Thomas A. Schlacher, Faculty of Science, Health & Education, The University of the Sunshine Coast, Maroochydore DC, Qld-4558, Australia. E-mail: firstname.lastname@example.org
Submarine canyons increase seascape diversity on continental margins and harbour diverse and abundant biota vulnerable to fishing. Because many canyons are fished, there is an increasing emphasis on including them in conservation areas on continental margins. Here we report on sponge diversity and bottom cover in three canyons of South-eastern Australia, test the performance of biological and abiotic surrogates, and evaluate how biological data from detailed seabed surveys can be used in conservation planning in these habitats. The biological data on sponge assemblage structure and species richness were obtained from 576 seafloor images taken between 148 and 472 m depth, yielding 65 morphospecies. Seafloor characteristics were similar within and between canyons, being almost exclusively composed of sediments with very few rocky substrates of higher relief. This environmental homogeneity did not, however, translate into biological uniformity of the megabenthos, and environmental factors were consequently poor predictors of biological features. By contrast, total bottom cover of sponges was highly correlated with species richness and served as a good proxy for species-level data in this situation. Design strategies that employ information on cover or richness of sponges provided a large dividend in conservation effort by dramatically reducing the number of spatial units required to achieve a specified conservation target of 50–90% of species to be included in reserves. This demonstrates that image-derived data are useful for the design of reserves in the deep sea, particularly where extractive sampling is not warranted. Using biological data on the sponge megabenthos to identify conservation units can also minimise socio-economic costs to fisheries because of a smaller geographic and bathymetric ambit of conservation areas.
Most of the world’s continental margins are incised by submarine canyons. These are highly complex seascapes of great topographical diversity, often characterised by heterogeneous sedimentary features (Trincardi et al. 2007). Canyons are major conduits for the transport of material from the shelf to the abyss (Puig et al. 2003) and are global deposition centres of sediment on continental margins (Oliveira et al. 2007). Canyons can also act as sinks for terrestrial carbon transported by river plumes over the continental shelf, and thus play an important role in land–ocean coupling on margins (Waterson & Canuel 2008). Canyon seascapes can interact strongly with hydrodynamic regimes on continental margins where they intensify mixing and amplify currents (Martín et al. 2006; Turchetto et al. 2007). They are sites of both upwelling and downwelling (Wåhlin 2002; Kämpf 2005), and canyon-induced changes to hydrodynamic conditions can result in increased biomass of the pelagos (Albaina & Irigoien 2007).
Canyon sediments can be richer in organics than the surrounding slope and contain denser deposits of fresh phytodetritus (Garcia et al. 2007), and accumulations of large plant detritus (seagrass) and algae (kelp) have been found inside canyons (Vetter & Dayton 1999). Given such enhancement of trophic resources and amplified currents – which should benefit suspension feeders – canyons are predicted to be favourable habitats for benthic consumers. Benthic biomass in canyons does not, however, exceed that of the abutting slope in all cases (Houston & Haedrich 1984), and faunal responses appear to differ between taxa and functional groups (Vetter & Dayton 1999).
Although dense and diverse communities of megabenthos are not uncommon in canyons, the specific faunal patterns depend on the balance between food imports, seafloor complexity, and habitat instability (Schlacher et al. 2007). Habitat instability due to sediment gravity flows and other geological events occurs naturally in canyons, but can also be triggered by fishing activities (Palanques et al. 2005; Martín et al. 2007). Thus, despite providing favourable conditions for benthic organisms, including organic-rich sediments, physical disturbance of the canyon seafloor – when frequent or severe – may result in lower benthic abundance, as has been demonstrated for meiofauna and foraminifera in such settings (Garcia et al. 2007; Koho et al. 2007). Habitat diversity and complexity is another influential driver of benthic biomass and diversity in canyons (Schlacher et al. 2007). Steep slopes, rocky outcrops, accumulations of boulders and other consolidated substrata increase seafloor heterogeneity and add habitat complexity to the otherwise relatively uniform continental slope sedimentary environment; demersal fishes and invertebrates preferentially occur around complex structures in greater abundance (Morais et al. 2007).
The topographic complexity of canyon seascapes enhances habitat heterogeneity at regional seascape scales, and greater abundance and diversity of fishes may therefore be associated with margins where canyons are common (Marques et al. 2005; Stevenson et al. 2008). Many canyons are commercially fished, but areas of highly rugged topography can provide some refuge from bottom-contact fishing at local scales (Yoklavich et al. 2000, 2002). Generally, canyons provide important fish habitats and critical nursery areas on continental margins (Stefanescu et al. 1994; Brodeur 2001), with some species preferentially spawning inside canyons (A. Williams, unpublished data).
Because canyons can influence the distribution of continental margin megabenthos, their locations are influencing the design of Australia’s National Representative System of Marine Protected Areas (NRSMPA). The first deepwater component of the NRSMPA was declared in July 2007 off South-eastern Australia, with the remainder to be completed by 2012. This imperative, and the fact that most of Australia’s continental margin remains unsampled for benthic biodiversity, has required that physical characteristics of the marine environment – including seabed features such as canyons – be used as surrogates for biological distributions to design the NRSMPA. Off South-eastern Australia, where numerous canyons incise the continental slope, marine reserves need to be differentiated to reflect the variety of benthic communities they support (Williams et al. 2009a).
New sampling tools such as multi-beam sonar (swath) mapping (Kloser et al. 2007) and quantitative photographic methods (Shortis et al. 2008) are increasing our ability to map biodiversity in the deepwater environment, and to validate the use of feature-scale surrogates for biological distributions. A focus for biodiversity mapping in Australian waters is the continental margin, particularly the upper slope (∼ 200–700 m), because in this zone high levels of fishing can overlap with rich and vulnerable biodiversity (Williams et al. 2009a). Targeted mapping of canyon megabenthos addresses a clear need for conservation planning because feature-scale habitats will remain drivers of the NRSMPA design, with canyons the single most prominent feature-scale habitat on the continental margin. In this paper we demonstrate the benefits of biodiversity mapping to conservation planning using data on the distributions of sponges from three canyons off the west coast of Tasmania.
Sponges are a widespread, common, abundant and diverse component of the marine benthos, especially where hard substrates are a major component of the seafloor. They occur in virtually every major subtidal habitat type, extending from shallow inshore regions over the continental shelf and margins to the deep-sea floor (Hooper & van Soest 2002; Becerro 2008; Pansini & Manconi 2008). Because sponges strongly modify many key structural and functional traits of benthic communities and ecosystems, they are generally regarded as a good model taxon to examine ecological patterns and processes (Bell 2007). Sponges can dominate the biomass of megabenthic assemblages (Ward et al. 2006) and, as a speciose group, contribute significantly to benthic biodiversity at all spatial scales (Janussen & Tendal 2007; Sorokin et al. 2007; Van Soest et al. 2007).
The functional roles of sponges are diverse, including benthic–pelagic coupling via filter feeding, nutrient recycling, modulation of competitive interactions, and provision of trophic resources (Bell 2008). One of the most significant ecological interactions of sponges is the creation of habitat and the modification of seafloor characteristics. Sponges, as structure-forming epibenthos, add and enhance structural habitat availability, complexity and quality (Beaulieu 2001b; Henkel & Pawlik 2005; Pirtle 2005; Tissot et al. 2006); this function is thought to be especially important in habitats of low topographical relief or where biogenic structures are otherwise rare (Barthel 1992; Beaulieu 2001a). This range of ecological characteristics marks sponges as a good candidate taxon for exploring strategies to conserve megabenthic biodiversity using photographic imagery.
Submarine canyons enhance seascape diversity at regional scales and provide unique habitat settings for diverse, abundant and fragile faunal assemblages on continental margins (Schlacher et al. 2007). These biodiversity and habitat values are under threat from bottom fishing, prompting a relatively recent emphasis to include canyons in conservation planning. Such conservation measures commonly use reserves (i.e. marine protected areas, MPAs) as the tool of choice (Harris 2007). Thus, the identification and spatial allocation of spatial conservation units requires data on the distribution of the ecological features to be protected.
Given the emerging and growing need for spatial information of biodiversity on continental margins in Australia, our primary objective in this paper is to document the diversity, distribution, and community structure of the sponge megabenthos in canyons. We document both bathymetric and geographic patterns of alpha-diversity, and assess beta-diversity in terms of differences in sponge community composition between canyons, between depth strata, and between types of megabenthic assemblage that sponges are associated with. Because conservation planning commonly uses both physical and biological surrogates to predict the distribution of biota, the performance and utility of environmental factors (e.g. depth, sediment properties) as well as simple biological measures (e.g. total bottom cover of sponges) were evaluated for their surrogacy performance in the canyon benthos. Finally, we model how different levels of information content influence the number and geographic placement of conservation areas needed to achieve a set conservation target. Thus, the utility and performance of megabenthic community data in the context of conservation planning for submarine canyons form the focal point of this paper.
Material and Methods
Field survey and data acquisition
Three submarine canyons incising the continental slope off the west coast of Tasmania were surveyed with a towed camera system (Fig. 1). The cruises formed part of fishery surveys in 2004 and 2005 that examined the types and distributions of shelf-edge habitats used by two commercial fisheries: bottom trawling for finfish and trapping for giant crabs Pseudocarcinus gigas (Williams et al. 2009b). Video and photo transects were directed down-slope to maximise the variety of habitats sampled along transect lines, and to document bathymetric changes in environmental characteristics and megabenthic assemblages.
The King Island canyon extends from the shelf edge at 200 m depth to > 1000 m, approximately in a north-easterly direction (Fig. 1A,B). The canyon is comparatively small, narrowest (1.5 km) at the 700 m isobath, and widening at its head to 3 km (300 m isobath). It is favoured for crab trapping, but has no history of demersal trawling. The photo transect in this canyon was taken down the north-north-eastern flank of the canyon head (∼ 200–450 m) at an acute angle to the channel axis (Fig. 1A,B).
The Ling Hole canyon extends from the shelf edge at 150 m to abyssal (> 2000 m) depths in a north-easterly direction (Fig. 1C,D). It is a relatively large canyon with its narrowest width (4.5 km) at the 700 m isobath, widening at its head to 6.1 km (200 m isobath). It is regularly fished by bottom trawlers. The photo transect in this canyon was taken along the south-eastern flank, parallel to the canyon axis, between 150 and 450 m depth (Fig. 1C,D).
The Pieman canyon extends from the shelf edge at ∼ 150 m to abyssal depths in a approximately north-easterly direction (Fig. 1E,F). It is a relatively large canyon, 5.0 km at the 700 m isobath, widening to 7.4 km at the 300 m isobath, and with narrower head of 4.6 km width at the 200 m isobath. It is fished seasonally by mid-water trawlers, targeting blue grenadier (Macruronus novezealandiae), but bottom contact by the trawls is believed to be minimal. The transects in this canyon were taken across the canyon head, near-perpendicular to the canyon axis. Two transects, located on opposite flanks of the canyon, were surveyed from 170 to 470 m, where ‘Pieman 1’ sampled the south-eastern flank and ‘Pieman 2’ the north-western flank.
The camera platform (Shortis et al. 2008) carried video cameras and a Canon 30D 8-megapixel digital stills camera that was mounted at a 50° angle on the camera platform. It was towed on a fibre-optic cable approximately 2 m off the seabed, with altitude regulated by hauling or paying out wire using a remote winch control while viewing a video-feed in real time. An ultra short base line (USBL) tracking beacon mounted on the camera system provided its position in relation to the vessel; telemetry data on platform pitch and roll data, combined with data from the vessel-mounted motion reference unit and differential GPS, enabled the position of the camera platform on the seabed to be estimated with an accuracy of approximately ± 5 m. Data for this paper were extracted from the digital still camera triggered manually by an on-board camera operator; images were collected at intervals of ∼ 25 s, with longer intervals over large patches of seabed with low structural complexity and sparse epibenthic fauna.
Still images provided very good details of seafloor features and epibenthic megafauna (Fig. 2). Our principal goal was to document the diversity, bottom cover and fine-scale spatial variation of sponges on the canyons’ seafloors. For this, we examined the entire set of 576 still images acquired along four transect lines in three canyons (Fig. 1).
Each image was first scored for the predominant substrate type, geomorphological attributes and megabenthic fauna types (Table 1, Fig. 2) using the classification of Williams et al. (2009a). Secondly, each distinct sponge specimen in the images was identified, using morphospecies as the operational taxonomic units (OTU). Species identification was done at 300% magnification.
Table 1. Categories used to visually classify seabed substrate, geomorphology, and megabenthic assemblages types in canyons.
Very fine sediment with silty appearance (< 0.1 mm)
Intermediate between mud and coarse sand (0.1–1 mm)
Coarse sediment with distinctly grainy appearance (1–4 mm)
Intermediate between coarse sand and cobble (4–60 mm)
Large, distinct rocks with clearly defined edges visible within the field of view (60 mm–3 m).
Consolidated hard sediments expanding beyond the field of view (> 3 m)
Seabed has a smooth, unstructured appearance
Seabed has distinct, regular ripples
Seabed has irregular small surface formations, such as burrow openings of bioturbating infauna, faunal traces, mounds or depressions
Small patches of rock appearing through sediments, mostly covered by sediments; no clear rock edges visible
outcrop, small (< 1 m)
Low rocky feature emerging from the surrounding sediments with edges or overhangs visible and smaller than 1 m
outcrop, large (> 1 m)
Low rocky feature emerging from the surrounding sediments with edges or overhangs visible and larger than 1 m
We assembled a library that contained images of the distinct morphospecies annotated with descriptions of key diagnostic features that could be visually distinguished. ‘Voucher images’ were added progressively to the library as new morphospecies were encountered. Prior to finalising the dataset, we critically examined this library and amalgamated any OTUs when there was uncertainty about the uniqueness of assigning them to a single morphospecies. In total, 65 morphospecies of sponges could be distinguished. Although we have taxonomic data on the composition of the sponge fauna in the canyons (Schlacher et al. 2007), actual species names could not be assigned with confidence to specimens from images.
Data analysis and modelling
The spatial scale at which alpha-diversity is defined varies considerably in marine benthos research. Here we used the number of positively identified morphospecies (OTUs) per image as the principal measure of alpha-diversity for all analyses. This spatial scale is conceptually analogous to ‘point’ samples or ‘point richness’ based on individual grab samples for soft-sediment assemblages (Gray 2000). Because the bathymetric and geographic distribution of biodiversity is often a key input to conservation planning and it is usually measured as species richness per unit area, we analysed changes in species richness with depth. Relationships between sample depth and sponge species richness per image were examined with non-linear regression models, run separately for each of the major macrobenthic assemblage types to avoid possible confounding of depth patterns across different assemblages.
Beta-diversity of sponge assemblages was examined using similarities in species composition expressed as Bray–Curtis (B–C) resemblance coefficients (identical to the Sorensen or Dice coefficient for presence/absence data). Because communities on continental margins often show zonation with depth (Carney 2005) and can differ geographically between seascape features (Schlacher et al. 2007), we tested hypotheses about differences in community composition according to two a priori defined factors: (i) geographic position of samples in different canyons (canyon effect), and (ii) depth strata (bathymetric effect). In addition, we also tested whether distinct megabenthic assemblage types that can be distinguished macroscopically (Fig. 2, Table 1) harbour sponge assemblages of distinct species composition (assemblage effect). All categorical tests on beta-diversity used analysis of similarities, ANOSIM (Clarke 1993). ANOSIM tests for the effects of depth and megabenthic assemblage types were run separately for each canyon to remove possible confounding by geographic location. Furthermore, each analysis for a main effect included the second factor as a crossed factor to account for possible confounding of depth with megabenthic fauna types and vice versa.
The degree to which spatial variation in environmental variables matches corresponding faunal patterns of sponges was examined with the RELATE procedure (Clarke & Ainsworth 1993). Environmental variables used in this analysis included depth (100 m strata), substrate type, geomorphology and megabenthic assemblage types. The main objective of this analysis was to assess whether spatial heterogeneity in sponge communities can be broadly predicted from abiotic surrogates (depth, sediments, relief), biotic surrogates in the form of broad megabenthic assemblage types, or a combination of both. Consequently, all possible combinations of abiotic and biotic predictors (surrogates) were examined, and their match to biological patterns assessed based on matrix rank correlation using the RELATE routine implemented in the PRIMER software package (Clarke & Gorley 2006). This analysis was done for biological patterns derived from species composition (i.e. traditional community-level data) as well as biological patterns derived from coarser, aggregate measures of species richness and bottom cover.
Conservation value assessment
To assess how different levels of information content influenced the selection of sites for conservation planning, we modelled the outcomes of site prioritisation under five likely scenarios for operational biodiversity conservation planning on the deep continental margin. These simple simulation models used a five-step sequence of decreasing information content (Table 2) as follows:
Table 2. Attributes of modelled strategies for selecting spatial units to maximise the sponge species richness in conservation planning for the canyon benthos.
randomisation and selection procedure
1 – random
Baseline to compare performance of alternative strategies.
2 – depth gradient
Bathymetry – position of samples
Random selection of one representative each for every 100 m depth band.
Because species have limited occurrence across the full bathymetric gradient, inclusion of all depth bands should increase species representation for the spatial configuration selected.
Repeat step 1 until all units have been used.
3 – fauna coverage
Visual identification of dominant fauna types from images
Random selection of one representative unit each for every megabenthic assemblage type.
Because species are associated with specific megabenthic assemblages, inclusion of all assemblage types is predicted to increase species representation in the final selection of conservation units.
Repeat step 1 until all units have been used.
4 – sponge cover
Percentage bottom cover of sponges
Rank units by bottom cover and select units from highest to lowest cover.
Because bottom cover of sponges is an adequate proxy for diversity, preferential selection of units with high sponge cover is predicted to increase biodiversity for the spatial configuration selected.
Randomise for all tied ranks and repeat step 1.
5 – sponge species richness
Rank units by species richness and select units from highest to lowest richness.
Species richness of sponges per sample unit is a direct measure of biodiversity. Therefore, preferential selection of units rich in species is predicted to maximise biodiversity for the spatial configuration selected.
Randomise for all tied ranks and repeat step 1.
1Random. No information on the spatial distribution of biological resources is available, but conservation areas still need be allocated for canyons. Essentially, this scenario represents planning in a data-free situation, assumes no knowledge of environmental factors that may influence the distribution of biological resources (e.g. depth), and therefore ‘selects’ spatial units randomly. Consequently, it serves as the null model against which the performance of alternative strategies is compared.
2Depth gradient. The bathymetry of margin habitats and potential candidate sites is available (e.g. from swath mapping or other surveys), but data on the spatial distribution of the biota are lacking. The basic rationale for using depth strata to inform the spatial selection process rests on the assumption that many species do not occur across the full depth gradient of the continental margin (Carney 2005). Therefore, by systematically including a range of depth zones, the number of species contained in the final selection is predicted to be greater than that from purely random selection.
3Megafauna coverage. Spatial data on the broad types of megafaunal assemblages (e.g. ophiurid beds, bryozoan thickets) that are distributed over the candidate area have been obtained (e.g. visual classification from video transects and/or still images), but without species identification of the benthos due to financial or other constraints. Assuming that different species are associated with different megabenthic assemblage types (sensuGutt & Starmans 1998), selecting sites to achieve equal representation of all recognised megabenthic assemblage types is thought to improve the chance of including a greater fraction of the total regional species pool.
4Cover. Aggregate, quantitative measures of communities have been obtained and are geo-referenced, but the level of analysis did not include analysis of species identities; this situation commonly arises when high labour costs for species-level taxonomic work cannot be accommodated, taxonomic expertise is not available, or the fauna is largely undescribed. Here we simulated the outcomes of site prioritisation when total bottom cover of sponges is known but species identifications have not been performed. The use of aggregate measures (e.g. bottom cover, total biomass) to prioritise sites in terms of diversity assumes that species diversity is monotonically related to the aggregate measure (sensuGoldberg et al. 2006) and therefore can serve as a surrogate for detailed species-level data. In the present situation, preferentially selecting samples with high sponge cover is predicted to maximise species richness in any selection if high bottom cover is an adequate predictor of high species richness.
5Species richness. The principal objective of the great majority of marine protected areas is to conserve biodiversity (Lubchenco et al. 2003). Species richness is, arguably, the least ambiguous metric for biodiversity (Gaston 2000), and is widely used in terrestrial (Orme et al. 2005) and marine (Roberts et al. 2002) conservation settings. Thus, in concert with rates of endemism and threats, maximising the protection of species richness is at the centre of most conservation efforts (Myers et al. 2000). Consequently, the final scenario represents the most data-rich situation where species richness has been quantified for all geo-referenced sites. Maximising species richness for a conservation area then simply adds candidate sites in decreasing order of observed richness until the specified conservation target is met.
For all five scenarios above we set conservation targets to represent 50%, 75% and 90% of the regional species pool of sponges in protected areas. Because there are no complete inventories of sponges for continental margins in this region, we used a purely operational definition of the ‘regional species pool’ as being the total of sponge species records obtained from all samples in all canyons in this study. Under each of the five modelled scenarios above, images were selected according to the criteria defined in Table 2. As practical spatial conservation units we chose 100-m depth bands, and these units were included in the final selection if they overlapped with geo- and depth-referenced image samples selected under the criteria of each scenario (Table 2). Because of tied ranks or multiple permutations of depth bands or community types in the selection process, there is often no unique solution for simulation runs under the five models. Therefore, differences in the performance of strategies can be evaluated by comparing the means of modelled conservation effort (in terms of spatial units required) from repeated model runs using ANOVA followed by Student–Newman–Keuls (SNK) post-hoc tests (Underwood 1981).
Photo transects covered a depth range from 148 to 472 m, with a mean depth of 300 m (Table 3). Images tended to be from slightly greater depths in the Ling Hole canyon, but overall the sampled depth ranges were broadly comparable amongst canyons. Importantly, the distribution of analysed images (Fig. 3) showed no monotonic increase or decrease in sample coverage with depth over the full depth gradient sampled, either for all transects combined (rs = 0.20; P = 0.43) or for individual transects (King Island: rs = 0.05, P = 0.85; Pieman 1: rs = −0.18, P = 0.54; Pieman 2: rs = −0.05, P = 0.84; Ling Hole: rs = 0.31, P = 0.29).
Table 3. Environmental attributes of canyons on the island margin of Tasmania (Australia) surveyed for epibenthic megafauna using video and high-resolution photo transects.
Two transects located on opposite flanks were surveyed in the Pieman canyon (cf.Fig. 1).
number of images analysed
distance of camera tow along transect (m)
depth coverage of images
range (minimum–maximum) (m)
quartile range (Q25–Q75) (m)
mean ± SD (m)
311 ± 82
349 ± 95
294 ± 96
287 ± 96
300 ± 94
fine sand (%)
coarse sand (%)
highly irregular (%)
outcrop < 1 m (%)
outcrop > 1 m (%)
The substrate was dominated by unconsolidated sediments that constituted 94–98% of all image records across the three canyons (Table 3). The substrate of King Island and Pieman canyons was dominated by coarse sand that was observed in 94–98% of all images taken in these two canyons, whereas Ling Hole canyon substrates were predominantly mud and fine sand (83% and 14% of images, respectively; Table 3). Larger cobbles and boulders were uncommon (0–2%), as were hard, rocky bottoms, which were present in only 2–4% of the images analysed in all canyons (Table 3). Most (70–89%) of the seafloor in all canyons showed a highly irregular surface topography. Flat areas without visible rippling or other relief were only regularly observed in the King Island canyon (27%), and occurred much less frequently in the other canyons (0–9%). Evidence of surface disturbance linked to wave energy could only be observed in 8% of records in the Ling Hole canyon. Similarly, outcrops were uncommon (0–3%) in all canyons (Table 3).
A total of 65 sponge morphospecies were identified from the 576 images analysed. The distribution of morphospecies amongst images (i.e. frequency of occurrence) was highly right-skewed, with one-third of morphospecies (n = 20) having been recorded in no more than three images, and 10 morphospecies (15%) occurring in a single image only. Five morphospecies each were recorded in two and three images. Half of all morphospecies occurred in 10 images or less, which represents 1.7% of the total sampling effort, the maximum range being 194 images (33% of total coverage) for a single sponge morphospecies. At the canyon scale, 22 morphospecies (34%) were recorded from a single canyon, 32 morphospecies (49%) were distributed in two canyons, and only 11 morphospecies (17%) occurred in all three canyons.
Most morphospecies occurred in a broad depth band between 200 and 350 m (Fig. 3). Upper and lower bathymetric boundaries did not differ in terms of their relative variability (Levene’s test of coefficients of variation of boundaries, t(2,105) = 0.32, P = 0.75), indicating that sampling did not introduce a directional, asymmetrical bias for estimates of either the shallow or deep distributional limits of sponges in canyons. Only 5% of morphospecies occupied at least 80% of the sampled depth range, 38% of species occurred across half of the depth gradient sampled, whereas three-quarters of species had bathymetric records that represented 20% or less of the full depth gradient sampled. Although depth ranges increased exponentially with site occupancy, they were significantly narrower (difference in Akaike Information Criterion = 242, P < 0.001; F(4,347) = 89, P < 0.001) than would be expected by chance, assuming an equal probability for each species to occur at any depth within the limits of sampling (Fig. 4). Thus, although common sponge morphospecies generally did not occupy very narrow depth ranges, their bathymetric ambit was nevertheless compressed.
Bathymetric patterns of species richness and bottom cover of sponges generally did not show a monotonic change with depth and were strongly influenced by the type of megabenthic assemblage type that sponges were associated with (Fig. 5). Richness and cover in images characterised by low, attached fauna of mixed taxonomic composition showed a unimodal peak at around 300 m, straddled by lower values at shallower and greater depths (Fig. 5A,F). A similar pattern was evident for sponge richness and cover associated with bryozoan thickets and turfs (not recorded from the Ling Hole canyon), complemented by a second peak at the shallowest depths sampled (Fig. 5B,G). No distinct bathymetric patterns in sponge richness and cover were found for images dominated by bioturbators, but many of these images contained a few morphospecies which covered small areas of the seabed (Fig. 5C,H). Images that had abundant brittle stars and encrusting fauna (only recorded in the Pieman canyon) tended to show an increase in sponge richness and cover from the shallowest depths to ∼ 275 m, which coincides with the lower distributional limit of megafauna communities dominated by ophiurids (Fig. 5D,E,I,J).
The close concordance in bathymetric patterns between sponge richness and cover results from the strong correlation between the seafloor area covered by sponges and the number of species that contribute to this bottom cover. Species richness was found to be closely linked to total bottom cover by sponges in each transect, as well as for data pooled across transects and canyons (rs = 0.97, P < 0.001; Fig. 6). Thus, sponge cover is a good predictor (R2 = 82–97%) of species richness for sponges in the canyons studied.
The presence of attached megabenthos appeared to promote the co-occurrence of sponges on the canyons’ seafloors, but effects on species richness varied amongst assemblage types. Species richness of sponges was substantially elevated in images dominated by bryozoan thickets and turfs, and was lowest in soft-sediment areas characterised by bioturbators (Fig. 7); other megafaunal assemblages supported levels of sponge richness intermediate between bryozoans and bioturbators (Fig. 7). Fewer sponge morphospecies were recorded in images with abundant brittle stars (‘ophiuroid beds’) but few other sessile fauna. However, where ophiuroids co-occurred with other attached fauna, sponge richness was comparable to images without abundant brittle stars. Thus, it appears that the presence of brittle stars per se may not be inimical to the development of rich and dense sponge assemblages in canyons. Rather, sponge assemblages became more diverse and abundant in areas where encrusting fauna, such as bryozoans, stabilised the seafloor. Thus, in situations where different types of megabenthic assemblage occur in different canyons, or where assemblages have restricted geographic or bathymetric distributions (e.g. ophiuroid-dominated assemblages only found in the Pieman canyon), differences in sponge species richness between megabenthic assemblages are likely to translate into spatial differences in benthic diversity overall.
Beta-diversity and surrogates
Community composition of sponges was only weakly structured by geographic separation of sites located in different canyons. Although we found some partition between canyons based on differences in sponge species composition, beta-diversity was small when the possible confounding factors of differences in depth (two-way ANOSIM with crossed factor depth: R = 0.048, P = 0.06) and megabenthic assemblage types (two-way ANOSIM with crossed factor megabenthos: R = 0.272, P < 0.01) were accounted for. Similarly, there was only a weak influence of depth on species composition across any transect (King Island: R = 0.102, P = 0.03; Ling Hole: R = −0.019, P = 0.47; Pieman 1: R = 0.194, P < 0.01; Pieman 2: R = 0.161, P < 0.01) that was independent of differences in megabenthos types. In contrast to the generally weak structuring effects of depth and geographic affinity (i.e. canyon-to-canyon), sponge assemblage composition differed significantly and more strongly amongst the megabenthic assemblage types associated with sponges (King Island: R = 0.567, P < 0.01; Ling Hole: R = 0.669, P < 0.01; Pieman 1: R = 0.573, P < 0.01; Pieman 2: R = 0.373, P < 0.01).
Spatial patterns derived from biological information on sponges (i.e. species identity, richness and cover) were substantially different from those based on environmental information (i.e. bottom type, depth, geomorphology) or combinations of environmental data and megabenthic assemblage types (Table 4). Thus, environmental variables were found to be poor predictors of the spatial structure evident in sponge assemblages. This poor match between environmental characteristics and biological features was pronounced irrespective of the level of detail at which the biological information was analysed. Abiotic data showed a low degree of spatial concordance with either compound community measures (i.e. sponge richness and bottom cover) or with community composition based on species identities (Table 4). By contrast, megabenthic assemblage types associated with sponges were more strongly correlated with biological sponge data, but the addition of abiotic data weakened this concordance (Table 4). Remarkably, spatial patterns derived from measurements of richness and cover (excluding detailed species identification) are highly correlated with patterns based on species composition (ρ = 0.872, P < 0.001). Thus, abiotic environmental data performed poorly as surrogates for biological patterns of sponges in canyons, whereas compound measures of community characteristics (species richness, total bottom cover) were good proxies of community patterns derived from species identities (Fig. 8).
Table 4. Congruence between spatial patterns of sponges and abiotic surrogates (depth, geomorphology – Geo., substrate type – Subs.), biotic proxies (megabenthic assemblage types associated with sponges), or both.
megabenthic assemblage typesc
A – species composition
B – species richness & cover
Congruence is tested by inter-matrix correlations of similarity matrices reflecting spatial patterns of single predictors or combinations of surrogates against matching matrices based on either species composition of sponges (A) or species richness and bottom cover of sponges [B; bold entries denote ρ values at P < 0.001 (***); all others are at P > 0.05 (n.s.)].
a Geomorphology (Geo.) = flat/unrippled, rippled, irregular, subcrop, outcrop < 1 m, outcrop > 1 m.
b Substrate type (Subs.) = mud, fine sand, coarse sand, cobble/boulder, rock.
Conservation effort, measured in terms of the number of spatial units (defined as a 100-m depth band in any of the three canyons) required to achieve a specified conservation target of species richness, was significantly smaller for strategies that employ biological information of the target taxon than for strategies based on representation of bathymetric strata or megabenthic assemblage types, which were more costly. For example, strategies that employ random assignments of sites to potential conservation units were four times less economical when 50% of species are the target to be included in conservation areas (Table 5). Importantly, selection strategies that used information on bathymetry or megabenthic assemblage types associated with sponges did not, in terms of spatial economy, perform better than random site selection: both required between twice and four times more spatial units than when selections were made based on sponge cover and richness data (Table 5). Because the simple, compound measure of total sponge cover was a good predictor of species richness (Fig. 6) there was no significant difference in the required spatial conservation effort, at a target of 50% of the regional species pool, between strategies based only on sponge cover or those based on actual species counts (Table 5). At conservation targets of 75% and 90% of species, strategies based on species counts and bottom cover were statistically significantly different, but contrasts were small at one to two spatial units (Table 5).
Table 5. Number of spatial units required to capture a given conservation target (i.e. 50%, 75% or 90% of the regional species pool) under five different strategies used for site selection.
target: 50% species
target: 75% species
target: 90% species
mean ± SEM
mean ± SEM
mean ± SEM
A ‘spatial unit’ is here defined as a 100-m depth band in any of the three canyons surveyed. Tabulated values are means and standard errors (SEM) from 10 simulations, with mean values among strategies compared by analysis of variance.
Superscripts (a–d) denote homogeneous groups, as defined by SNK (Student–Newman–Keuls) tests, amongst strategies within each conservation target.
***P < 0.001.
12 ± 0.58
16 ± 0.40
18 ± 0.40
13 ± 0.33
17 ± 0.35
19 ± 0.37
megabenthic assemblage coverage
12 ± 0.47
16 ± 0.61
20 ± 0.42
sponge cover ranked
3 ± 0.15
6 ± 0.15
8 ± 0.10
sponge richness ranked
3 ± 0.17
4 ± 0.00
9 ± 0.27
Spatial units that were prioritised by the best performing strategy (i.e. selection method that resulted in the smallest number of spatial units required to achieve a set conservation target), differed significantly in terms of the relative frequency of macrobenthic assemblage types which they contained (Table 6). Bryozoan turfs and thickets were massively over-represented (84–95%) in sample selections compared with their much lower proportional representation in the available sample pool (27%, Table 6). Thus, the optimal prioritisation strategy for sponge species richness selected nearly exclusively sites where sponges occurred in bryozoan thickets and turfs (Table 6). Conversely, despite constituting 30% of the seafloor in the canyons, sites dominated by bioturbators were never selected for inclusion in conservation areas to represent sponge species richness at a specified target level (Table 6). Similarly, areas characterised by low/encrusting fauna and ophiuroids were moderately abundant in the benthos of the canyons (10–15% frequency of occurrence) but were selected only in very low numbers (2–4% of sites) at the highest conservation target set, which represents 90% of sponge species (Table 6). Ophiuroid beds were included in site selections in proportions (16%) roughly equal to their occurrence in the canyons (18%) when half of sponge species were targeted to be represented in conservation areas; more ambitious targets (i.e. 75% and 90% of sponge species) included significantly fewer sites characterised by brittle stars (Table 6).
Table 6. Proportions of megabenthic assemblage types in selections of spatial units to achieve three conservation targets (50%, 75%, 90% of regional species pool) under a selection strategy based on ranked species richness per sample, compared with proportion of each assemblage type in the available pool.
megabenthic assemblage types
available in sample pool
***P < 0.001; **P < 0.01; n.s. P > 0.05 for difference of proportions test using Bonferroni corrections; values in bold denote significantly higher proportions in the selection.
ophiuroids and encrusting fauna
As expected, prioritisation strategies that use biological information on species richness were clearly superior in terms of the number of sponge species included for a given spatial effort (Fig. 9). However, our simulations indicated that the use of sponge bottom cover (i.e. excluding data on species diversity) resulted in only marginally lower species representation at spatial ambits of 50 samples or more. There was no difference in species accumulation rates between a complete random selection of samples and strategies that sought to achieve equitable representation of depth bands or megafaunal assemblage types (Fig. 9).
Not only did prioritisation strategies that excluded data on sponge species richness and cover result in substantially higher costs (i.e. number of spatial units to be included for a given conservation target), but the spatial arrangement of identified conservation areas was substantially different. For example, for selections that would protect 75% of species richness, strategies excluding biological information would require 176 km2 of conservation areas across all three canyons that encompass the full depth gradient sampled (Fig. 10). By contrast, when data on sponge cover are available and are used for site prioritisation, only 27.2 km2 of seafloor in two canyons are required, representing a saving of 149 km2, or 85% (Fig. 10). Moreover, no areas deeper than 350 m would require inclusion in conservation areas (Fig. 10). Finally, when sponge species richness data are available and form the input for spatial prioritisation, only 16.9 km2 of seafloor were identified as potential conservation areas to achieve representation of 75% of the regional species pool; in this situation a depth band between 200 and 350 m located in the Pieman and King Island canyons appears to be sufficient to include three-quarters of the regional sponge species pool inside potential conservation areas (Fig. 10).
Sponges are also sensitive to the impacts of bottom-contact fishing (Auster & Langton 1999; Freese et al. 1999). They can therefore serve as biological indicators of human disturbance on the canyon benthos. Observed reductions or marked spatial contrasts in sponge cover, biomass or diversity – as observed in some areas of the present study – could thus be interpreted to result from the impacts of fishing. However, such biological changes may equally reflect habitat conditions that are unsuitable for the establishment and persistence of sponge assemblages, irrespective of whether putative fishing impacts occurred.
Performance of surrogates
The use of surrogates is frequently advocated in conservation planning as a more tractable, easier or cheaper alternative to traditional methods of data acquisition at the species level (Sarkar et al. 2006). Abiotic surrogates, which commonly include geomorphic features, bathymetry, sediment composition, seabed mobility and exposure, have become an important input to the planning process for protecting marine resources and biodiversity in Australia (Jordan et al. 2005; Harris 2007) and elsewhere (Pickrill & Todd 2003).
Surrogates are particularly relevant in conservation planning when biota have difficult or intractable taxonomy, where the cost of species-level resolution cannot be met, or where the timetable and geographical scope of conservation planning provide insufficient time to achieve species-level data collections. Because all these factors apply to deepwater reserves on the Australian continental margin, we examined whether environmental attributes can predict biological patterns and whether compound measures of sponge communities (i.e. total cover) are adequate surrogates for species-level data in the context of conservation planning.
We found that environmental factors (bottom type, depth, geomorphology) were poor predictors of the species composition, richness and cover of sponges in canyons. In fact, much of the canyon seafloor is remarkably uniform, consisting of unconsolidated sediment of very low relief (Fig. 11). Yet this apparent environmental homogeneity (sensuStevens & Connolly 2004) does not translate into a biologically uniform benthos (Fig. 11). This contrasts to some extent with the reported concordance between spatial patterns of physical factors (i.e. depth, exposure, % mud, % gravel) and the distribution of megabenthic communities on the Northern Australian shelf, where moderately strong correlations (r = 0.60) between the distributions of physical and biological data were found.
Ideally, meaningful and robust abiotic surrogates can reliably predict biological information. Poor performance of abiotic surrogates is likely when the key environmental factors that determine species distributions are poorly defined, or when relationships between physical and biological data are derived from investigations at spatial scales much smaller than their application in regional planning (Post 2008). Physical surrogates may also be ineffectual when key properties of the environment that are instrumental in structuring the biota are not measured or included in the analysis, or when biological interactions (e.g. competition, trophic amensalism, recruitment variability, etc.) override environmental factors (Constable 1999). In these situations, biological assemblages will differ between areas that have apparently similar environmental conditions: a case of ‘false homogeneity’ in the use of abiotic surrogates (sensuStevens & Connolly 2004) which applies to the sponge assemblages in the canyons.
Sponge distributions were predicted much more successfully when co-occurring megafaunal types were used as surrogates. However, although the correlations between sponge community structure and associated megafauna types were significant, the absolute values of coefficients were small. Conversely, spatial patterns of sponge cover (that did not require species identification) were highly correlated with richness patterns based on species composition and can therefore be regarded as a good biological surrogate in this taxon. Accordingly, there was no significant difference in the required conservation effort between strategies based only on sponge cover or those based on actual species counts. The implication of this finding is that the cost of research to inform the conservation process could be reduced using a strategy based on sponge cover (relatively low cost of data collection) as a surrogate for actual species counts (relatively high cost).
Virtually all sponge species recorded in our study were consistently associated with bryozoans; only four species – each having a single occurrence – were found to be associated with other megafaunal types. Thus, it appears that the distribution of bryozoan thickets and turfs on the upper continental slope may provide a useful surrogate for the distribution of sponges. Erect and rooted forms of bryozoans stabilise the sediments as their calcareous skeletal material accumulates on the seafloor (Hageman et al. 2000). We postulate that the mechanism underlying the close match between the distribution of sponges and bryozoans is that bryozoans provide attachment sites for sponges. The potential to use bryozoan distribution as a surrogate for other taxa has practical applications across temperate Australia, because the distribution of bryozoa – at a broad community level – is better known compared with other megabenthic taxa (Wass & Yoo 1983; Bone & James 1993; Aminia et al. 2004).
We recorded virtually no bryozoan turfs and thickets in the Ling Hole canyon, where sponge richness and diversity was much lower. Fishing with bottom-contact gear is common in this canyon, where virtually all of the canyon floor receives some level of bottom trawling annually. Images and catch data show that bottom trawl gear has the potential to completely remove the megabenthos characterised by bryozoans, including sponges (A. Williams, personal observation). Thus, bottom trawling is a plausible explanation for the virtual absence of bryozoan-dominated megabenthic communities at this location.
In contrast to the richness and comparatively high cover of sponges when bryozoans structure the seafloor, much smaller numbers of sponges, at greatly reduced bottom cover, were recorded in areas dominated by ophiuroids and bioturbators, possibly a consequence of a trophic-group amensalism effect (Snelgrove & Butman 1994). Although sponges can be common in unconsolidated substrates on margins and in the deep sea (Janussen & Tendal 2007), many species prefer hard substrates as attachment sites (Hooper & van Soest 2002). Thus, in areas dominated by bioturbators the generally softer sediment is likely to be an inferior habitat for sponges.
We considered whether the poor performance of environmental surrogates in this situation could be related to the homogeneity of the habitats surveyed. These canyons, when compared to some others in this region (Schlacher et al. 2007), appear to have relatively uniform substrata of unconsolidated sediments where hard bottoms of higher relief such as rocky outcrops and boulders are rare. Finer resolution of sediment properties could, theoretically, be achieved by physical sampling and grain size analysis, possibly using image-based techniques to derive sediment size statistics (Butler et al. 2001). However, an accurate spatial match between biological data (derived from images) and sedimentary features at a fine scale is logistically extremely challenging or impossible for physical samples (i.e. grabs, cores) given the large number of images analysed and the inability to accurately locate physical sample at slope depths. Acoustically derived information on seabed properties has the potential to provide additional surrogates for the distribution of the deep megabenthos. Multi-beam sonar provides backscatter (seabed hardness) and slope (Kloser et al. 2007; Schlacher et al. 2007), and sub-bottom profiling shows the substrata that underlie surficial sediments (Kirkwood et al. 2005) – important where sediment veneers overlie consolidated substrata used as attachment sites by megabenthos such as sponges. However, the spatial resolution of acoustic data in continental slope depths (typically gridded in ≥ 20 m cells) is coarse compared with the spatial resolution of the biological data obtained here.
Practical implications for conserving megabenthos in canyons
We showed that a reserve design strategy which employs information on cover and richness of the target taxon (sponges) provided a large (up to 90%) dividend in spatial conservation effort by dramatically reducing the number of spatial units required to achieve a specified conservation target of species richness. Strategies that employed random assignments of sites to potential conservation units, or used bathymetry or co-occurring megabenthos types as surrogates, were much less economical.
The spatial arrangement of identified conservation areas was substantially different between strategies, with those excluding biological information requiring areas in more canyons over a much greater depth range and area (Fig. 10). Data on sponge cover or richness enabled any given conservation target to be met with a smaller spatial subset. For example, using actual richness data on sponges to achieve representation of 75% of the regional species pool required only 16.9 km2 in a depth band between 200 and 350 m located in the Pieman and King Island canyons, compared with 176.2 km2 when no biological information is used to select potential conservation area. Although these findings clearly demonstrate the efficacy of using diversity and distribution data of sponges in a conservation planning context, they should, ideally, be corroborated with other megabenthic taxa. Sponges are often an appropriate proxy for the megabenthos on continental margins based on their diversity (Schlacher et al. 2007) and prominent functional (Bell 2008) and structural (Tissot et al. 2006) roles and have apparent utility in conservation planning (this paper). This may, however, not be the case for all canyons, particularly those where very fine sediments or bottom instability are inimical to the establishment of rich sponge communities. Thus, the choice of model taxon as biological surrogates in conservation planning is dependent on local and regional habitat characteristics and dominant assemblages.
It is also the case that strategies which use other criteria (e.g. community composition, species abundances or biomass), may require additional or complementary sets of conservation areas. Nonetheless, in a location such as Western Tasmania where diverse and vulnerable biodiversity exists adjacent to commercial fishing grounds, these results indicate how biologically informed reserve design in the deep sea can improve the efficacy of selecting and allocating target areas to conservation units. Importantly, these approaches provide an option to lower the socio-economic impact of new reserves. In this specific case, conserving a narrow depth zone in two canyons would incur a small direct cost to the trawl and longline fisheries working the sediment plains in deeper water. Because minimising the socio-economic impacts is a stated management goal for deepwater reserves in Australia (Anonymous 2003), there is a great potential to expand the concepts of biological surrogacy and design strategies, utilising different levels of information, to features other than canyons and in other regions.