The Bay of Biscay is part of the North-East Atlantic Ocean, located west of France and north of Spain. It is a passive margin containing three parts; the Celtic, Armorican and Aquitaine margins (Zaragosi et al., 2000). Within this study, the Bay of Biscay was defined using geological boundaries (see Bourillet et al., 2006) therefore including all three margins, as the Celtic margin is part of the Celtic Sea according to hydrographical terms.

The morphology of the Celtic and the Armorican margins (Figure 1), limited by the Goban Spur in the north and the Conti spur in the south, is characterized by spurs and canyons (Bourillet et al., 2006). The continental shelf is wide, up to 200 km for the Armorican margin and more than 250 km for the Celtic margin (Bourillet et al., 2006). These two margins can be divided into three zones based on the geomorphology of the continental margins, the influences of sedimentation and hydrodynamic regimes (Figure 1): (i) the Celtic and northern Armorican margin, (ii) the central Armorican margin, and (iii) the southern Armorican margin.

The morphology of the first zone—the Celtic and northern Armorican margins—is complex and comprises some shelf-incising submarine canyons (Figure 1). These canyons contain several morphological features, such as cliffs, mainly formed in the head of the canyons by regressive erosion from the bottom to the top of the canyon (Bourillet et al., 2010). There are three main deep-sea drainage systems: (i) the Petite Sole drainage basin on the Celtic margin, comprising the canyons from Sorlingues to Hermine, (ii) the Chapelle drainage basin, comprising the canyons from Blackmud to Guilcher, (Bourillet and Lericolais, 2003) and, (iii) the West Brittany drainage basin, comprising the canyons from Brest to Douarnenez (Zaragosi et al., 2000, 2001). Zone 1 is under the influence of tidal currents of the English Channel (Zaragosi et al., 2001) and is linked with the drainage basins of rivers, such as the Seine, via the Channel paleo-river (Bourillet et al., 2003).

The second zone is the central part of the Armorican margin (Figure 1). This zone includes an alternation of large and narrow canyons. The large canyons, considered to be formed first, cut the continental shelf, while the heads of the narrow canyons, formed during a later stage, are located halfway down the slope (Bourillet et al., 2010). Two drainage systems are found within this part of the Armorican shelf: (i) the South Brittany drainage system, including the canyons from Audierne, south of Douarnenez Canyon, to Blavet Canyons (Zaragosi et al., 2001), and (ii) the Gascogne drainage system from the Belle-île to Yeu Canyons (also including Croisic, Saint-Nazaire and Pornic Canyons) (Bourillet et al., 2006). These canyons are not under the influence of the English Channel any longer and sediment is transported from the continental shelf to these canyons (Zaragosi et al., 2001).

The third zone comprises the southern part of the Armorican margin (Figure 1). Even though canyons still reach the continental shelf in this part of the Bay of Biscay, they are smoother than those on the Celtic and northern Armorican margins. The flanks of the canyons in this zone are regular and sedimentary, while the thalwegs are continuous with sloping banks. Cliffs are either scarce or not present in this zone (Bourillet et al., 2010). The Rochebonne drainage basin, comprising the canyons from Sables d'Olonnes, north of Rochebonne Canyon, to Oleron Canyon, north of the Conti Spur, is the only drainage system within this zone (Bourillet et al., 2006) and includes Rochebonne and Ars Canyons.

The third margin of the Bay of Biscay, the Aquitaine margin, from the Conti Spur to Capbreton Canyon, can be considered as a fourth zone of the Bay of Biscay (Figure 1). The continental shelf is narrow, with only 70 km from the shore (Bourillet et al., 2006). The continental slope of this margin is a “tectonic-dominated” margin instead of a “canyon-dominated” slope (Bourillet et al., 2006). The dominant relief, the Landes plateau, is surrounded by two majors canyons: Cap-Ferret and Capbreton. The “gouf de Capbreton” is an exceptional example of a shelf-incising canyon with a head close to the beach and a very gentle along-slope profile (Cirac et al., 2001; Bourillet et al., 2007). The flanks of the canyons on this margin have a weak slope and they do not present steps or cliffs. The Cap-Ferret and Arcachon thalwegs are broader than those of other canyons, north of this zone/margin (Bourillet, 2010; De Chambure et al., 2013).

In the Bay of Biscay several water masses with different origins and densities can be found (de Madron et al., 1999; van Aken, 2000a,b): (i) the Eastern North Atlantic Water (ENAW), (ii) the Mediterranean Outflow Water (MOW), (iii) the Labrador Sea Water (LSW), and (iv) the Northeast Atlantic Deep Water (NEADW). The ENAW is usually found between 200 and 600 m water depth. It originates from the Labrador Current and may contain a significant amount of Antarctic Intermediate Water (AAIW) transported to the NE Atlantic by the Gulf stream-North Atlantic Current. The more saline and denser MOW is deeper and generally flows between 700 and 1,300 m water depth. This water mass is more pronounced in the south of the Bay of Biscay and becomes less noticeable in the north of the Bay of Biscay toward Porcupine Seabight. The third water mass is the LSW and reaches to ~2,000 m depth. It is usually mixed with the MOW by internal waves near the continental slope and therefore this water mass is less apparent. The fourth and last layer is the NEADW originating from multiple water masses, including MOW and LSW. It occurs approximately between 2,000 and 2,600 m water depth.

There are several currents and other hydrological processes occurring in the Bay of Biscay that modify the temperature, salinity and density of the water masses. The slope current runs along the continental slope of the Bay of Biscay. It arrives in this basin via the Portuguese and Spanish continental slopes (Pingree and Le Cann, 1989, 1992; Koutsikopoulos and Le Cann, 1996) and is thought to be a density driven current formed by the MOW (Pingree and Le Cann, 1989). Even though the slope current is influenced by the complex morphology of the continental slope, it moves dominantly northwards/polewards (Pingree and Le Cann, 1989, 1992; Koutsikopoulos and Le Cann, 1996).

Other important processes in canyons of the Bay of Biscay are tidal currents and internal waves (Koutsikopoulos and Le Cann, 1996; Mulder et al., 2012). Tidal currents can have speeds up to 1 m/s (de Madron et al., 1999; Mulder et al., 2012; Khripounoff et al., 2014) and are weaker in the southern region of the Bay of Biscay (lower than 45°N) (Koutsikopoulos and Le Cann, 1996). These currents follow a semi-diurnal tidal frequency (e.g., Pingree and Le Cann, 1989; Koutsikopoulos and Le Cann, 1996; de Madron et al., 1999; Mulder et al., 2012) and can extend deep into the canyons (Pichon et al., 2013). Internal waves cause the resuspension of sediment and create nepheloid layers, as observed in Cap-Ferret Canyon on the Aquitaine margin (de Madron et al., 1999).

Currents have an influence on the primary production or on other biological traits within the Bay of Biscay. Tidal currents and upwelling are thought to promote water mixing and the release of nutrients. This creates a favorable environment for phytoplankton growth, thus enhancing primary production (Pingree and Mardell, 1985; Huthnance, 1995). Tidal currents and internal waves can transport the primary production into canyons, thus providing food for filter-feeders such as CWCs (Huthnance, 1995; Amaro et al., 2015, 2016).

Habitat type was visually assigned for images based on the coral biota classification system created within the CoralFISH project (Davies et al., in press). This classification system was created as a tool to standardize habitat observations across the CoralFISH regions in the North-east Atlantic and Mediterranean. The definition of a habitat given by this classification was an area “where a coherent suite of conspicuous epibenthic organisms, including CWCs, extending throughout a minimum estimated area of 25 m² as observed by underwater cameras.” Coral biotopes (or habitats) were based on the dominant coral species or group of species, the presence of scleractinian framework, usually L. pertusa and M. oculata, and the substrate type. Large geological features, such as vertical walls and boulders, were also included in the classification.

The CoralFISH coral biota classification system is a hierarchical system consisting of three levels (Davies et al., in press): level 1, the broadest level, includes the dominant of coral type(s) and substrate type, whereas level 3, the most detailed level, includes the coral species constructing the habitat, the substrate type and possible geological features as well as conspicuous non-coral species. Due to the large variation and thus the high number of level 3 habitats over a large study area, only the first level of this classification was used (Biotope level 1).

Substrate cover and fauna were analyzed on a subset (2,350 images) of the 4,191 images on which CWC habitats were observed (see Section Results). The subset was created by selecting images at an interval of ~1 min from the beginning of the dive. As the images acquired using the Scampi system were taken manually and at irregular intervals, images taken 15 s before or after the 1-min interval were considered for the subset.

Substrate cover, including colonial scleractinians, was measured using a 100 point grid, which was placed over the subset images using the software COVER (C. Carré, Ifremer; see Gomes-Pereira et al., 2016). Substrate types were assigned to each point and a relative substrate percentage cover of the visible part of the image was calculated for each subset image.

For each subset image, fauna were enumerated and identified to the lowest taxonomic level possible. It is difficult to identify deep-sea species to species level from images, because of a lack of associated identifications based on morphological characteristics or genetics. Therefore, they were identified as morphotypes and given an Operational Taxonomic Unit (OTU). To aid identification, a species catalog was used (Howell and Davies, 2010: http://www.marlin.ac.uk/deep-sea-species-image-catalogue/) and updated as part of a collaborative project between Plymouth University, Ifremer and NOAA. For this current study, the focus was coral morphotypes. Where possible, species were identified and/or confirmed by coral taxonomists using voucher-specimen collected during the BobEco cruise (see Acknowledgments).

For the reef-forming scleractinian species L. pertusa, M. oculata, and Solenosmilia variabilis percentage cover was measured using the same method as substrate cover. These three species were not included in any result concerning the abundances of individual organisms. However, they were included in a number of analyses (see Section Statistical Analysis). In many instances, the resolution of images did not allow for discrimination between L. pertusa and M. oculata, therefore were treated as one morphotype, since previous literature has shown that these species nearly systematically co-occur in the Bay of Biscay (De Mol et al., 2011; Arnaud-Haond et al., in press).

A digital Terrain Model (DTM) was available for the whole Bay of Biscay at a resolution of 100 m (Figure 1). DTMs were also available for four boxes (bob-boxes) at a higher resolution of 15 or 25 m (Bourillet et al., 2012), representing each of the four zones of the Bay of Biscay (Figure 1; Section Margin Morphology). The boxes Bob-1 and Bob-2 have a surface of ~10,000 km² and are located on the northern and central part of the Armorican margin, respectively. The third box, Bob-3, has a surface of ~1,900 km² and represents the southern part of the Armorican margin. The forth box, Bob-4, with a surface of ~7,000 km², is located on the Aquitaine margin. The resolutions of Bob-1, Bob-3, and Bob-4 are 15 m, however, the resolution of the Bob-2 bathymetry is lower (25 m), due to poor weather conditions during the time of acquisition. The bathymetry for the dives on the Celtic margin and Athos Canyon (Aquitaine margin) was only available at the lower (100 m) resolution. For the consistency across the whole dataset, depth-values were extracted from the lower (100 m) bathymetry resolution. This resolution was also used to measure the 3D-length of the dives and habitat segments.

Terrain derivatives were extracted from the bathymetry using the ArcGIS extension Benthic Terrain Modeler v. 3.0 (Wright et al., 2012). Slope, direction (northness or cosAspect, eastness or sinAspect), curvature (general, plan and profile), Surface to Planar, BPI and a measure of rugosity (VRM = Vector Ruggedness Measure) were calculated. Neighborhood sizes of ~200, 300, and 500 m were chosen for the fine scale BPI and 1 and 1.5 km for the broad scale BPI. For the VRM, similar neighborhood sizes (except 200 m) were used.

Geomorphological classes were produced for the entire Bay of Biscay, using bathymetry data, its derivatives, such as slope, canyon network extraction as well as expert interpretation (De Chambure et al., 2013). The geomorphological classes are available on the same resolutions as the DTMs and mapped by Bourillet et al. (2012) and are using the Coastal and Marine Ecological Classification Standard (CMECS) code (Madden et al., 2008). A total of 21 classes have been created on the higher (15/25 m) resolution for the bob-boxes and 15 classes on the lower (100 m) resolution for the whole Bay of Biscay.

One of 15 geomorphological classes at a 100 m resolution was attributed to each image for the whole dataset in ArcGIS 10.2 and one of 21 classes at a 15/25 m resolution for each image in the bob-boxes. Due to the uneven and sometimes low number of images per geomorphological class, classes were merged for statistical robustness according to three different criteria: (i) Morphology with three attributes: Canyon, Interfluve and Upper slope, (ii) Slope, with four intervals depending on the resolution: at 15 m resolution: <10°, 10–20°, 20–40°, and >40°; at 100 m resolution: <10°, 10–15°, 15–25°, and >25°, and (iii) Location, with two attributes: northwestern and southeastern flank of a canyon/interfluve (Table 2).

Seawater potential temperature and salinity data were publically available (Copernicus Marine Environment Monitoring Service: http://marine.copernicus.eu/) at a resolution of 0.083° latitude (~10 km). These variables were available for different depth layers (from 0 to 5,500 m depth). The monthly means from 15/12/2009 to 15/12/2011, available at the lowest depth layer as possible, were used to calculate mean potential temperature (in Kelvin) and salinity (in PSU). A geotiff (geo-referenced image) was created for both variables and a value extracted for each image of the data set in ArcGIS. Temperature values were transformed to °C.

The seawater density can describe the mixing of water masses. The potential density anomaly (sigma-theta or σ_Θ), using the potential temperature instead of in situ measured data, was calculated according to Dullo et al. (2008).

A raster of the sigma-theta was calculated using the rasters of the potential temperature and salinity data in the software program R. A value was extracted for each image using ArcGIS from a geotiff.

Several metrics—abundance, density and diversity—were used to characterize the habitats and the engineering coral species. Mean coral densities per habitat type and per segment were calculated. The density was expressed per image, instead of per linear transect or surface measure for three reasons: (i) only a subset of images were analyzed for fauna which makes it difficult to give densities per linear meter, (ii) no lasers were available on both optical techniques, and (iii) the altimeter on the Scampi frame was not reliable enough to give good estimations of area of images, complicating the expression of densities by surface measure. A spearman correlation was calculated to investigate whether segment length and mean densities (in individuals/image) per segment were correlated.

Rarefaction analysis (individual based) was used to compare differences in richness of corals between habitats. The Hurlbert's index (Hurlbert, 1971) is given as a diversity index, comparing the diversity of the habitats using a random sample size equal to the smallest number of individuals observed in one of these habitat, and thus, limiting the influence of unequal sample sizes. The three scleractinians measured as percentage cover were excluded from these analyses.

A Spearman correlation tested the relationship between the mean percentage of scleractinian cover and the total abundances of other coral types, i.e., antipatharians, gorgonians and seapens, per habitat segment.

Principal Component Analyses (PCA) were computed to investigate the (dis)similarities in species composition of the different habitats. Two PCAs were performed: (i) a PCA on a covariance matrix using the Hellinger-transformed abundances of each morphotype per habitat, excluding the reef-forming scleractinians L. pertusa, M. oculata, and S. variabilis, and (ii) a PCA on a correlation matrix using the raw total abundances of morphotypes and the mean percentage scleractinian cover per habitat. The later included the three previous mentioned scleractinians. The Hellinger-transformation gives a low weight to the morphotypes with a high abundance (Legendre and Gallagher, 2001).

Coral habitats were mapped in the canyons surveyed in the Bay of Biscay. The proportion of each coral habitat per canyon was used to describe the distribution of these habitats per canyon.

A PCA was computed to investigate the (dis)similarities in habitat types observed in the canyons using the Hellinger-transformed total linear (i.e., lengths) of each coral habitat (Legendre and Gallagher, 2001).

A specific principal component analysis with respect to instrumental variables, the Between-Class Analysis (BCA; Doledec and Chessel, 1987), was used to investigate if the habitats were characterized by different environmental settings. The BCA is a multivariate analysis which partitions and maximizes the variance between groups of one qualitative variable. A matrix of environmental variables (including latitude, temperature, sigma-theta, slope, northness, eastness, general, plan and profile curvature, Surface to Planar, BPI, and VRM) per habitat segment was used as the response variable and the habitats (all habitats separately or categorized into biogenic, hard substrate and soft substrate habitats) were used for partitioning. This analysis aims to discriminate the segments based on their environmental conditions and how much of this variation is explained by the habitats.

Chi-square tests were used to test for differences in coral distribution according to the three qualitative variables: morphology, slope and location. Chi-square tests were undertaken using the image data, rather than habitat segments because they involve the geomorphological classes (see Section Environmental Data) and habitat segments can cross geomorphological classes.

All analyses were performed using the open source software R. The R packages “ade4” (Dray and Dufour, 2007) and “vegan” (Oksanen et al., 2016) were used for the diversity measurements, PCAs and BCA.

The distribution of CWC habitats is heterogeneous in the canyons of the Bay of Biscay. The majority of canyons in this study hosts four or more and up to seven, coral habitats in the same canyon. The heterogeneity in the distribution of coral habitats seems to be largely driven by the substratum type, both at the scale of the Bay of Biscay and at the scale of canyons; the most important patterns being (i) the absence of live scleractinian habitats in canyons on the southern Armorican margin and the Aquitaine margin and (ii) the dominance of habitats on soft substrate in canyons incising the Aquitaine margin.

At the canyon scale, canyons can be divided into three groups based on their dominant substrate type, depending on their coral habitat composition. The canyon grouping matches with the mean percentage of substrate cover within each group (soft, hard, and/or scleractinians).

In general, the canyons in the southern part of the Bay of Biscay are smoother and more sedimentary without geomorphological features known for their hard substrate when compared with canyons in the northern or central parts of the Bay of Biscay which present falls and cliffs (Bourillet et al., 2010). The canyons on the Aquitaine margin also seem to have a different sedimentation regime, because of their shorter distance to the shore (Mulder et al., 2012; Schmidt et al., 2014), similar to Nazaré Canyon (de Stigter et al., 2007). The southern canyons, e.g., Cap-Ferret and Capbreton Canyons exhibit higher sedimentation rates and more recent sediment input than canyons on the Armorican margin (Mulder et al., 2012; Schmidt et al., 2014). This sediment was not being remobilized due to the low internal wave energy and lower current speeds in the canyons of the Aquitaine margin (Koutsikopoulos and Le Cann, 1996; Mulder et al., 2012), preventing erosion and, therefore, the exposure of hard substrate. This erosion is observed in Blackmud, Audierne and Guilvinec Canyons on the northern and central Armorican margin (Mulder et al., 2012; Khripounoff et al., 2014). Such a regime of higher sedimentation toward the southern Bay of Biscay likely explains both the quasi absence of scleractinian habitats and the dominance of soft substrate habitats on the southern Armorican margin and the Aquitaine margin.

Variations in sedimentation regime at canyon scale may also account for some of the variability in habitat composition and explain the statistically significant differences observed in the distribution of scleractinian and seapen habitats between the flanks of canyons. The Blackmud Canyon provides a good example. Due to its location on the Armorican margin, elevated currents and lack of recent sedimentation (Mulder et al., 2012), this canyon was a good candidate for scleractinian habitats. However, only the southeastern flank could have been explored, which turned out to be dominated by seapens on soft substrate.

This asymmetry in canyons—one eroded flank and one sedimentary flank—has been previously observed in the Bay of Biscay (Van Rooij et al., 2010; De Mol et al., 2011; Huvenne et al., 2011; Sanchez et al., 2014), in the Mediterranean (Orejas et al., 2009; Fabri et al., 2014) and off Canada (Mortensen and Buhl-Mortensen, 2005). The observed asymmetry is due to a dominated current, e.g., a westerly current in Cassidaigne Canyon, or dense shelf water cascading in Cap de Creus and Lacaze-Duthiers Canyons (Orejas et al., 2009; Fabri et al., 2014). In the Bay of Biscay, the slope current flows polewards along the slope (Pingree and Le Cann, 1989, 1992; Koutsikopoulos and Le Cann, 1996) in a dominant southeastern—northwestern direction, eroding the northwestern flank from canyons and exposing hard substrate on this flank. This important current in the Bay of Biscay has therefore been suggested to favor the development of species needing hard substratum on the northwestern flank compared to the more sedimentary southeastern flank (Van Rooij et al., 2010). In the present study, however, the occurrence of mixed corals on hard substrate on the sedimentary southeastern flank contradicts previous observations and hypothesis. This habitat occurred more often on steep areas with slopes of more than 20°, suggesting that this apparent exception may be due to the steep topography itself or the accelerated currents it forms preventing deposition of sediment. This local prevention of sediment deposition may potentially lead to the occasional exposure of hard substrate on the southeastern flank of canyons.

To summarize, the distribution patterns of coral habitats—a dominance of soft substrate habitats in canyons of the Aquitaine margin and an absence of live scleractinian habitats in the southern Bay of Biscay—can be related to the substrate heterogeneity, influenced by differences in morphology and hydrology, at the scale of the Bay of Biscay. Coral habitat distribution also varies at a canyon scale, due to an asymmetry of hydrological regimes within canyons, leading to eroded northwestern flanks that enhance the colonization by scleractinians and more sedimentary southeastern flanks favoring soft sediment habitats.

Coral habitats in the Bay of Biscay are dominated by the three reef-forming scleractinians as well as the antipatharian Leiopathes spp., the gorgonian N. versluysi and the seapen K. cf. stelliferum. Coral assemblages, however, are much more diverse than these few dominant species. A total of 62 coral morphotypes were identified in this study. The composition of coral assemblages and the correlations between coral densities or cover all indicated a clear dichotomy between soft sediment dominated habitats and hard substrate/biogenic habitats.

Corals are usually specialized in either hard or soft substrate. Seapens are adapted to live in soft sediment by a peduncle which anchors the colony in the sediment (Williams, 1995). Most antipatharians, colonial scleractinians and most gorgonians need hard substrate to settle (Roberts et al., 2009; Wagner et al., 2012). Members of the gorgonian family Isididae, e.g., A. arbuscula, can also occur in soft sediment due to a root-like holdfast (e.g., Mortensen and Buhl-Mortensen, 2005; Wienberg et al., 2009).

The biogenic and hard substrate habitats share coral morphotypes and clustered together in the PCA excluding the three reef-forming scleractinians. The coral assemblage of colonial scleractinians on soft substrate habitat also clustered with these habitats, despite the dominance of soft sediment in this habitat. Coral species preferring hard substrate could settle on the scleractinian framework emerging from the soft sediment. These corals, therefore, can be present in a habitat with a non-optimal substrate cover, as was observed in Nora Canyon in the Mediterranean Sea (Taviani et al., in press). With the addition of the reef-forming scleractinians in the analysis, the biogenic reef assemblages differed from those on hard substratum; in reef, L. pertusa/M. oculata colonies are present, while S. variabilis is absent. Additionally, the reef composition is characterized by different gorgonian morphotypes than that of hard substrate habitats and it also includes antipatharians and solitary corals, that are almost absent in hard substrate habitats.

While the biogenic and hard substrate habitats shared some species, the coral assemblages of the soft substrate habitats (excluding colonial scleractinians) differ more from each other as shown by the PCA (Figure 4A), and thus, show a higher species turnover. The alpha-diversity of these habitats is also generally lower than that of the coral assemblages of biogenic and hard substrate habitats. The mixed corals on soft substrate habitat is characterized by the gorgonian A. arbuscula, also characterizing gorgonians on soft substrate, but it also includes the seapen K. cf. stelliferum, characterizing the seapens habitat. The few segments of this mixed coral habitat are surrounded by either gorgonian or seapen habitat, what could suggest that the mixed coral habitat may function as a “transition” zone between the gorgonian or seapen habitats. Even though these three habitats share K. cf. stelliferum and A. arbuscula, the PCA does separate the seapen habitat from the other two soft sediment habitats, due to other seapen species, e.g., D. gracile, which were exclusively seen in the seapens habitat.

The submarine canyons in the Bay of Biscay host a large range of coral habitats, making them an important target for marine management and conservation. Coral habitats are threatened by human activities of which litter and fisheries are the main impacts in the Bay of Biscay.

Litter, including lost fishing gear, is largely present in the canyons of the Bay of Biscay (Galgani et al., 1995; van den Beld et al., in press). Corals, boulders and other features forming relief on the seafloor are important structures that can trap litter (e.g., Galgani et al., 2000) as shown by the presence of more litter items in areas with a seafloor relief formed by geological or biological features than without a relief (Watters et al., 2010; Bergmann and Klages, 2012; Schlining et al., 2013; van den Beld et al., in press).

Bottom trawling is probably one of the largest anthropogenic threats to CWCs (Fosså et al., 2002; Hall-Spencer et al., 2002; Benn et al., 2010) and this fishing activity is increasing on the rims of canyons (Martín et al., 2014b). Trawl gear can damage CWC communities by reducing or changing coral abundances, diversity and community or the removal of structuring species (reviewed by Clark et al., 2016), as shown for coral reefs of the NE Atlantic (Hall-Spencer et al., 2002) and Solenosmilia thickets on Australian seamounts (Althaus et al., 2009). In addition to damages caused by physical contact with trawl gear, trawling can have an indirect impact on CWCs. It causes extensive sediment resuspension in the water column that is transported further down the canyon and it changes the seafloor relief by smoothening canyon flanks through repeatedly scraping off the seafloor resulting in homogeneous slopes and low rugosity (Puig et al., 2012; Martín et al., 2014a). These effects can also have an impact on CWCs and associated community.

Canyons are less accessible for trawling gears than the interfluves and upper slope due to their steep and complex topography and, therefore, may function as a natural refuge for CWCs (Huvenne et al., 2011; Fernandez-Arcaya et al., 2017). The results of this present study support this hypothesis. Most (live) habitats, including coral reef, occurred more frequently in the canyons than on the interfluves or upper slope. A consistent exception was coral rubble that was more frequent on the interfluves and upper slope than in the canyon. Furthermore, live coral habitats are more frequently observed on steeper areas, hardly accessible to trawling, while coral rubble is more often observed on flatter areas. Similarly to the last result, the highest number of corals in Whittard Canyon has been observed on areas with steep slopes (Morris et al., 2013).

Impact by fisheries may have an influence on the fields of rubble on shallow areas, such as the upper slope, and the presence of live corals in canyons. Previous evidence from the 1920s show that fishermen trawling around and on the continental slope of the Bay of Biscay, had caught L. pertusa and M. oculata in their nets (Joubin, 1922).

Natural causes also influence the distribution of coral species and habitats. Live corals, such as scleractinians, may prefer steep topography over flatter areas, which could be related to for example accelerated currents. Furthermore, environmental changes over time, such as sea temperature elevation since the Last Glacial Maximum, can cause the death and breakdown of CWCs. In Guilvinec and Penmarc'h Canyons, age measurements suggested that the dead scleractinian corals, occurring in shallower waters (200–300 m), are older (~7–8 ka) than the live corals deeper (600–700 m) in the canyon (~1–2 ka) (De Mol et al., 2011). The authors suggested that both trawling and natural events could cause these dead corals.

Besides coral rubble, the seapen habitat is also observed in shallow waters (from 234 m water depth), on areas with a slope value less than 20° and is more common on interfluves than in the canyon (100 m res.). Some seapen species can retract completely within the sediment, as has been observed for Protoptilum carpenteri (Packer et al., 2007; Baker et al., 2012) and K. stelliferum, Pennatula phosphorea, and Virgularia mirabilis (De Clippele et al., 2015). This may suggest that some seapen species are resilient to trawling compared to other corals. It may be also possible that seapens (re)colonize an area more rapidly than colonial scleractinians. However, it may also be possible that trawling does not take place as shallow as certain seapens, as it is suggested for Funiculina quadrangularis (not able to withdraw in the sediment) occurring at 240 m water depth in Mediterranean canyons (Fabri et al., 2014).

In summary, our results support the hypothesis that canyons may function as natural refuges for coral habitats. Trawling may cause the observed distribution of live habitats favoring steep slopes, but the influence of natural causes cannot be excluded. Seapens exhibit specific features that may make them more resilient to trawling and may explain their similar distribution to coral rubble, compared to other coral habitats.

Submarine canyons are “hotspots” for coral habitats and could serve as natural refuges for certain coral habitats. However, conservation measures are rare in the Bay of Biscay. Until the present day, there are two measures on the Aquitaine margin. A fishing restricted area is in place around Capbreton Canyon since 1985, prohibiting certain types of fishing around this canyon (Sanchez et al., 2013). Under the EC Birds Directive, a Special Protection Area (SPA) has been designated at the head of Cap-Ferret Canyon, but does not include management measures of relevance for benthic habitats.

This study fed into a proposal to define sectors for a network of Natura 2000 sites to protect reef habitats under the Habitats Directive (MNHN-SPN and GIS-Posidonie, 2014). The designation of a Natura 2000 network comprising submarine canyons is a step forward in the protection of deep-sea habitats in the French Atlantic.

The Natura 2000 management measures will, however, not apply to soft sediment coral habitats, because this type of habitat does not fall under the Habitats Directives. Although the diversity was low, the (possible) unique species compositions may make them potential candidates for protection. The OSPAR commission does have “Seapen and burrowing megafauna communities” and “Coral gardens” listed as threatened and/or declining habitats (OSPAR, 2008) recognizing its potential vulnerability to anthropogenic impact, but most of the seapen communities remain unprotected by any form of legislation. Seapens can have an important value for humans, supported by the presence of burrows made by e.g., the commercially important crustacean Nephrops norvegicus (langoustine) (OSPAR, 2010) and associations of fish larvae, e.g., Sebastes spp., with seapens (Baillon et al., 2012).