A First Approach to Assess the Impact of Bottom Trawling Over Vulnerable Marine Ecosystems on the High Seas of the Southwest Atlantic

Applied Microsystems SV Plus equipment. Data processing included the removal of anomalies and the necessary sound velocity corrections using the Kongsberg–Simrad swath bathymetric software package NEPTUNE. Valid data were gridded at 50×50 m cell size resolution on a SUN workstation. The seismic parametric system Topas very high resolution seismic profiles along all ship tracks. Sub-bottom penetration varied, according to the lithology, between 150 and 250 m. Morphometrical data were obtained using ArcGis (ESRI) and Fledermaus software (Interactive Visualization Systems [IVS]) to provide final 3D images of the seafloor morphology. this


Introduction
The Southwest Atlantic (SW Atlantic), corresponding to FAO Statistical Area 41, includes a total continental shelf area of approximately 1.96 million km 2 of which a large portion lies off the Argentine coast (the Patagonian Shelf) and extends beyond Exclusive Economic Zones (EEZs) in the region [1][2][3]. This area is therefore integrated in the Southeast South American Shelf Large Marine Ecosystem (SSASLME) [4,5]. Currently, this region is the only worldwide significant area for high seas (HS) fisheries not covered by any Regional Fisheries Management Organisation (RFMO) [3].
Several authors [2,3,[16][17][18][19][20][21][22][23] have studied the potential disturbance of the seabed by bottom otter trawls and the possible negative effects on the structure of benthic communities. In recent years, several resolutions of the United Nations General Assembly [24][25][26][27][28] on sustainable fisheries made a call to States and RFMOs to identify vulnerable marine ecosystems (VMEs) and determine whether bottom fishing activities would cause a significant adverse impact on such ecosystems.
Sensitive species such as deep-water corals and deep-water sponges are found throughout the world oceans. Thus, the importance of habitat-structuring organisms is not restricted to shallow water, but also to shelf-break, hydrothermal vents, seamounts, and even the once considered constant and uniform deep-sea basins. Deep-water corals are vulnerable organisms occurring in the upper bathyal zones throughout the world and threatened by human activities, particularly fishing and oil exploration [29][30][31]. Fishing has a significant adverse impact (SAI) on deep-water coral communities in all oceans [32][33][34][35], particularly in the Northeast and Northwest Atlantic [36][37][38][39][40], Northeast Pacific [41,42], and Southwest Pacific [43][44][45][46]. In the SW Atlantic, the HS are one of the areas where deep-sea science has, to date, not been very active.
Protection of VMEs is a significant element of the management framework for bottom fisheries in high seas areas of the world ocean and its identification for selecting suitable protection areas is a challenge that conventional fisheries science cannot alone solve satisfactorily. Instead, it requires a multidisciplinary approach [21,22,47]. From the point of view of management of bottom fisheries and the governance of high seas areas, the situation in the PS poses an added problem as there is no any RFMO in force [2]. In its 2014 report [48], the Global Ocean Commission (GOC) recognises that continued scientific research is necessary to assess the cumulative impacts of human activities on the high seas so that informed decisions can be made about reversing the degradation of the global ocean.

Materials and methods
In accordance with the aforementioned UNGA resolutions [24][25][26][27][28] and the FAO deepwater guidelines [54], the Spanish Institute of Oceanography (Instituto Español de Oceanografía [IEO]) conducted from October 2007 to April 2010 a series of 13 multidisciplinary research cruises on the HS of the SW Atlantic, to identify VMEs and to assess the potential interactions with fishing activities. This paper presents the results of the five first cruises, consistently with UNGA resolutions (paragraphs 80 and 83 to 87 of resolution 61/105 (2007) and paragraphs 117 and 119 to 127 of resolution 64/72 (2010) in [27,28], which support making publicly available information on interactions between bottom fisheries and VMEs in the HS.
The use of spatial management tools to preserve the marine biodiversity of species inhabiting the HS has been broadly discussed in recent years [55]. To make such spatial management possible, our immediate objectives are: assessing specific biodiversity (mainly describing new species to science); describing the different habitats, ecosystems and deep-sea geomorphological features identified; and analysing their interactions and relationships to protect the full range of potentially different habitats.
The explored area during the five cruises conducted between October 2007 and April 2008 (Table 1) was located on the southern part of the HS of the SW Atlantic, to the east of the Argentinian EEZ 200 miles limit and between 44° 40'S and 47° 51'S up to the 1500 m depth contour ( Figure 1). The rest of the study area (up to 42°S) was surveyed during the eight following cruises (October 2008-April 2010), but the analysis of the information concerning VMEs collected during those last cruises, is still ongoing. In the right image of Figure 1 a non coloured area in the shelf can be roughly appreciated around 45°30'S and between 60°00'W-60°40'W, for which it was not possible to collect multibeam bathymetry data (no data) due to bad sea state conditions. The exploration of this area was carried out during one of the cruises conducted in 2009. Nevertheless, this type of data is not relevant for the present study, for which several trawl and CTD stations allowed the collection of pertinent information. The blue lines in the left image of Figure 1 corresponding to the 600, 1000 and 1500 m depth contours.
Key concepts for definition of VMEs were applied according to the FAO International Guidelines for the Management of Deep-Sea Fisheries in the High Seas [54]. These guidelines classify marine ecosystems as vulnerable based on several criteria: (1) uniqueness or rarity; (2) functional significance of the habitat; (3) fragility; (4) life-history traits of component species that make recovery difficult; and (5) structural complexity. For an adequate identification of VMEs, the two approaches in operation since 2008 by the NAFO Scientific Committee and the NAFO Working Group on Ecosystem Approach to Biodiversity in Ecosystems -Linking Structure and Function 4 Fisheries Management (WGEAFM) were applied in this study [56,57]: (1) the examination of cumulative catch data by ranking the biomass of VME taxa in each trawl from lowest to highest and then plotting the increase in cumulative biomass with each additional trawl; and (2) the use of Geographical Information System (GIS) to map the density of vulnerable species and groups' by-catch [58].
The study area included part of the outer shelf and upper and middle slope of the PS and was divided into thirteen depth strata in order to obtain a higher resolution in the description of vulnerable organisms. The research cruises involved five scientific disciplines: cartography, geology, benthos, fisheries, and hydrography.
This study used data from three main sources: i) Information from the five research cruises (geological, echosounder and oceanographic data; benthos and fish samples; fishery catch data [cpue]); ii) Data from commercial fishing activity collected by onboard scientific observers from 1989 to 2007 (fishery footprint); and iii) Commercial information on historical landings and effort data (provided by the Spanish fishing sector), as well as catch data for the main commercial species during the period 2000-2007 (logbooks filled in by captains of the fishing vessels, and provided by the Spanish General Secretariat for Fisheries [SGP]).
Geophysical and geological data were collected following internationally accepted standards and protocols for habitat mapping [20,59]. Full sea floor coverage using swath bathymetry provided a very high resolution of sea floor morphology. The backscatter data from multibeam echosounder together with high resolution seismic reflection profiles made available valuable data on the seabed sediments types. These data provided the geomorphological and acoustic basis to design a ground-truth planning strategy allowing for precise habitat mapping. Navigation during the surveys was via differential GPS Simrad GN33 using satellite corrections integrated into an inertial-aided Seapath 200 system. Swath-bathymetric data were acquired using a hull-mounted Kongsberg-Simrad EM 302 multibeam echosounder (288 individual beams, angular coverage up to 150°) operating at a frequency of 30 kHz. To correct the multibeam bathymetry, we carried out systematic casts of direct sound velocity profiles on the water column with an Applied Microsystems SV Plus equipment. Data processing included the removal of anomalies and the necessary sound velocity corrections using the Kongsberg-Simrad swath bathymetric software package NEPTUNE. Valid data were gridded at 50×50 m cell size resolution on a SUN workstation. The seismic parametric system Topas 18 produced very high resolution seismic profiles along all ship tracks. Sub-bottom penetration varied, according to the lithology, between 150 and 250 m. Morphometrical data were obtained using ArcGis (ESRI) and Fledermaus software (Interactive Visualization Systems [IVS]) to provide final 3D images of the seafloor morphology.
Samples of benthic fauna analysed in this study were collected with the Lofoten bottom trawl gear itself. Benthic fauna samples were sorted on board and preserved (70% ethanol or 4% buffered formaldehyde-seawater solution) for further identification analysis. Even if the bottom-trawl by-catch collected information did not allow for a detailed habitat mapping of VMEs, it provided a valuable indication of VME presence/absence that can be used to propose conservation measures, such as candidate areas for bottom fishery closures [23]. Sediment samples were collected using net collectors attached to the Lofoten fishing gear (Atlantis 2008 cruise) and with an USNEL type box-corer (BC) (maximum breakthrough of 60 cm; effective sampling area of 0.25 m 2 [50×50 cm]). A few samples were taken using a Bouma type box-corer (effective sampling area of 0.0175 m 2 [10×17.5 cm]). Both gears are designed to take undisturbed samples from the top of the seabed, and are suitable for almost every type of sediment. Sediment temperature and redox profiles (Eh) were immediately performed for the box-corer sample after each station. In the laboratory, the granulometrical analysis of the sediment was carried out by dry sorting the coarse fraction (>62 µm) and the sedimentation of the fine fraction (<62 µm). The organic matter content was assessed after calcinating (at 500°C for 24 h) and drying the sediment sample.
The hydrographical conditions in the studied area during the Atlantis 2008 cruise were characterised by means of a Seabird-25 CTD probe (SBE-25), equipped with oximeter, fluorometer and PAR detector. The survey schedule was optimized by systematically deploying the CTD at fishing stations below 500 m, but not always at greater depths. At each cast, the CTD was deployed to 5 m depth and stabilised for approximately 3 min. Once stable, the CTD was brought back to the surface and started profiling at a constant speed of 1 m⋅s -1 . The SBE-25 worked in auto-contained mode at a frequency of 8 scans⋅s -1 and the downloaded data were converted into physical units and pre-processed by using the SeaBird software (SeaSave/SBE DataProcesing-win32) with standard calibration values. Quality control and post-processing was performed with MATLAB.
Atlantis 2008 stratified bottom trawl survey enabled the assessing of the biomass and bathymetric distribution of the main commercial and most abundant fishery stocks by means of the swept area method. The survey used a stratified random design with strata boundaries definedby latitude and depth ranges, depth strata 1-7 located south of parallel 45°S and depth strata 8-13 sited north of the referred parallel (Table 2). Scheduled fishing stations (hauls of 30 min) were performed using a Lofoten bottom trawl net fitted with a rockhopper mix train with bobbins and rubber separators, suitable for deep-water fishing over irregular bottoms. Mean trawl speed was of 3.2 knots and trawl direction followed the bathymetric profile in the upper slope, but was variable in the outer shelf and middle slope.
Data recorded by scientific onboard observers from 1989 to 2007 between latitude 42°S and 48°S were used for mapping only the Spanish fishery footprint, since fishing data of other fleets were unavailable to us. The IEO observers' program placed one observer per selected vessel to cover 12% to 15% of the whole fleet. Table 3 summarize the activities (number of hauls year -1 ) of the IEO observers on the HS of the SW Atlantic, were Divisions 42 and 46 correspond to the areas roughly around parallels 42°S and 46°S.
Data used for each fishing haul corresponded to the middle tow position, since it offers more relevant information than the initial or final positions. All middle tow positions were imported into ArcGIS 9.3 mapping software to plot all trawl tows as straight lines between the reported start and end positions. They were then exported to a grid of 5'×10' min blocks, and any block including at least two tows was retained for mapping the bottom trawl footprint. Proper identification of the areas where VMEs are present followed the methodology used by the NAFO in its Regulatory Area [60]. Threshold catches, defined as catch levels of significant concentrations of invertebrates to be considered as possible VME areas, were assessed by analysing the cumulative biomass frequencies. Cumulative catch curve method was chosen to calculate the threshold catch. The cumulative frequency was plotted for all capture sets where taxa, considered as vulnerable by the International Guidelines for the Management of Fisheries [54] and by the Convention for the Protection of the marine Environment of the North-East Atlantic (OSPAR), were identified. The threshold selection for each taxon was made on the basis of minimum/maximum catch, density and morphological characteristics. Once a location of significant concentrations of vulnerable organisms was defined (key location), a 2 nm radius buffer zone around it was drawn to provide a safe margin of error on site.
The Random Forest algorithm for classification (RF) was used to predict the potential distribution of vulnerable benthic species by rating environmental conditions on the basis of previous observations.
RF is a non-parametric statistical method for data analysis that makes no distributional assumptions about the predictor or response variables [61], showing high prediction accuracy classifying rocky benthic communities [62] and beating other methods commonly used for ecological prediction [61,63]; The algorithm calculate the suitability of a given habitat for a given species based on known affinities with habitat characteristics, stored as raster maps, and called independent ecogeographical variables (EGV). According to HSI values, a map of species' expected distribution is produced, a value ranging from 0 to 1 showing the probability that the habitat of a given location is suitable for the species occurrence [64]. Thus, for a particular location, high HSI values mean high chances of the species' occurrence. To perform this mapping, presence/absence data from different vulnerable benthic organisms found in the study area were used as dependent variables of the different EGV.
Gathering accurate sampling presence/absence data is a critical part of the study, since the absence of a species in a given location can be due to several reasons: the species is present but is not observed, the species is absent even though the habitat is suitable, or the species is absent because of the unsuitability of the habitat. Only the last reason is considered as a "true absence" [65,66]. As presence data were aggregated into one single group named "vulnerable organisms", the resulting HSI predicted the potential habitat of any of the considered vulnerable organisms in the HS of the SW Atlantic under study.
The RF method offers the possibility to calculate an accurate unbiased estimator, using Out-Of-Bag (OOB) observations as an internal validation data set [67], computed from the resulting confusion matrix [68]. Accuracy is the proportion of the total number of predictions that were correct and this accuracy indicator is offered to the user as a measure of the model's predictive performance. It is determined using the equation: Where TS is the number of truly suitable locations, i.e. suitable locations correctly classified by the model; TU is the number of truly unsuitable locations, in other words unsuitable locations that have been correctly classified; and N is the total number of observations. Biodiversity in Ecosystems -Linking Structure and Function 8 Data was analyzed using the R statistical software [69] and the "Random Forest" package [70] and predictions were exported to a shapefile format using the "maptools" package [71]. GIS visualization of results was performed using the ESRI ArcMap 10.0 software.
Selected environmental variables involved in the study included depth, slope, sea bottom temperature, substrate characteristics and topographic position ( Table 4). The topographic position was sorted into six categories: shelf (1), outcrop areas on the shelf (2), high slope (3), low slope (4), abyssal flats (5), and canyons (6).  CTD stations' sea bottom temperature data were interpolated for the whole area using the local polynomial interpolation function (LPI) implemented in the ArcGIS 10.0 software. Slope was derived from the bathymetry high resolution data, and after studying the semivariogram, substrate characteristics were interpolated from granulometrical measures for the whole area using a universal kriging interpolator (Unpublished).
All the explanatory data were extracted for the presence/absence data locations, subsequently exported and analysed with the R statistical software using the BIOMOD package [72]. Several presence/absence models were performed: Generalized Additive Models [73,74]

Geomorphology
Geomorphological and geophysical data from the five research cruises revealed that the outer shelf was mantled by 15 m high sand ridges, and was 60 to 67 m deeper than the maximum 120 m lowering of sea-level during the last glacially induced regression. This difference in depth indicates that the PS had experienced subsidence in the Holocene. These ridges are relict and were probably constructed during the post-glacial transgression by the north flowing Falkland (Malvinas) Current, since they are resting on shell layers of <35,000 to 11,000 years old [77].
The upper continental slope descends from the shelf break, located at depths from 200 to 750 m, and is scarred by iceberg plough marks whose orientation and morphology suggest that icebergs carried northwards by the Falkland (Malvinas) Current were probably responsible for this erosion during the last glaciations [78].
Scattered over the study area (south of 45˚S) we found pockmarks, carbonate mounds formed by deep-water corals, northwards furrows, areas of smooth topography and sediment waves indicating that deposition on this part of the middle slope is controlled by bottom currents [79].
Seven submarine canyons were identified on the middle slope surveyed (Figure 2). Canyons 1 to 6 were cut by turbidity currents, whereas canyon 0 resulted from the combined effect of turbidity currents and coalescence of pockmarks (formed by the expulsion of thermogenic gas). These gas and fluid seepages contributed to the formation of canyons and to the partial detachment of blocks from the canyon walls. Thus, the thermogenic gas responsible for the formation of the identified pockmarks on the middle slope could be deep-seated, probably related to the Falkland Rift Basin, north of the Falkland (Malvinas) Islands [80,81]. The association of gas seepage with deep-water corals has been reported by [82] in pockmarks off Brazil. If such association also occurs on the Patagonian margin, those communities may be quite widespread in our study area.

Benthic communities
Bathelia candida (Moseley, 1881) was found to be one of the main reef builder species in the study area, providing habitat for diverse associated fauna of sponges, crustaceans, echinoderms, molluscs, and other cnidarians. The benthic megafauna caught during the cruises included invertebrates as well as Phyla Chordata and Hemichordata. Phyla Cnidaria and Porifera were dominant in terms of biomass (46% and 30%, respectively [ Figure 3A]). The high abundance of Cnidaria is remarkable, since 33.7% of the biomass of this phylum corresponded to the Class Octocorallia, including significant groups such as gorgonians (sea fans), alcyonaceans (leather corals) and pennatulaceans (sea pens). In addition, the VMEs dominated by suspensivore and/or filter feeding organisms are habitats with high biodiversity and many resources. A large part of the benthic samples contained erect sponges, octocorals, colonial scleractinians, calcified antipatharians, and hydrozoans (Family STYLASTERIDAE), all of them slowgrowing organisms considered as vulnerable by the UN and the OSPAR standards (see Table  7).
Bathymetric strata differences clearly arise by comparing the composition of the sampled benthic megafauna (Figures 3B-F Figure 3E]) recorded high biomass and numbers of octocorals, sponges, colonial scleractinians (Bathelia candida), and large hydrocorals. Octocorals included colonies of various genera belonging to families PRIMNOI-DAE and ISIDIDAE. As aforementioned, the increase and proliferation of these species create complex 3D structures providing the ideal habitat for a wide range of organisms. In those strata, the large amount of filter feeders and suspensivore sessile organisms is an indication of the presence of unaltered, complex and structured ecosystems. In the future, ROV and other submersible camera systems could confirm these assumptions.
Strata 7 and 13 (1001-1500 m, [ Figure 3F]) were the most difficult ones for trawling. Numerous tows failed to produce valid results. In these strata, the highest proportion of animals was of benthopelagic crustaceans, usually making diel migrations, even though they were normally present on the seafloor. Benthic cnidarians were dominated by octocorals of the Order PENNATULACEA (Verrill, 1865), with a wide bathymetric distribution, adapted to live on soft substrates.

Sediments
Sediment data obtained during Patagonia 1207, Patagonia 0108, and Atlantis 2008 cruises showed that fine sands were generally predominant throughout the study area, with low contents of organic matter and sediment sorting varying from poor to moderately good. In more detail, the bathymetric sedimentary classification would be as follows: Depths <200 m: fine sand (mean diameter=210 µm) with low organic matter content (mean value=1.14 %), moderately sorted.
Depths from 201 to 400 m: fine sand (mean diameter ranging from 150 to 189 µm) with low organic matter content (mean value=1.06%), moderately well sorted.
Depths from 401 to 700 m: very fine sand (mean diameter from 110 to 120 µm) recording the highest organic matter content (mean value ranging from 2.23% to 2.35%) and also the highest percentage (up to 44.50%) of silt and clay (<62 µm). Sorting was poor to moderate.
Depths from 701 to 1500 m: fine sand sediments similar to those of the shallowest stratum (mean diameter 160 to 190 µm), with low organic matter contents (mean value ranging from 1.43% to 1.68%). Moderately sorted.
Depths >1501 m: the deepest stratum, located in the bottom of submarine channels and canyons, was characterised by the presence of heterogeneous sediments mainly composed of fine sand (mean diameter=200 µm), with low organic content (mean value=1.68%) and poor sorting. This stratum showed the highest percentage (up to 39.5%) of coarse particles (>500 µm).

Fishery footprint
The statistical analysis of the bottom trawl footprint plot generated with the georeferenced fishery data obtained by the IEO scientific observers (between 1989 and 2007, 9013 fishing operations) showed that most of the commercial hauls of the Spanish fishing fleet in the study area (99.85%) took place at depths below 300 m ( Figure 4). A First Approach to Assess the Impact of Bottom Trawling Over Vulnerable Marine Ecosystems on the High Seas… 13 http://dx.doi.org/ 10.5772/59268

Model selection
Predictive accuracy of the models was evaluated through multiple cross-validation procedures, splitting the original data three times into two random subsets for calibration (80% data) and evaluation (20% data). The mean area under the receiver operating characteristic (ROC) curve (AUC) obtained from the three repetitions served to assess the predictive performance index of the model. AUC ranks from 0.5 to 1, null accuracy or perfect accuracy of the model, respectively [83]. Table 5 shows the best predictive performance score of the RF model, which was subsequently chosen for vulnerable species modelling.

Variable influence
The Mean Decrease Gini method, implemented in the BIOMOD package, was used to measure the importance of the dependent variable. This exploration tool shows graphically the total decrease in node impurities from splitting on the variable, averaged over all trees. Thus, for classification, the node impurity is measured by the Gini index [72]. The higher the value in the X axis, the higher importance the indicated variable will have on the classification of the dependent variable. Figure 5 show that the topographic position is the main variable affecting the distribution of vulnerable organisms in the HS of the PS, followed by the slope and the sea bottom temperature. Comparatively, the sea floor granulometry has a negligible effect on the distribution of the vulnerable organisms.
In addition to this, it is possible to visualize how each environmental variable, independently from any other, influences the response variable using partial dependence plots [73], which graphically represents the relationships between each predictor variable and the predicted occurrence probabilities of the vulnerable organisms obtained from the RF model. Figure 6 show that bathymetry has a positive effect between 500 and 1000 m depth. Regarding the topographic position, the highest interactions with the presence of vulnerable organisms were observed in canyons (6), followed by abyssal flats (5) and the slope (3 and 4). On the shelf, only outcrop areas (2) were positively correlated with the dependent variable.    Table 6 shows the presence data of benthic vulnerable organisms from the 169 sampled locations. Predicted values were plotted to produce a habitat suitability map showing survey sampling stations with presence/absence data and the vulnerable organisms' probability of occurrence (Figure 7).

Conclusions
Multibeam acoustic data showed that the upper slope and uppermost middle slope were scarred by iceberg plough marks. The middle slope surveyed was entrenched by seven submarine canyons [78]. Pockmarks and other seismic and morphologic evidence of gas/fluids seepage were pervasive throughout the entire survey area and more intense in the southern middle part [80]. Water coral communities associated with those pockmarks could be quite extensive in the study area.
The highest benthic biodiversity was found between 800 and 1500 m depth. Biodiversity was higher along the continental margin (per an equal number of individuals, and in terms of abundance) than biodiversity found along the continental shelf. Our results have confirmed the existence of close ecological relationships between Patagonian deep-sea fauna and Antarctic fauna of shallow waters. Benthic megafauna collected included invertebrates, chordates, and hemichordates. There was a clear dominance in biomass and diversity of the Phyla Porifera and Cnidaria. Most species of these groups are considered as vulnerable according to UN and OSPAR criteria: sponges, octocorals, colony scleractinians, anthipatarians, calcified hydrozoans (Family STYLASTERIDAE), and erect bryozoans ( Table 7).
Shallow waters (<400 m) are the strata having sustained most of the fishing pressure for almost 50 years. Below 400 m we recorded the lowest biomass, abundance and diversity values, most likely due to this fishing pressure. In these strata we noted the presence of sparse organisms with erected growth, a high dominance of pectinid mollusks (Zygochlamys patagonica), and minor presence of species considered as indicators of VMEs.
Intermediate depths (401-1000 m) showed an important increase in number and biomass of vulnerable organisms, with outstanding numbers, densities and biomasses of octocorals, sponges, colony scleractinians (Bathelia candida) and big hydrocorals (Errina spp., Cheiloporidion pulvinatum [Cairns, 1983], Sporadopora sp., and Stylaster densicaulis [Moseley, 1879]). Also remarkable was the presence of sponges of the Family CLADORHIZIDAE (Dendy, 1922), a group of a great zoological importance because they are carnivorous and have developed a trophic adaptation to live in the ocean's depths.
The model accuracy is acceptable (0.876). Although the modelling' accuracy values were higher when considering each organism, this was an expected fact due to the different environmental preferences of the studied organisms. However, HSI mapping is a useful conservation management tool enabling an initial observation of how environmental conditions control the spatial distribution of vulnerable organisms in the study area. The research will proceed further when data from all 13 survey cruises undertaken in the area are analysed.
The main environmental conditions affecting presence of vulnerable organisms seems to be connected to the topographic position, slope and bathymetry. Sea bed granulometry appeared

Class Hexactinellida Schmidt, 1870
Euchelipluma sp.  to have a negligible effect on the presence of vulnerable organisms, contradicting published research results on this subject in other geographical areas, where substrate characteristics determine to a large extent the presence or absence of a particular benthic species [84][85][86].

Rossella antarctica
Our study only calculated the general trends of the granulometrical parameters, while bathymetry, slope and topographic position were variables derived from high resolution data, strongly correlated with the response variable. Therefore, local conditions are the main factors ruling the potentiality of a habitat to host benthic vulnerable organisms in the HS of the PS.
The use of the Random Forest model offers both higher classification accuracy and determination of variable importance, and more stability where small perturbations of the data exist [76]. RF is a predictive classification and algorithm that does not make any distributional assumptions about the predictor or the response variables. It also handles situations in which the number of predictor variables exceeds the number of observations, offering a powerful non parametrical alternative for ecological modelling [64].
The vulnerable species groups, communities and habitats described here are mainly distributed beyond the 500 m depth contour. The presence of organisms considered as vulnerable is almost negligible in the fishing area. This fact is almost certainly due to bottom trawling operations of international fleets taking place in the study area for nearly 50 years. Also, the fishing grounds are far away from the geographical location of the main geomorphological features such as canyons, trenches, gas and fluid seepages observed in the middle slope, and identified as potential sites for VMEs.
The fishery footprint plot shows that the historical activity of the Spanish bottom trawler fleet has been located in the shallowest depth strata, at depths not generally exceeding 300 m. On this basis we think that the adverse impacts of current bottom fishing activities on VMEs are negligible or small. However, the displacement of the fishing fleet to target deep sea species at greater depths (were the existence of VMEs has been observed) could have a negative impact on those ecosystems. With this in mind and following the FAO deep-water guidelines, the potential threat of such a fishing strategy should be assessed.
Apart from Spanish fishing fleet, other bottom trawling fleets from different nations (former Soviet Union, Poland, GDR, Bulgaria, etc) have been operating intensively in the SW Atlantic (including our study area) from mid 60's until mid 80's, both over the continental shelf and slope [87][88][89][90][91][92]. Even if no data were made available to us for assessing the eventual negative impact of these fleets on VMEs, some experiences in other geographical areas such as the North Atlantic, Southwest and East Pacific, seamounts off Tasmania, and waters off New Zealand [31,45,92], have shown that high fishing pressure exerted by a large number of bottom trawlers over a long period of time could relevantly affect these VMEs. We therefore think that probably the almost 50 years of intensive bottom trawling in this SW Atlantic area by the abovementioned fleets could have contributed to the low presence of VMEs in the study area at depths lower than 500 m.