability were most strongly correlated with spatial distance, and high dispersal taxa were weakly correlated with all models. Perhaps the latter taxa disperse over scales that exceed those in this study. A previous investigation of communities from a much wider range of sites throughout the Taieri catchment showed that geographical location was generally influential in accounting for community composition but not in the case of invertebrates with strongly flying adults (Townsend et al. 2003).

There have now been a series of studies reporting limited or no support for neutrality. The majority of these have concentrated on sessile organisms: rainforest trees (Clark & MacLachlan 2003; McGill 2003; Tuomisto, Ruokolainen & Yli-Halla 2003; Alonso & McKane 2004), grasses (Adler 2004), boreal forest (Alonso & McKane 2004), barnacles (Wootton 2005), understorey plants (Gilbert & Lechowicz 2004) and parasites (Poulin 2004). Sessile organisms operate under a set of constraints that may favour dispersal limitation and thus neutrality: they rely on broadcasting propagules into the environment, and the eventual site occupied is strongly dependent on the initial settlement site. Sessile organisms are also often limited by space, so that lottery effects of establishment may predominate, and many (such as rainforest trees or grasses) may be considered ecologically equivalent, apparently satisfying the requirements of neutral theory. Despite these tendencies, local ecological factors have generally been found to better predict sessile communities than neutrality. Our study incorporated mobile organisms that may have greater control over where in the landscape they and their offspring occur. Aquatic insects can select suitable habitat by choosing oviposition sites (e.g. Timm 1994; Winterbourn 2003) or by habitat selection as larvae (e.g. Holomuzki & Messier 1993; Townsend et al. 1997b). For that reason we expected that associations between local ecological conditions and macroinvertebrate communities may be stronger than occur for sessile organisms. While we did find associations with ecological conditions, the strong role indicated for spatial separation of sites suggests that on this scale dispersal limitation is important even for highly mobile organisms.

Scale, as in other areas of ecology, has become an increasingly important consideration with regard to neutrality (Adler 2004; Alonso & McKane 2004). Given the importance of dispersal in neutral theory, the scale of investigation can be expected to affect whether neutral patterns are observed. Some rigorous tests of neutral theory have been carried out on relatively small spatial scales (1–10 km) (Adler 2004; Gilbert & Lechowicz 2004). However, Alonso & McKane (2004) raised the contention that neutral theory may be valid only at extremely large spatial scales. We attempted to assess whether dispersal limitation was a relevant factor at the scale of this study by incorporating dispersal ability in the analysis. It may be expected that species with poor dispersal may be influenced to a greater degree by neutrality than those with high dispersal. Our results were consistent with this – the slope of the negative relationship between spatial separation and community similarity was steepest for low dispersal species, less steep for moderate dispersal species and nonsignificant for high dispersal species. The moderate dispersal group was also more strongly influenced by local ecological conditions than the low and high dispersal groups. The biological trait ‘dispersal ability’ deserves to be incorporated into future tests of neutral theory. In fact, we suggest that suitably designed studies at an appropriate scale may reveal a role for dispersal limitation, along with local conditions, even in plant and other sessile communities.

Neutral theory relates to within-trophic level diversity (Poulin 2004). Therefore our test of the theory for an entire invertebrate community goes beyond that imagined by Hubbell (2001). Despite this, the entire community showed evidence of a negative relationship between spatial separation and community similarity. This is likely due to aerial dispersal being the primary means of reaching different locations. It is also probable that lottery effects that favour the founding of populations in a particular location may cause the development of local nodes of abundance for some species. The apparent strong effect of dispersal on species in this study is not surprising given the nature of the communities. For instance, many of the predatory species show high dietary overlap (Thompson & Townsend 2004), are taxonomically closely related (primarily occurring in the trichopteran family Rhyacophilidae), and are morphologically difficult to distinguish (R. Thompson pers. obs.). These species are likely to be largely functionally equivalent and therefore ‘exchangeable’ in communities as stochastic events remove species. The effect of phylogenetic relatedness in determining the role to which species compete and coexist is worthy of attention in future studies. High degrees of phylogenetic relatedness within communities may favour ecological equivalence and therefore increase the relative importance of neutral factors.

The finding that dispersal limitation is an important factor in determining community structure in these streams cannot, in isolation, be taken as confirmation that neutral theory is correct. Bell (2000) emphasized that neutral patterns (i.e. random accumulations of differences between sites) may result from non-neutral processes and showed (Bell 2001) that metapopulation dynamics can provide neutral type patterns through populations ‘blinking in and out’ at a landscape scale as local extinctions and recolonizations take place. This scenario does not require the explicit assumptions of neutrality. Distinguishing between explanations is likely to require a direct assessment of life-history traits, such as individual reproductive success, that underpin the neutral model (sensuWootton 2005). We can also not rule out the possibility that differences in the invertebrate communities result from an unmeasured ecological factor that was correlated with distance. Two things appear to make this unlikely, (1) there is no readily identifiable unmeasured ecological factor, and (2) it seems unlikely that this factor would be correlated with distance when no other ecological factor in the analysis was.

Given that neutral theory alone cannot explain species-abundance relationships in this data set, can we integrate some of the assumptions of neutrality into a niche-based framework? Sophisticated niche assembly models have already been generated that explain species–abundance patterns well (Mouillot, George-Nascimento & Poulin 2003; Sugihara et al. 2003; Tilman 2004). Of these, the hierarchical niche model (Sugihara et al. 2003), like the neutral model, requires underlying generalizations (such as constant variance) that we know are unlikely to be true. Like the neutral model, however, it also incorporates processes whose importance is supported by our results, such as the interaction between multiple ecological factors to generate a niche. The manner in which niche and neutral models interact provides a rich opportunity for mathematicians, modellers and empirical scientists to collaborate. The production of a synthetic model incorporating neutrality and niche assembly has already been attempted (Etienne & Oliff 2004; Schwilk & Ackerley 2005). A continuation of this process needs to take into account the results presented here – that trophic groupings and species traits are important factors in determining the weighting of niche and neutral factors.

The literature addressing neutral theory has largely been phrased in terms of either fully accepting or fully rejecting the theory. In part this is due to the way in which the original theory was established – complete neutrality provides no room for any of the processes that niche theory relies upon. On the other hand, the important processes that underlie neutral theory – dispersal limitation, stochastic loss and speciation – have largely been ignored by community ecologists entrenched in a niche paradigm (but see Ricklefs & Schluter 1993). If neutral theory is treated as an idealization of the real world (in much the same way that many physical laws are unrealistic idealizations) then many of its unrealistic assumptions become acceptable (Harte 2003), and incorporating other theories such as niche theory to explain variability about the idealization becomes conceptually tenable. The current results are compatible with both neutral theory and niche theory: dispersal limitation apparently contributes to differences in local community structure, as do differences in local ecological conditions. These results clearly show that local ecological conditions play a part in determining local patterns of macroinvertebrate abundance and diversity. However, the simultaneous influence of spatial distance in determining patterns indicates an equally important role for dispersal limitation. It is no longer a matter of neutral theory or niche theory, but how they operate together.

Acknowledgements

This data was collected by the Taieri and Southern Rivers Research Program at the University of Otago, Dunedin, New Zealand in a project funded by the New Zealand Foundation for Research Science and Technology. We would like to thank Mike Scarsbrook and Ngaire Phillips (National Institute of Water and Atmospheric Research, Hamilton, New Zealand) for the use of their data. Colin Bates, Ben Gilbert, Diane Srivastava, Brian Starzomski and the DiJo Group at the University of British Columbia provided valuable discussion and comments on drafts. RT was funded during the analysis by the Biodiversity Research Centre postdoctoral fellowship at the University of British Columbia.