- Top of page
- Materials and methods
- Supporting Information
Continental margins, the areas of seabed between the coast and the base of the continental slope, provide some of the most environmentally dynamic and heterogeneous habitats for marine species (Levin et al., 2010). As continental margin habitats are often shaped by both terrestrial and oceanographic influences, and have topographically complex surfaces as a result of geological processes (Levin & Dayton, 2009; Levin & Sibuet, 2012), the distributions of marine species should reflect a variety of environmental associations in these areas. In addition, it could be expected that environmentally heterogeneous areas will have high assemblage diversity, as assemblage structure should change along the gradients provided by environmental variables (Ricklefs, 1987).
Although depth is commonly identified as the most important correlate of species distributions in continental margin environments (Carney, 2005), a number of other topographic and oceanographic variables, including slope, temperature, dissolved oxygen and organic matter, interact with depth and play an important role in structuring assemblages (Levin et al., 2010; Pitcher et al., 2012), as well as maintaining biogeographical barriers to dispersal and migration (e.g. Wares et al., 2001). For instance, along the south-west Australian continental slope, species richness is associated with bathymetry, bottom temperature and oxygen concentration, the latter two variables representing water-mass type and tidal currents (Williams et al., 2010), while on the Chilean continental margin, distributions of sublittoral benthic organisms are associated with water-mass type, bottom-water oxygen concentration and sediment organic loading (Sellanes et al., 2010). Such studies show that continental margins are not monotonous mud slopes, as previously conceived (see review by Levin & Sibuet, 2012), but are regionally diverse areas that are often shaped by a variety of factors. However, relationships between environmental heterogeneity and assemblage diversity are poorly understood.
It has been suggested that comparisons between contrasting continental margins are needed to provide insights into the role of regional environmental heterogeneity in driving changes in species composition (Levin & Dayton, 2009). The Challenger Plateau and Chatham Rise, extending from the west and east coasts of New Zealand, respectively, provide a unique opportunity to identify the role of complex topographic and oceanographic complexity in driving assemblage composition, as they have contrasting oceanographic environments but similar ranges of latitude (38–45° S) and depth, and are known to have shared species. Challenger Plateau lies beneath relatively homogeneous warmer subtropical waters, whereas Chatham Rise has a more complex geomorphology (Nodder et al., 2012) and is bisected along its east–west axis by the Subtropical Front (STF; Heath, 1972). Chatham Rise is an area of high primary productivity (Nodder & Northcote, 2001), which supports major commercial fisheries for species including orange roughy (Hoplostethus atlanticus) and hoki (Macruronus novaezelandiae) (Bull et al., 2001). Previous studies of both benthic and pelagic fauna across Chatham Rise have identified species composition changes associated with the STF and depth (reviewed by Carney, 2005). For example, mesopelagic fishes, decapods, euphausiids and squid displayed differences in species composition from the north to the south of Chatham Rise (Robertson et al., 1978).
We hypothesize that benthic assemblages will be more diverse in an area of high topographic and oceanographic complexity, and examine this by comparing benthic invertebrate assemblages across Chatham Rise and Challenger Plateau using data from the most extensive survey of benthic fauna to date in this region, in combination with regional-scale environmental variables. To investigate relationships between turnover in benthic assemblage composition (beta diversity; Legendre et al., 2005; Ferrier & Guisan, 2006) and environmental variables, we used three methods: one based on individual species distribution models (SDMs) using boosted regression trees analysis (BRT; Friedman et al., 2000), and two community modelling methods, namely generalized dissimilarity modelling (GDM; Ferrier et al., 2007) and gradient forest analysis (GF; Ellis et al., 2012; Pitcher et al., 2012). Using these relationships, we then predicted turnover in assemblage composition across unsampled space in the study region. The predicted patterns in assemblage turnover, as described by eight environmental variables, are here defined as ‘biophysical patterns’. While the statistical approaches in this paper differ technically, the predictive outputs all represent expected patterns of assemblage composition mapped over space (Leaper et al., 2011). Thus, the primary purpose of applying these three methods was to assess whether the assemblage patterns inferred by each were consistent across the region, and if so to give greater confidence to their predictions, rather than to compare the methods per se. An ability to describe and predict compositional change across large continental margin areas with respect to variables describing the topographic and oceanographic environment, using sparse ecological data, is relevant to the management of these areas, as they are increasingly becoming the foci of commercial fishing activities and mineral extraction (Levin & Dayton, 2009).
- Top of page
- Materials and methods
- Supporting Information
The sites sampled on Chatham Rise encompassed a wider range of environments in terms of seabed temperatures (tempres), primary production (vgpm and poc) and tidal current speeds compared with the sites sampled on Challenger Plateau (Table 1). Across all surveyed sites, 833 benthic invertebrate OTUs were identified, but only a few occurred frequently (only 116 OTUs with > 13 individuals). Regional (gamma) diversity was higher on Chatham Rise (DTIS 191 OTUs, SEL 319 OTUs for n = 40 samples) than on Challenger Plateau (DTIS 171, SEL 275 for n = 40 samples), but average OTU richness per site (alpha diversity) was similar between areas (Chatham DTIS 24 and SEL 17, Challenger DTIS 21 and SEL 21). In addition, while a proportion of OTUs occurred in both areas (DTIS 41%, SEL 19%), the percentage of OTUs unique to Chatham Rise was higher than for Challenger Plateau (DTIS 17% and 4% unique, SEL 42% and 20% unique for Chatham Rise and Challenger Plateau, respectively).
Species distribution modelling
Fifty OTUs were modelled from the DTIS data, and the most important variables contributing to the models were bathymetry (bathy: 22% of overall contribution, Table 4), sea surface temperature gradient (sstgr: 13%) and depth-corrected temperature (tempres: 13%). Eighteen OTUs were modelled from the SEL data, and the most important variables contributing to the models were depth-corrected temperature (tempres: 26% of overall contribution), bathymetry (bathy: 15%) and primary productivity (vgpm: 14%). The BRT models had an average AUC of 0.80 for species modelled from the DTIS and of 0.83 for species modelled from the SEL, excluding species models where AUC < 0.74 (Table 2).
Table 2. The number of operational taxonomic units (n OTU) of the benthic taxa across both Challenger Plateau and Chatham Rise used in each analysis and the average performance results from the community-level modelling approaches (SDM, species distribution modelling approach; GDM, generalized dissimilarity modelling; GF, gradient forest analysis) for species sampled with the epibenthic sled (SEL) and deep towed imaging system (DTIS). The performance of SDM is given by an average area under the receiver operating curve (AUC) across the boosted regression trees models with an AUC > 0.74, the GDM is assessed by the percentage of deviance explained (% Dev expl), and the GF is assessed with the R2 or goodness-of-fit. The data used for the modelling are also given (P/A, presence/absence; Abund, abundance)
|% Dev expl||20||22|
|% Dev expl||30||22|
|R 2 ||0.2||0.18|
Mapped probabilities of occurrence from the BRT models showed that some OTUs were narrowly distributed and had specific associations with topographic or oceanographic features. For instance, the gastropod mollusc Poirieria zelandica was strongly associated with conditions on the crest of Challenger Plateau, whereas the echinoid Enypniastes eximia was associated with conditions in deeper water on the flanks of the plateau (Fig. 2; Table 3). Similarly, on Chatham Rise, the brittle star Ophiomusium lymani was associated with conditions deep on the southern flank of the rise, and the alcyonacean Taiaroa tauhou with shallower depths on the southern flank (Fig. 2; Table 3). By contrast, other taxa were more broadly distributed and had less specific environmental associations. These taxa included the errant polychaete Hyalinoecia longibranchiata, which was associated with conditions on the crests and shallow southern flanks of both Chatham Rise and Challenger Plateau, and the decapod crustacean Metanephrops challengeri, which occurred in shallow waters on the crests of both features, as well as in continental shelf areas closer inshore (Fig. 2; Table 3). All BRT model results of species used to describe overall assemblage composition can be seen in Appendix S2.
Figure 2. Probability of occurrence as predicted by boosted regression tree models of six benthic invertebrate species having different environmental associations along the two margins – two species associated with Challenger Plateau: (a) the echinoid Enypniastes eximia, and (b) the gastropod mollusc Poiriera zelandica; two species associated with Chatham Rise: (c) the alyconacean Taiaroa tauhou, and (d) the ophiuroid Ophiomusium lymani; and two species broadly distributed across both areas: (e) the errant polychaete Hyalinoecia longibranchiata, and (f) the decapod Metanephrops challengeri. The relative abundance of each species is shown. Spatial predictions were restricted to areas shallower than 1850 m. A sinusoidal equal-area projection was used for all maps. See Table 1 for a definition of the variables.
Download figure to PowerPoint
Table 3. The relative influence of the environmental variables (percentage) and optima from the individual boosted regression tree (BRT) models for selected taxa of benthic invertebrates across Chatham Rise and Challenger Plateau. The spatial predictions derived from these individual operational taxonomic unit (OTU) models are shown in Figs 2–4. For all BRT model results, see Appendix S2
|OTU and phylum||Gear||Variable||Percentage||Optima|
|Poirieria zelandica, Mollusca||SEL||tempres ||61||1.7|
|sstgr ||13||0.002°C km−1|
|vgpm ||12||485 mg C m−2 d−1 |
|Enypniastes eximia, Echinodermata||DTIS||poc ||21||8 mg Corg m−2 d−1|
|bathy ||40||1203 m|
|mld ||4||61 m|
|Taiaroa tauhou, Cnidaria||SEL||tempres ||30||−0.8|
|vgpm ||22||700 mg C m−2 d−1 |
|sstgr ||14||0.02°C km−1|
|mld ||12||46 m|
|Ophiomusium lymani, Echinodermata||SEL||bathy ||43||1198 m|
|vgpm ||11||453 mg C m−2 d−1|
|Metanephrops challengeri, Arthropoda||DTIS||bathy ||43||460 m|
|vgpm ||13||648 mg C m−2 d−1|
|mld ||12||54 m|
|sstgr ||10||0.01°C km−1|
|Hyalinoecia longibranchiata, Annelida||SEL||tidcurr ||20||0.2 ms−1|
|poc ||17||16.7 mg Corg m−2 d−1|
|bathy ||14||581 m|
On Challenger Plateau, fitted surfaces describing associations between the NMDS ordination (three dimensions, stress: SEL Chall = 17.9, DTIS Chall = 17.6) and both bathymetry and depth-corrected temperature were concordant (approximately ‘parallel’ contours from smaller to bigger values, Fig. 3), indicating that the survey composition data showed similar associations with these variables. In contrast, across Chatham Rise (NMDS three dimensions stress: SEL Chat = 20.9, DTIS Chat = 20.8), the fitted surfaces differed in shape and direction (Fig. 3), indicating that the survey composition data showed different associations with these two variables in this region.
Figure 3. Two-dimensional non-metric multi-dimensional scaling (NMDS) ordinations of Bray–Curtis dissimilarities between benthic community composition at sites sampled across Challenger Plateau (Chal, left panels) and Chatham Rise (Chat, right panels), for video (deep towed imaging system, DTIS, upper panel) and epibenthic sled (SEL, lower panel) data. Bubble plots show the relative value of the bathy and tempres values across the NMDS ordination. Contours show the fit between the values from the NMDS ordination and depth (bathy) and depth-corrected temperature residuals (tempres). See Table 1 for a definition of the variables.
Download figure to PowerPoint
The GDM models identified that, of the eight variables examined, bathymetry (bathy contribution: 42% DTIS and 45% SEL) and depth-corrected temperature (tempres: 19% DTIS and 14% SEL) were the most important variables contributing to compositional turnover (Table 4). The contribution of the remaining variables was lower (0–14%), with sea surface temperature gradient making no contribution. The averaged fitted response curves derived from the SEL and DTIS GDM curves combined (Fig. 4a, BOTH) showed that compositional changes were greatest in the 0–500 m bathymetry range and across the transition from colder than expected to warmer than expected waters (tempres −1 to 1; Fig. 4a), given their depth. Compositional change was also highest at the lower end of the productivity scale (poc 0–15 mg Corg m−2 d−1 and vgpm 400–500 mg C m−2 d−1), and at mixed layer depths greater than 60 m. The fitted response curves were generally higher for SEL than for DTIS, except for primary productivity (vgpm), and the greatest differences between the gears occurred for bathymetry and slope.
Figure 4. Fitted functions from (a) generalized dissimilarity modelling and (b) gradient forest analysis, representing the turnover in assemblage composition of benthic invertebrates with respect to each of the eight environmental variables across Chatham Rise and Challenger Plateau. The functions representing compositional turnover with respect to environment are shown for each gear code [epibenthic sled (SEL) and deep towed imaging system (DTIS)] and for both gear codes combined (BOTH). Variables with the highest maximum fitted values have the greatest contribution to compositional turnover in each model. See Table 1 for a definition of the variables.
Download figure to PowerPoint
Table 4. Overall percentage contribution of the environmental variables to the boosted regression tree (BRT) species distribution models (SDMs) and the community-level modelling approaches [generalized dissimilarity modelling (GDM) and gradient forest analysis (GF)] using data from the deep towed imaging system (DTIS) and the epibenthic sled (SEL) of benthic taxa across Chatham Rise and Challenger Plateau. For the full names of the abbreviated predictor variables, see Table 1. The two highest contributions for each are shown in bold
| ||DTIS ||SEL |
|tempres ||12||26 |
|bathy ||22 ||15 |
|sstgr ||13 ||11|
|tempres ||19 ||14 |
|bathy ||42 ||45 |
|tempres ||17 ||21 |
|mld ||17 ||17 |
The GF models (DTIS n = 102 OTUs, SEL n = 114 OTUs) identified depth-corrected temperature (tempres: 17% DTIS and 21% SEL) and mixed layer depth (mld: 17% DTIS and 17% SEL) as the two most important variables contributing to turnover in assemblage composition (Table 4). In contrast to the GDM results, all other variables also made relatively high contributions (8–15% per variable, Table 4) and rates of assemblage change were more constant across the full range of each environmental variable (Fig. 4b). The combined cumulative response curves from both gears showed that the rate of assemblage turnover was greatest across the temperature gradient, particularly in the lower part of the range (tempres −1 to 0), and decreased in the order mixed layer depth, primary productivity, depth, tidal current speed, particulate organic carbon flux, seabed slope, and sea surface temperature gradient (steepness of curves in Fig. 4b). Also in contrast to GDM, cumulative response curves from the DTIS GF model had higher maximum values than the SEL GF model for bathymetry, mixed layer depth, particulate organic carbon and tidal current speed, indicating that the DTIS models had a higher overall explanatory power (R2 = 0.20) than the SEL model (R2 = 0.18) and that most of the explained variation in the model was attributable to these variables.
The predicted outcomes from the GDM and GF analyses consistently identified similar relationships between the benthic assemblages and environmental gradients (‘biophysical patterns’) across the region (Pearson correlations between distances in GF and GDM transformed ‘biological space’: BOTH gears, r = 0.63; SEL, r = 0.64; DTIS, r = 0.67). In addition, the mapped assemblage patterns obtained from combining the species distributions predicted by the BRT models were also similar to the results of the two community modelling methods (Fig. 5; Appendix S3). Mapped biophysical patterns from all three models indicated that environmental conditions and associated benthic assemblages on the shallow part of Challenger Plateau were distinct from those on Chatham Rise. Conditions in deeper waters surrounding Challenger Plateau, however, were similar to those along the deep northern edge of Chatham Rise and, in the GDM model, the strong influence of bathymetry and depth-corrected temperature led to this deep biophysical type extending south of Chatham Rise. Across Chatham Rise, there was a greater range of biophysical environments than on Challenger Plateau, with clear differentiation between the north and south flanks and, to a lesser extent, between the east and west ends, particularly in the SDM and GF models (Fig. 5). The greater assemblage diversity on Chatham Rise was associated with faster tidal currents, steeper slopes – particularly on the northern flanks of Chatham Rise – and colder sub-Antarctic waters along the southern flank (Fig. 5; Appendix S3). Although model performance measures were not directly comparable between the three methods (Table 2), all models explained a relatively small proportion of the total variation in the raw data (18–30%, Table 2).
Figure 5. Maps representing predicted patterns of assemblage composition of benthic invertebrates with respect to environment (i.e. ‘biophysical patterns’) across Chatham Rise and Challenger Plateau for all data combined (i.e. BOTH), as determined by an aggregation of the individual boosted regression tree (BRT) species distribution models (SDM, upper panel), generalized dissimilarity models (GDM, middle panel) and gradient forest analysis (GF, lower panel). The key to each is obtained by matching colours between the maps and the biplots, which show the first two dimensions of assemblage composition and vectors indicating the direction and magnitude of the eight environmental variables. For the aggregate SDM, the ordination is a non-metric multi-dimensional scaling of Bray–Curtis dissimilarities. For GDM and GF, principal components analyses (PCA) were used to display assemblage–environment associations. A sinusoidal equal-area projection was used for all maps. See Table 1 for a definition of the variables.
Download figure to PowerPoint
- Top of page
- Materials and methods
- Supporting Information
A comparison between two continental margins with similar latitudinal and depth ranges but contrasting oceanographic environments showed that the more environmentally heterogeneous Chatham Rise had greater assemblage diversity. Our study supports the hypothesis that increased assemblage diversity is linked to regional-scale environmental heterogeneity across continental margins. Because we used environmental parameters derived from broad-scale remote-sensing methods and oceanographic models, these findings could be general for continental margins in other regions.
All three methods identified that variables associated with water mass (i.e. seabed temperature and mixed layer depth) were strongly associated with the turnover in assemblages across the study area and that these variables were not constrained by bathymetry. These results accord with other studies that show that gradients in temperature, oxygen or productivity can be important for describing changes in species richness (Williams et al., 2010) and beta diversity (Gooday et al., 2010; Narayanaswamy et al., 2010; Pitcher et al., 2012), independent of the depth at which they occur. For example, benthic communities in the Gulf of Maine show changes in assemblage composition with sea surface temperature where warmer Gulf Stream waters transition to cooler sub-boreal waters independent of depth (Pitcher et al., 2012). Although the order of importance of these environmental variables differed between the three modelling methods used, this probably reflected technical differences between the models and environmental surrogacy between correlated variables rather than any real differences in the importance of the environmental variables.
The other environmental variables made a smaller contribution to explaining compositional change but enabled discrimination of other finer-scale biophysical patterns and thus were also important in describing turnover in benthic assemblages at relatively smaller scales, for example identifying assemblages associated with high primary productivity along the STF. This supports the suggestion that, while water-mass characteristics influence assemblage composition over large spatial scales on continental margins, strong environmental gradients at the interfaces between water masses contribute to increased beta diversity at smaller scales (Levin & Sibuet, 2012).
The contrasting assemblages on the northern and southern flanks of Chatham Rise, as shown by the biophysical maps, indicate that the STF creates a complex physical gradient that is not constrained by depth. These results support previous findings from Chatham Rise, which have identified changes in assemblage composition associated with the STF that are independent of depth (reviewed by Carney, 2005). This contrast also suggests that the STF forms a biogeographical barrier to species, possibly through physiological limitations associated with temperature (Adey & Steneck, 2001) or through the influence of water-mass circulation on larval dispersal. Evidence of a biogeographical barrier for benthic taxa in this area comes from studies of the sea star Patiriella regularis, which occurs close to the shore but shows a genetic discontinuity across the STF (Ayers & Waters, 2005).
Although more trawling has generally taken place on Chatham Rise than on Challenger Plateau, the frequency and the spatial distribution of fishing effort have inevitably varied over time. While it is highly probable that trawling will have effects on benthic assemblage structure, species or assemblage diversity alone does not provide an indication of how assemblages have changed, or are changing, as a consequence of trawling. In this study, we do not consider trawling as a factor influencing regional assemblage diversity because to do so it would be necessary to include quantification of changes in trawling disturbance over time. However, if we included fishing intensity, we would not expect our main conclusions to change, because no previous work has shown the contribution of fishing intensity to override the role of regional environmental drivers (Hewitt et al., 1998; Pitcher et al., 2007).
In this study, the fitted models and biophysical maps differed slightly between the two sampling methods, SEL and DTIS, for two main reasons. First, there were differences in the taxonomic resolution of the two data sets. Because the SEL retrieves physical specimens, which can be examined in detail, the resulting data are at a finer taxonomic resolution than the DTIS video data, for which species-level or genus-level identifications from moving images can be uncertain. In consequence, SEL samples had a high proportion of unique but low-abundance OTUs, whereas DTIS had fewer distinct OTUs but at higher abundances. In the case of GDM, a consequence of this is that the SEL data have greater Bray–Curtis dissimilarity values between sites than do the DTIS data. Second, there are differences in how the raw data are used by each modelling technique. The main differences here are that our BRT models were run with a minimum of 13 individuals and needed a minimum AUC to contribute to the final turnover predictions, GDM had the criterion that individuals occur at a minimum of three sites, and GF had a criterion that each OTU must occur in at least five samples and have an R2 > 0 to contribute to final GF turnover functions.
Although environmental gradients, as surrogates for biological distributions, provide a useful description of the overall patterns in compositional turnover across large spatial scales, these results and others (Leaper et al., 2011; Pitcher et al., 2012) have shown that the overall performance of models describing assemblage turnover using environmental surrogates alone is low, with generally less than 30% of total variance explained. Although it could be argued that the low performance of the statistical models arises from missing environmental variables, for example the influence of episodic changes in detritus flux (Nodder & Northcote, 2001), another recent study using up to 29 environmental variables found similar results, suggesting that the likelihood of missing important predictors is getting fairly low (Pitcher et al., 2012). This suggests that environmental variables are only part of the explanation, and that other factors, such as historical events, population dynamics, temporal environmental variability and species interactions, play important roles in structuring biological communities (Pitcher et al., 2012). The low local-scale prediction performance of regional environmental layers does not mean that they cannot make useful contributions for ecological understanding and management purposes, but suggests that outputs need to be interpreted with care as they provide a highly smoothed description compared with the raw data. Such approaches are of particular value for environments such as the deep seabed, where biological sample data are sparse but oceanographic variables can be quantified reliably across large areas. To date, they have been used to describe areas of unique ecological character at large spatial scales, including New Zealand's exclusive economic zone (Snelder et al., 2007) and Australia's continental shelves and slopes (Pitcher et al., 2012).