Cold-Water Corals and Anthropogenic Impacts in La Fonera Submarine Canyon Head, Northwestern Mediterranean Sea

We assess the occurrence and extent of cold-water coral (CWC) species Madrepora oculata and Dendrophyllia cornigera, as well as gorgonian red coral Corallium rubrum, in La Fonera canyon head (Northwestern Mediterranean Sea), as well as human impacts taking place in their habitats. Occurrence is assessed based on Remotely Operated Vehicle (ROV) video imaging. Terrain classification techniques are applied to high-resolution swath bathymetric data to obtain semi-automatic interpretative maps to identify the relationship between coral distribution patterns and canyon environments. A total of 21 ROV immersions were carried out in different canyon environments at depths ranging between 79 and 401 m. Large, healthy colonies of M. oculata occur on abrupt, protected, often overhanging, rocky sections of the canyon walls, especially in Illa Negra branch. D. cornigera is sparser and evenly distributed at depth, on relatively low sloping areas, in rocky but also partially sedimented areas. C. rubrum is most frequent between 100 and 160 m on highly sloping rocky areas. The probable extent of CWC habitats is quantified by applying a maximum entropy model to predict habitat suitability: 0.36 km2 yield M. oculata occurrence probabilities over 70%. Similar predictive models have been produced for D. cornigera and C. rubrum. All ROV transects document either the presence of litter on the seafloor or pervasive trawling marks. Nets and longlines are imaged entangled on coral colonies. Coral rubble is observed at the foot of impacted colonies. Some colonies are partially covered by sediment that could be the result of the resuspension generated by bottom trawling on neighbouring fishing grounds, which has been demonstrated to be responsible of daily increases in sediment fluxes within the canyon. The characteristics of the CWC community in La Fonera canyon are indicative that it withstands high environmental stress of both natural and human origin.


Introduction
1.4 mgÁl −1 , with high turbidity events reaching values of 40 mgÁl -1 and lasting 1 to 6 hours associated with sediment gravity flows triggered by trawling activities [42].
Exceptional oceanographic events are storms and dense-shelf water cascading, during which La Fonera canyon is able to funnel suspended sediment and associated nutrients and pollutants [38,42,45,46]. Concurrent dense shelf water cascading events and eastern storms may generate near-bottom current speed peaks of >50 cmÁs -1 , suspended sediment concentrations over 6 mgÁl -1 and instantaneous sediment fluxes >2.4 gÁm −2 s −1 at 370 m water depth in Illa Negra branch as measured in situ in March 2007 [46]. These values increase at 1200 m water depth, since sediment enters the canyon not only through its head but also across the canyon walls. Near-bottom temperature at 370 m water depth in Illa Negra branch is in the 13.2-13.4°C range, with values slightly below 12.6°C during cascading events [46].
As stated above, bottom trawling generates sharp increases in mass fluxes within the canyon [42,44,47]. Bottom trawling takes place on the shelf and along both canyon flanks, with special intensity in the northern rim and canyon wall along the Sant Sebastià fishing ground down to water depths of about 750 m (Fig 2). A mooring placed at 980 m depth within a tributary valley on the northern canyon flank revealed a daily occurrence of sediment transport events linked to the passage of the trawling fleet upslope the site, with maximum downslope velocities of up to 38 cmÁs −1 and concentrations of up to 236 mgÁl −1 close to the bottom [43,48]. Thus, bottom trawling-generated flows are equivalent in the long term to natural small-scale landslides [48], and are responsible of widespread erosion of recent sediment [49] and large-scale changes in the seascape of the region [48]. Bottom trawling can also contribute to the development of slope nepheloid layers, and thus its effects can propagate away from fishing grounds [50]. La Fonera canyon shows the highest mean concentrations of litter on the deep-sea floor ever reported, with a mean exceeding 15,000 itemsÁkm -2 calculated after ROV observations [51]. Litter mostly concentrates on the canyon floor at water depths exceeding 1000 m, whereas canyon walls seem to be less impacted.
In this paper, we present a systematic ROV exploration of different canyon environments within La Fonera canyon head aiming at detecting CWC communities and quantifying their extent by applying predictive habitat mapping techniques based on high-resolution swath bathymetry data. In addition, we describe the anthropogenic impacts that these habitats are currently experiencing. The results of this study may represent a first step for establishing protection measures in an area extremely pressured by bottom trawling.

Bathymetric data
Swath bathymetry data from La Fonera canyon head and neighbouring continental shelf were obtained during the "Euroleón" cruise onboard B.I.O. Hespérides in October 2007 using a Simrad EM-1002S multibeam echosounder. It works at a frequency of 95 kHz with 111 beams per ping, and was operated in equidistant mode, that is, with swath width independent of water depth and fixed to 500 m. Data were logged using Kongsberg's Seafloor Information System (SIS) and processed using Caris Hips and Sips software, yielding a bathymetric grid with a resolution of 4 m (Fig 2). Derivative backscatter data were processed using SwathEd software developed by the Ocean Mapping Group, University of New Brunswick, yielding a backscatter grid with a resolution of 20 m. A detailed morphological description of La Fonera canyon head based on these data was published by [36].

Benthic terrain modelling
Terrain classification techniques were applied to bathymetric data in order to obtain a semiautomatic interpretative map, as a basis to assess the relationship between coral distribution patterns and the morphological seabed expression at various locations. The benthic terrain model tool (BTM) creates a terrain model that ascribes each location into a zone describing the structure and seabed complexity that characterizes the physical environment, and is usually based upon two primary data sets: bathymetry (Fig 2) and backscatter ( Fig 3A). From the first of them, other bathymetric parameters such as seafloor gradient (Fig 3B), bathymetric position index (BPI) (Fig 3C and 3D), curvature ( Fig 3E) and rugosity ( Fig 3F) are derived, and are later combined through a set of standard algorithms to classify the benthic landscape and perform maximum entropy prediction of habitat suitability.
Curvature is calculated as the second derivative of the surface, and rugosity is defined as the ratio of surface area to planar area, thus measuring terrain complexity indicative of areas with high biodiversity potential. BPI is a second-order derivative of swath bathymetry, and indicates positive and negative terrain features [52], measuring where a given location is (in terms of depth) relative to the locations surrounding it. It calculates for each cell the mean depth around a minor given radius through a neighbourhood function and compares it to the mean depth around a major given radius. Positive cell values denote features that are higher than the surroundings and, inversely, negative values indicate features lower than the surroundings. The first ones may be associated to ridges or edges such as canyon rims, whereas the others relate to valleys or depressions. BPI rasters can be built at different scales with the aim to identify small or large scale features on the seascape.
Due to an excess of acquisition artifacts, backscatter and curvature grids were discarded during terrain modelling. Grid size of model input data sets was set to 4 m since ROV video footage showed that substrate and associated communities vary at a very small scale, so larger grid sizes would fail at distinguishing small scale heterogeneity. Based on the general bathymetry of La Fonera canyon head, BPIs were calculated at two scales: (1) a fine-scale bathymetric position index raster (FS BPI) was computed for an inner radius of 4 m and an outer radius of 20 m (Fig 3C), and (2) a broad-scale bathymetric position index raster (BS BPI) was built with an inner radius of 8 m and an outer radius of 60 m (Fig 3D). Since the range of BPI values decreases with size of analysis window, BPI rasters were normalized by subtracting the mean value and dividing by the standard deviation. Benthic terrain modelling was then applied to 99.65 km 2 of the bathymetric data set following an ad hoc classification schema (cf. section 4.1).

ROV data
ROV video images were collected using a Seaeye Lynx 1500 during the months of November 2009, and May and June 2010, operated from MV Bon Pigall owned by Fundació Argomaris, designed for a depth range of 400 m. According to national regulations, no specific permissions were required for the use of ROVs in La Fonera canyon at the time of the study. ROV tracks were planned based on the bathymetric data aiming at obtaining enough information of all potential habitats in the canyon head. For each morphologically similar potential habitat, different locations were explored so that surveyed points are widely distributed within the canyon head. Safety was a basic criterion in such a rough terrain that is in addition littered with lost fishing gear posing a significant additional threat to the ROV [51]. C. rubrum is included in Annex V of Council Directive 92/43/EEC of 21 May 1992 on the conservation of natural habitats and of wild fauna and flora. No samples were collected of any species.
A total of 21 ROV immersions were carried out in La Fonera canyon head (Table 1). ROV tracking was achieved using a Seaquest Tracklink 1505 USBL acoustic tracking system integrated with the ship's DGPS and motion sensor unit. During three of immersions, ROV positioning was lost due to system malfunctioning. Thus, 18 positioned ROV transects (Fig 2) were used to generate predictive maps of benthic habitats. The ROV was equipped with three video cameras: a low-resolution colour camera, a black and white long-range camera and a high-resolution colour camera. Images obtained by the first two were available on board live during the operations; the latter was only recovered when the ROV was back on board. A total of 420 Gb of video images were recorded (41 h 20 min approximately). Transects were performed upslope (i.e. from the canyon axis up to the canyon rim), at depths ranging between 79 and 401 m.
ROV navigation was incorporated to an ArcGIS project together with bathymetric and backscatter data. After filtering position fixes where ROV was not in contact with the seafloor, 7914 position fixes depict the ROV track, that is, a mean of more than three fixes per minute of total recording time (Fig 2). Each second of video footage was classified in terms of substratum (mud or fine sediment, sand and rock), presence or absence of CWC, anthropogenic impacts (litter, litter type, trawling marks) and any other relevant observation. That means that a given position fix, spanning for some seconds, can contain multiple observations. In this study, "litter" encompasses all man-made objects such as ropes, longlines, nets and other fishing gear, plastic bags, bottles, cans, boxes, and also cultural heritage objects such as a Roman amphora.
With the aim to evaluate the correlation between cold-water and red coral observations and the bathymetric parameters (depth, gradient, rugosity, curvature, BS and FS bathymetric position index) as well as backscatter data, each of these parameters has been extracted for all position fixes with positive ROV observations.

Maximum entropy prediction
Maximum entropy prediction of habitat suitability was performed with Maxent software, version 3.3.3k available online [53,54], using the bathymetric parameters in combination with ROV positive observations of the three coral species considered in this study, namely M. oculata, D. cornigera and C. rubrum. In addition, the same model was used to predict litter accumulations using litter positive ROV observations. The parameters used were depth, gradient, rugosity, curvature, and fine scale and broad scale bathymetric position indexes. Backscatter data were not incorporated into the calculations since nadir noise and gain changes are too dominant for using data for quantitative analysis and obtaining proper predictions.

Benthic terrain modelling
The region over which the benthic terrain modelling has been applied comprises the La Fonera canyon head and surrounding shelf and slope (Fig 2). The canyon head down to an axial depth of 1000 m occupies mostly the western and southern parts, whereas a wide and smooth sloping shelf with a poorly defined shelf break extends over the northeastern part. Based on depth, slope and the two standardized BPI rasters, a classification scheme incorporating both marginscale and fine structure elements has been specially designed for the study area (Table 2). When applied (Fig 4), it is able to differentiate between (1) flat areas on the shelf, walls and canyon floor, (2) sloping areas, (3) elevated areas on the shelf, walls (divides) and canyon floor, and (4) depressed areas on the canyon walls (gullies) or along the canyon axis. Being overall a topographically rough area, flat areas are defined as those less than 6°in gradient (8°when including shallow depths, i.e. the canyon rim) and with BPI values within the first standard deviation (between -100 and 100), except for the shelf, where BPI values of 60 limit flats and shelf crests, and for the canyon rim, with positive BS BPI values. Flats are classified primarily based on their depth position. The 130 m isobath defines the limit between the inner and the outer shelf, and the 160 m isobath limits the outer shelf and the ramp, that is, a very gently sloping area connecting the Roses shelf and the upper open slope. All flats between 275 m and 475 m water depth are then considered canyon wall flats, and those located at depths exceeding 475 m represent flat-bottomed sections of the canyon floor.
Sloping areas are defined to exceed 6°in gradient (8°in shallow depths, i.e. the upper canyon wall) and to have BPI values also within one standard deviation, except for the upper canyon wall that, as it happens with the canyon rim, has positive BPI values due to their close location to the much deeper lower wall and canyon axis. Sloping areas are classified taking into account their gradient, characterizing gentle slopes below 15°and steep slopes above 15°and below 60°. Near-vertical walls are those cells with gradients exceeding 60°, disregarding depth and BPI values. Depressed areas are accounted at all depth ranges and at all slope gradient ranges, except for near-vertical walls, when either one or both BPI's are negative below one standard deviation. If BS BPI is below one standard deviation but FS is not, it is considered that the cell is located nearby the axis of an incised, v-shaped section of the canyon. Conversely, if BS BPI is within one standard deviation and FS is not, it can be assumed that the cell is in a local depression within a broader flat or homogeneously sloping area, for example a gully in the canyon wall. If both BPIs are negative below one standard deviation, the cell is located in a very narrow depression or at the foot of the canyon wall. Crests are differentiated in shelf crests, located at depths shallower than 160 m and with BPI's exceeding 60, and as slope crests and divides, at larger depths, when either one or both BPI's are positive above one standard deviation. If FS, BS or both BPIs are positive, broad crests, narrow crests and crests within the canyon wall, that is, divides between gullies, are distinguished, respectively.
This semi-automatic terrain classification accurately describes the distribution of the different benthic terrains in the study area, with the exception of the flat canyon floor, seldom classified as gentle slope, and of the head tips of canyon branches, classified as shelf crests, which could be better described as shelf irregularities, disregarding their nature as either crests or depressions. The areal distribution of the different classes is indicated in Table 3. Within the study area, the ramp is the most common terrain class (29.1% of all cells), extending over the very gentle sloping area connecting the Roses shelf and the upper open slope (Fig 4), followed by the flat inner shelf (16.4%), representing to the smooth and sedimented regions of the inner shelf, without accounting shelf crests and irregularities, which barely sum 2%. Almost 51.7% of the cells correspond to flat areas and almost 28.6% to sloping areas, among which 1% (0.94 km 2 ) have gradients exceeding 60°. Depressions account for 10.1% of the classified area, but it is worth noting that some parts of the canyon axis are classified as flat canyon floors. Crests represent 9.6% of the area. Flat and uniformly sloping regions such as the shelf or a regular slope (e.g., the ramp and the eastern wall) are also characterised by curvature values close to zero (Fig 3E), whereas those places where the seafloor gradient abruptly changes have high positive or negative curvature values, denoting terrain irregularities such as complex, gullied canyon walls or escarpments. Rugosity values close to 1 (Fig 3F) are indicative of zones whose 3D area equals their plan area, and thus rather simple, homogeneous topographies; on the contrary, high rugosity values over 10 indicate regions whose 3D area is ten times the size of their plan area (e.g. canyon walls of Illa Negra and Sant Sebastià branches), providing an idea of a highly complex and heterogeneous seascape. The three bathymetric-derived data sets (curvature, rugosity and benthic terrain classification) provide a complete picture of the terrains in La Fonera canyon, their distribution and heterogeneity.
ROV inspection effort was focused on specific terrain classes within the study region (Table 4). Inspection efforts concentrated on the canyon rim (6.7% of position fixes with respect to 1.1% of terrain class extension), upper wall (17.6% of position fixes with respect to 1.7% of terrain class extension) and the floor of the incised branches (12.1% of position fixes with respect to 4.1% of terrain class extension). Near-vertical walls were also inspected (2.3% of position fixes) since they were good candidate locations for CWC. On the contrary, very few images were obtained from the ramp (0.6%) even though it is the most common terrain class in the study area, and no ROV images were obtained from the outer shelf. The main reason is that taking into account background information from nearby canyons (cf. section 2), these two terrain classes have very little chances to host significant CWC colonies. Also, the deep canyon floor remained uninspected as it was beyond the ROV depth range.

Substrate types
ROV images display a notable habitat variety, in terms of bottom topography and substrate types, in La Fonera canyon head and branches. ROV observations have allowed classifying the substrate of the study area into three broad groups: (1) rocky outcrops and large boulders, (2) sandy bottoms and (3) fine sediment-covered bottoms (Fig 5). Given the limitations of this classification, especially regarding the distinction of grain sizes, the second and third categories could be more accurately described and as coarse sand to fine gravel and as clay to medium sand mixtures bottoms, respectively. For ease of simplicity and conciseness we will refer to "hard substrate", "sandy bottom" and "muddy bottom" from here onwards.
Hard substrates are represented either by rocky outcrops or large boulders. Rocky outcrops dominate the lower and middle walls in the western canyon branches and gullies, as well as the eastern wall of Cap de Begur branch. They also form some linear escarpments along the eastern La Fonera canyon wall at the western end of Sant Sebastià fishing ground ( Fig 5A). Hard substrate communities are formed by a variety of encrusting organisms ( Fig 5B). Very irregular rock formations, with small caves and holes occupy the lower and middle part of the western wall of Cap de Begur branch, often forming overhangs (Fig 5C), whereas more regular, subrounded morphologies are imaged along Illa Negra branch (S1A and S1B Fig). The upper parts of the walls are more often occupied by decimetric to metric boulders, frequently with a thin mud cover (S1C Fig), which attains a relevant thickness able to smooth the original rocky relief in the linear escarpments near the Sant Sebastià fishing ground ( Fig 5D). The transition between rock outcrops and muddy bottom is abrupt at the foot of the walls especially in the inner parts of the canyon head (S1D Fig), although substrate changes are generally gradual (S1E Fig).
Sandy bottoms are mainly found in two environments: at the canyon floor along the axes of Illa Negra and Cap de Begur branches at distances shorter than 1 km from their tips and depths below 260 m, and on the canyon rims around these two branches ( Fig 5A). The first one is characterised by coarse bioclastic sand and gravel (Figs 5E and S1F), whereas sand is finer on   Most ROV observations were made over muddy bottoms (54%), while sandy bottoms and hard substrates were surveyed equally (22% and 24%, respectively) ( Table 5). Regarding the terrain classes defined on the basis of benthic terrain modelling, muddy bottoms cover all (100%) the surveyed inner shelf, ramp and canyon wall flats, and most (>80%) of the shelf crests, gentle slopes and broad slope crests. Muddy bottoms are also the dominant bottom type (>45%) in the canyon rims, footwalls and steep slopes, although a significant number of observations (>12%) in these terrain classes also display sandy bottoms. Sandy bottoms occupy most (67%) of the incised canyon terrain class, although a significant part is covered by mud (26%). Hard substrates characterize most of near-vertical walls (84%), and a majority (>50%) of upper walls, gullies, divides and narrow slope crests (Table 5).
M. oculata was observed in 12 out of the 21 ROV immersions, D. cornigera in 14 and C. rubrum in 10 of them. Out of a total of 7914 position fixes, M. oculata colonies have been imaged in 270 of them (3.41%), D. cornigera in 121 (1.53%) and C. rubrum in 169 (2.13%), but they are far from being homogeneously distributed within the canyon environments (Fig 6). Observations do not account for colony health, i.e., a positive observation could indicate the presence of a single, small, isolated, broken or sediment capped colony, whereas another positive observation could refer to a well-developed, large colony.
M. oculata colonies span over water depths ranging between 129 and 369 m (mean 230 m), and are most frequent between 180 and 300 m water depth (Fig 7), on highly sloping areas (mean 42.8°). Observed M. oculata colonies form apparently fragile frameworks but are in general in good shape except when they are directly or indirectly impacted by human activities (cf. Cold-Water Corals in La Fonera Canyon Head    Cold-Water Corals in La Fonera Canyon Head (S1 Video). Thus, most sightings of these species were made in the most abrupt parts of the canyon walls, although some were made in places like the canyon axis. These locations are in general less exposed to impacts from commercial trawling. Medium-sized, apparently decimetric colonies develop on the eastern wall of Cap de Begur branch, being especially dense at Cold-Water Corals in La Fonera Canyon Head the foot of the canyon wall where they look like hanging palm tree forests (Figs 8B and S3A). These colonies often co-occur with D. cornigera (Figs 8C and S3B) and host a wide variety of fauna including polychaete P. tubularia, ophiurioid Ophiothrix sp., oyster N. cochlear, echinoderm C. cidaris, decapods Munida sp. and P. elephas, holothurians and bony fishes (Figs 8C and S3C). M. oculata colonies also develop just opposite across the Cap de Begur branch on the western wall, although less densely (Fig 8D). In this wall, colonies reach shallower water depths (130 m, S3D Fig), also living together with Paguroidea and deep-water oysters N. cochlear (Figs  8D and S3E). M. oculata colonies extend further south on the western wall, as imaged during transect 22 (Figs 5C and S3F), than on the eastern wall, where they were no longer detected during transect 20 ( Fig 6A). Nevertheless, beyond the confluence of the Cap de Begur and Illa Negra branches, M. oculata colonies appear back on the uppermost section of the eastern wall, very close to Sant Sebastià fishing ground (Fig 6A). There, they coexist with D. cornigera and C. rubrum colonies (S3G and S3H Fig) and even grow over lost fishing gear (S4A Fig). Colonies of M. oculata are much scarcer in Illa Negra branch, and have not been observed in Sant Sebastià branch (Fig 6A). In spite of their scarcity, medium-sized living colonies were found during transect 9 along the middle and upper sections of the western wall of Illa Negra branch, nearby the confluence with Cap de Begur branch (Figs 6A and 8E).
D. cornigera colonies are sparser but evenly distributed at depth, ranging between 131 and 325 m water depth (mean 211 m) (Fig 7), on relatively low sloping areas (mean 29.3°). Sometimes found in association with M. oculata colonies (Figs 8B, S3B, S3C and S3H), they are more often observed as isolated 5-10 cm large colonies of a few individuals (Fig 8F) or as a  Cold-Water Corals in La Fonera Canyon Head single polyp (S4B Fig). As such, it is relatively abundant along the walls of Illa Negra branch (Figs 6B and S4B-S4E), but also appears on Cap de Begur (Figs 6B and S4F) and Sant Sebastià branches, as well as on the uppermost section of the eastern La Fonera canyon wall (S1E Fig), generally on sediment-covered rocks along gently sloping areas.
C. rubrum colonies are slightly more frequent than those of D. cornigera. They appear between 101 and 297 m water depth (mean 156 m), but are most frequent at the shallower part of this interval, between 100 and 160 m (Fig 7), on highly sloping areas (mean 43.2°). Only at a very small number of locations polyps were observed, like in Fig 8G, where C. rubrum grows on a N. cochlear deep-water oyster bank. In general, they appear as small, centimetric colonies located in highly protected places, such as overhangs and caves (S4G and S4H  Fig), which are far from reaching the size or the density of the well-developed M. oculata colonies. At 28% of locations where C. rubrum was found it was observed to coexist with M. oculata (S3G Fig), and only at two locations were the three species found together (S3H and S4B Figs).
Each positive observation has been related with the benthic terrain model classification. This allows identifying if each species concentrates on any of the benthic terrain classes. M. oculata colonies have been most reported on narrow slope crests (56 position fixes), near-vertical walls (54), steep slopes (52) and footwalls (35), although taking into account the number of observations carried out over each terrain class, this species preferentially settles on gullies (36% of observations on gullies included M. oculata specimens), near-vertical walls (29%) and narrow slope crests (12%) ( Table 5). D. cornigera colonies have been most observed on steep slopes (50 position fixes) and upper canyon walls (24), and taking into account the number of observations carried out over each terrain class preferentially settle on gullies (11% of observations on gullies included D. cornigera specimens) and steep slopes (5%). Contrarily to M. oculata, D. cornigera colonies are seldom observed on near-vertical walls (Table 5). Finally, ROV images mostly show red corals on upper canyon walls (108 position fixes), narrow slope crests (24) and near-vertical walls (21), and taking into account the number of observations carried out over each terrain class they are more abundant on near-vertical walls (11% of observations on near-vertical walls included C. rubrum specimens) and upper canyon walls (8%) ( Table 5). No specimens of these three species have been observed on the inner shelf, ramp, canyon wall flats, and only twice red corals have been observed on shelf crests.
Minimum, maximum and mean values of bathymetric parameters (depth, gradient, rugosity, curvature, BS and FS bathymetric position index) at locations were cold-water and red corals as well as litter have been observed are provided in Tables 6 and 7. Significant differences between the bathymetric locations and the characteristics of the terrains where these three species settle stand out. M. oculata colonies appear at a mean depth of 230 m, deeper than D. cornigera (211 m mean depth) and especially if compared to C. rubrum (156 m mean depth) ( Table 6). Both M. oculata and C. rubrum colonies develop in high gradient slopes, with mean values of 42.8°and 43.2°, respectively, which are much higher than the mean slope gradient were D. cornigera establishes (29.3°). There is also a significant difference in the mean curvature of the terrain where M. oculata establishes, with a value of 9.4 compared to where D. cornigera develops (-4.1) and the mean of all points within the study area (0.2). Backscatter values are also more elevated in areas colonized by M. oculata and D. cornigera ( Table 6). The presence of all coral species is also related to more positive values of both broad scale and fine scale BPIs ( Table 7). The most significant differences are those for C. rubrum colonies, which develop in areas with mean FS BPI values (405.6) and mean BS BPI values (230.1), almost ten times above the general value. All species develop in areas where rugosity is well over the mean (1.80 for M. oculata, with respect to a mean value of 1.19 for the study area) ( Table 7).

CWC prediction
A predictive model of M. oculata occurrence in La Fonera canyon head by maximum entropy modelling is presented in Fig 9A. Modelled occurrence probability of this species is high on the highly sloping sections of Cap de Begur and Illa Negra branches. Some 0.36 km 2 out of 6.94 km 2 of the modelled area between 75 and 410 m water depth yield probabilities over 70%, and 0.10 km 2 exceed 80% probability of occurrence of M. oculata colonies.
Similar predictive models have been produced for D. cornigera ( Fig 9B) and C. rubrum (Fig 9C). Occurrence probability of the first is in general higher than M. oculata, and its spatial distribution is broader. A total of 0.43 km 2 have probabilities over 70%, and 0.12 km 2 exceed 80% probability of occurrence of D. cornigera colonies. Its presence is highly dependent on local morphological changes in the slope; in general, it is absent in depressions (Fig 9B). On the contrary, distribution of C. rubrum is much more restricted in space, with occurrence probabilities exceeding 70% extending over 0.15 km 2 , but occurrence probabilities have higher values (1904 m 2 exceed 95%) at shallower depths than the other two species, especially on the upper parts of the canyon walls. The model predicts that this species is mostly restricted to the western wall of the canyon head, and should be almost absent in the eastern wall (Fig 9C). Jackknife tests of variable importance show that the variables that appear to have the most useful information by themselves regarding the presence of M. oculata and D. cornigera are seafloor gradient and rugosity, whereas for C. rubrum, these variables are fine-scale bathymetric position index and rugosity (S5 Fig). For the three species, the main environmental variable is water depth (S5 Fig).

Anthropogenic impacts
In parallel with cold-water and red coral identification, litter and seafloor disturbance by bottom trawling were accounted during video analysis. Litter was observed in all ROV immersions except for transect 58, which, although being devoid of litter items, is markedly impacted by Cold-Water Corals in La Fonera Canyon Head trawling (S2 Video). In 604 out of a total of 7914 position fixes, a wide variety of litter items was observed resting on the seafloor. This represents 7.6% of the video footage (Fig 6). Litter consists mostly of lost or entangled ropes and longlines (Figs 10A, 10B, S6A, S6B and S6C), nets (Figs 10D, 10E, S6D and S6E) and other fishing gear (71% of litter items), whereas the rest (29%) comprises plastic bags, bottles (Fig 10F), cans, but also other items such as a large plastic box (or an upturned boat, Fig 10F), Roman amphorae (Fig 10G), and a soccer ball. Ropes, longlines and nets often posed a serious risk for ROV navigation (S2 Video). Litter has been observed at all depths, although it is scarce at less than 100 m water depth (only 2 objects out of 623 position fixes), and at all slope gradients. It is especially abundant at two depth ranges: between 100 and 160 m and between 260 and 300 m water depth (Fig 7). There are less litter items in flatter areas, and are more frequent in topographically elevated places with respect to the surroundings, as indicated by the bathymetric position indexes ( Table 6). Litter has mostly been imaged resting on the upper canyon wall (244 items), in narrow slope crests (93) and steep slopes (82), and is proportionally more abundant in narrow deep crests (20% of observations on narrow deep crests included litter), near-vertical walls (17%), upper walls (17%) and shelf crests (15%). On the contrary, gentle slopes (1%) and the incised canyon classes (1%) display less litter (Table 5). Maximum entropy model predicts that 0.55 km 2 out of 6.94 km 2 of the modelled area between 75 and 410 m water depth yield probabilities over 70% of being directly impacted by litter (Fig 6D).
Ropes, longlines and nets relate to CWC, and specifically to M. oculata colonies, in two ways. On one hand, they have been observed entangled in coral patches (Figs 10B, S6A  Trawl marks have been identified at 322 position fixes, mostly in transects located within Sant Sebastià fishing ground (Fig 2). They are imaged as metric-wide linear depressions with overconsolidated mud clasts accumulations on their sides (Fig 11B-11E), among large expanses of the seafloor devoid of life except for scattered Cerianthus sp. (S2 Video). In general, those transects close to or within the fishing ground yielded images with an increased amount of suspended particulate material (Fig 11B and 11C). Some coral colonies located on the eastern La Fonera canyon wall have been observed to be partially covered by fine mud, to the extent that coral branches are only partly visible (Fig 11D and 11E and S2 Video).

CWC communities in La Fonera canyon
The shallow (<260 m) and deep (>260 m) sections of the axis of La Fonera canyon head are mostly covered by sand and finer sediment (Table 5), respectively, and are poorly occupied by benthic communities with erect invertebrates. Canyon walls are rocky and abrupt (Table 5), with abundant caves and overhangs, and host a large variety of benthic fauna. The canyon rim is often covered by sand but also occupied by boulders and outcropping rocks (Table 5), the latter colonised by numerous organisms. Medium-sized colonies of M. oculata as well as more sparsely distributed D. cornigera colonies have been imaged in La Fonera canyon head, which In general, larger and healthier M. oculata colonies in La Fonera canyon are located in the rougher rocky areas of the canyon walls with higher slope gradients. These areas have been classified as near-vertical walls, narrow slope crests and gullies by benthic terrain modelling incised canyon). G: Top of a roman amphora, almost totally buried by sediment at the axis of Illa Negra branch, with unidentified fish (299 m, footwall). based on morphological parameters, although the latter have only been sparsely surveyed. Often, these colonies appear in overhangs, as also observed by [23] in the canyons of the Gulf of Lion, and are oriented horizontally or downwards. Smaller M. oculata colonies have been observed in more isolated blocks and linear escarpments along the canyon walls. Among the areas within La Fonera canyon where CWC have been imaged, the occurrence of M. oculata on the western wall of Cap de Begur branch, at depths not exceeding 150 m, is among the shallowest locations in which CWC have been found, together with the Strait of Gibraltar and Cassidaigne canyon [15,23].
On the contrary, D. cornigera preferentially settles on moderately sloping areas, which have been classified as steep slopes. Red coral C. rubrum colonies prefer shallower waters (<160 m) and thus are present on the upper but also highly sloping sections of the canyon walls, classified as upper walls and near-vertical walls. Studies conducted in front of Cap de Creus promontory between 50 and 230 m depth showed that red coral is less abundant below 120 m water depth, where it has a more scattered distribution with isolated colonies, similar to our observations in La Fonera canyon, in contrast to dense patches observed in shallower water (50-85 m) [34]. Deep (>200 m) occurrences of C. rubrum were also recorded, in coexistence with M. oculata. Co-occurrence of red coral with scleractinian corals L. pertusa and M. oculata has been reported at depths ranging between 200 and 460 m in the Gulf of Lion [23] and the Sicily Channel [19,30].
A characteristic of the CWC community in La Fonera canyon is the absence of L. pertusa, in contrast with Cap de Creus and Lacaze-Duthiers canyons [20,23,35], at least within the depth range investigated. Although this species is scarce and in decline in the Mediterranean [16,24], its absence in the shallow reaches of the canyon, together with the dominance of middle-sized, fragile M. oculata colonies, the poor development of the small-sized D. cornigera colonies, and the lack or low occurrence of gorgonians and other usual species of these communities may be indicative that La Fonera CWC community is close to the environmental limits in which these species can develop, and /or withstands a high environmental stress. Among the three species, M. oculata presents higher growth rates and is better adapted to such stressing environments [55].

Favourable environments for CWC communities in La Fonera canyon
Morphological parameters derived from bathymetric data, such as slope, seafloor roughness, topographic indexes, curvature and slope directions, are proven surrogates for sedimentary environments, and are commonly used in multivariable classifications of the seabed and species distribution modelling [56,57,58]. [57] evidenced how the use of high-resolution multibeam bathymetry, in their case obtained with a ROV-mounted multibeam system in the Porcupine Seabight, yielding a 0.5 m bathymetric grid cell size at 800-900 m water depth, allowed resolving bathymetric details that are relevant in structuring the distribution of CWC. In our case, a hull-mounted multibeam system operated in equidistant mode was able to resolve such details at depths not exceeding 400 m. These parameters are indeed related to the factors affecting the distribution of CWC. This relation is quite direct regarding coral-suitable substratum, since hard substrates will most probably occur on high sloping areas, often with relatively high roughness, whereas favourable water-mass properties and food availability will be related with water depth and current patterns, the later partially controlled by seabed topography.
Based on these parameters, maximum entropy modelling allows inferring the possible extension of CWC suitable habitats in La Fonera canyon. Nevertheless, modelling results should be taken with caution since the distribution of ROV observations in the study area does not represent equitably the different types of terrains. For example, ROV surveying effort in the outer shelf and canyon floor is nil, and the ramp only accounts for 0.6% of all observations although representing 29.1% of the area (Tables 3 and 4). In addition, ROV observations are constrained to a given depth interval (deeper than 79 m, shallower than 401 m) and locations outside that interval lack observations. Therefore, beyond these water depths maximum entropy predictions are unfounded. In addition, maximum entropy prediction of habitat suitability only takes into account positive observations. Nevertheless, not all favourable grounds identified by modelling are indeed occupied by CWC. Large sections of transects located along areas where corals are predicted to occur did not show evidence of coral presence, and no large and healthy colonies similar to those observed in the eastern wall of Cap de Begur branch were observed in Illa Negra branch. This could be related to the model lacking a key environmental parameter that ultimately determines coral presence or absence when topographic conditions define a suitable habitat. Such environmental parameter could be related to food availability, sediment concentration and/or hydrodynamics.
One common characteristic of the canyons conforming the Northwestern Mediterranean CWC province is that they have their heads incised in the continental shelf, especially Cap de Creus and La Fonera canyons, whose heads are located at short distances from the coastline. This favours capturing dense shelf water when generated [38,45,46] and probably also sediment-laden waters of coastal origin during eastern storms as it happens in Blanes canyon further south [59], enhancing hydrodynamics and food arrival. Thus, canyon morphology as well as meteorology and oceanography configure particularly favouring conditions that determine the extent of this CWC province, even though water temperature is generally slightly beyond the optimal conditions for CWC development [60]. As a counterpart, this configuration can also involve an excess of sediment arrival. Thus, CWC do not occupy all morphologically favourable grounds throughout these canyons, but only specific locations where hydrodynamics prevent high sedimentation rates [22]. At these locations, hard substrates are sediment-free and ready for larval settlement. [19] suggested that canyon systems might better facilitate the extent of coral assemblages into shallower depths than open slopes due to seasonal or episodic cascading of water masses loaded with nutrients, food, and organic matter. CWC prefer highly energetic hydrodynamic environments, but although Illa Negra branch presents high current speeds (>50 cmÁs -1 ) during dense shelf water cascading events triggered or enhanced by storms [46], corals are scarce in this branch compared to Cap de Begur branch. [20] related CWC communities in Cap de Creus canyon to the energetic dense shelf water cascading flows carrying periodically nutritive particles in suspension from the shelf environments and to the reduced sediment accumulation rates caused by this process, but this seems not to be the case of Illa Negra branch where cascading events are likely milder. On the contrary, the proximity to the coastline, with the Illa Negra branch tip at a distance of barely 800 m from shore, could promote the arrival of an excess of sediment intercepted from the coastal sediment drift during storms with detrimental effects on coral development. Note, for instance, the thin sediment veneer covering rock outcrops in S1A, S1C, S4C and S4D Figs compared to rock outcrops in Cap de Begur branch ( Fig  5B and 5C). This branch would be, conversely, more protected from the arrival of excess sediment but still within the effects of high energetic processes, similarly to the more distant sections of Illa Negra branch where corals grow (Fig 8E).

Litter in La Fonera canyon and interaction with CWC
La Fonera canyon has been reported to host one of the largest mean concentrations of litter on the deep-sea floor [51]. Although [51] concluded that canyon walls seem to be relatively less impacted than the canyon floor in terms of number and density of litter items, our observations indicate that within the inspected depth range of 79 to 401 m, litter is specially abundant on the canyon walls and crests, whereas is relatively less abundant along the floor. Objects in the canyon walls are mostly linear-shaped items entangled in rock outcrops, such as ropes, longlines and nets; whereas litter in the canyon floor are round-shaped items more easily transported such as light plastics, bottles or cans. High energetic events such as storms and cascading, which are most probably the key elements that favour the extension of the Northwestern Mediterranean CWC province, also favour the transport of light and round-shaped items down the walls to the canyon floor, and then down-canyon, and concentrate them along the canyon floor surveyed by [51]. On the other hand, those linear-shaped objects that can easily be entangled, generally lost and abandoned fishing gear, and other heavy litter items, simply stay close to their original locations in the upper canyon walls, where fishing effort is larger.
Out of a total of 497 position fixes where any of the three species studied has been observed, litter is observed in the same field of view in 116 locations (23.3%). In general terms, the relation between CWC and litter presence is negative, with ropes, longlines and nets observed to be entangled and damaging coral colonies. Less protected areas and isolated colonies are highly vulnerable and much more impacted by this kind of litter, for example those located in the western limit of the Sant Sebastià fishing ground (Fig 2). Similar impacts to those observed in La Fonera canyon (entangled nets, coral rubble) have also been described in the neighbouring Cap de Creus canyon, where on average one fishing line was found every 10 m of video transect, often entangled on coral colonies [61].

Trawling impact
Impact of bottom trawling has previously been documented in many CWC communities such as those located west of Ireland and offshore Norway [62,63,64] and in the Northwestern Mediterranean Sea [20,23]. In La Fonera canyon, high sloping, rocky areas where the largest colonies have been imaged are less prone to be directly impacted (i.e. breaking and crushing) by trawling because they are already known as dangerous areas for the nets and avoided by local fishermen. In a way, the rougher and harder the terrain, the more self-protective against trawling it is.
In addition to direct impact by trawling, frequent bottom trawl activity alters the hydrodynamic and sedimentary conditions, particularly increasing suspension of sediment, further impacting suspension feeders such as coral species [2]. [65] demonstrated that although L. pertusa can tolerate fairly high sedimentation, mortality increases rapidly with extreme or persistent suspended sediment exposure even if food is available. [66] reported that the skeletal growth of this coral species is significantly lower under sediment concentrations of 25 mgÁl −1 . In La Fonera canyon area, bottom trawling along the Sant Sebastià fishing ground at water depths ranging between 170 and 750 m or deeper generates daily sediment transport events with concentrations of up to 236 mgÁl −1 close to the bottom [43,48] and contributes to the development of slope nepheloid layers, propagating suspended sediment away from fishing grounds [50]. This way, human activities add stress to a naturally stressing environment. Coral colonies partially suffocated by fine mud on the eastern La Fonera canyon wall could be the result of increased sediment settling due to trawling, in excess to the capacity of M. oculata in shedding it through ciliary action and mucus production where local currents are not strong enough to clean it away.
Anthropogenic impacts such as the presence of lost and abandoned fishing gear and both the direct and indirect impact of trawling may be nowadays intrinsic to CWC provinces since the conditions that control CWC growth as well as their role as other species' refuges for prey and nursery areas determine their high biodiversity and thus their appeal as fishing spots. The European Marine Strategy Framework Directive (MSFD) established in 2008 aims at assessing and reducing direct and indirect human impacts on marine ecosystems of Europe, with the objective of maintaining or achieving good environmental status by 2020. The General Fisheries Commission for the Mediterranean (GFCM) identified in 2008 sensitive habitats of relevance for the management of priority species that are potentially vulnerable, which included CWC L. pertusa and M. oculata communities. In this regard, the observed impacts of trawling on CWC in La Fonera canyon should be a wakeup call for increasing the effort put on studying the distribution of these species in submarine canyons of the Mediterranean Sea and controlling fishing activities in the areas where pristine colonies can still be found.

Conclusions
Colonies of CWC M. oculata and D. cornigera have been observed in La Fonera canyon head, in the Northwestern Mediterranean Sea, by means of ROV video imaging. These positive identifications extend further west and south previous observations in submarine canyons of the Gulf of Lions, particularly Cassidaigne, Lacaze-Duthiers and Cap de Creus canyons, thus defining an extensive CWC province in the Mediterranean Basin. M. oculata was imaged in 12 out of a total of 21 transects and D. cornigera in 14. M. oculata forms, on the walls of Cap de Begur branch, dense middle-sized healthy colony clusters hosting a wide variety of fauna, whereas in other locations within the canyon head colonies are relatively smaller and more isolated. D. cornigera colonies are small-sized and isolated. Red coral C. rubrum also is present in La Fonera canyon head, where it has been imaged in 10 ROV transects. L. pertusa has not been observed in any of the transects.
Benthic terrain modelling based on swath bathymetry data has allowed classifying the different environments of the canyon head, and identifying in which terrains the different species concentrate. M. oculata and C. rubrum grow mostly on rocky near-vertical walls, the later at shallower water depths, whereas D. cornigera prefers sedimented or rocky steep slopes. The three species have been observed to live together. Maximum entropy prediction of habitat suitability indicates that out of a study area of 6.94 km 2 between 75 and 410 m water depth, 0.36 km 2 yield probabilities over 70% of occurrence of M. oculata, 0.43 km 2 of D. cornigera and 0.15 km 2 of C. rubrum. Nevertheless, due to hydrodynamics and other conditions beyond canyon morphology, not all suitable modelled habitats are occupied by CWC. Particularly, corals are much scarcer in Illa Negra branch than in Cap de Begur branch, the former incised much closer to the coastline.
Litter items resting on the seafloor have been observed throughout the canyon head whereas direct impact by bottom trawling is evident on the upper eastern wall. Fishing items such as ropes, nets and longlines are entangled on CWC colonies, and detached coral fragments and coral rubble have been imaged at the foot of damaged colonies. Other colonies are partially covered by fine sediment, possibly as a result of enhanced sediment fluxes due to daily resuspension by bottom trawling. The characteristics of the CWC community in La Fonera canyon are indicative that it withstands a high environmental stress, of both natural and human origin. This study represents the first step of a necessary assessment of the distribution of CWC along one of the largest submarine canyons in the entire Mediterranean Sea and highlights the need of identifying and mitigating human impacts to these unique habitats.