Quantifying the effects of fragmentation of connectivity networks of deep‐sea vulnerable marine ecosystems

Protection of vulnerable marine ecosystems (VMEs) in the high seas has focussed on identifying concentrations of indicator species and prohibiting the operation of bottom‐contact fishing gears where those occur in significant concentrations. Most such species have planktonic larvae and depend on dispersal networks for inter‐generational persistence. Yet, connectivity amongst patches of VME has seldom been considered when spatial management measures are introduced. Here, the relative importance of individual patches for the maintenance of their connectivity networks is evaluated, and a prioritization scheme for management action is proposed. Effective conservation measures should maintain approximately natural network configurations whenever possible.


| INTRODUC TI ON
The distributions of benthic invertebrates, including those in the deep sea, comprise high-density patches set amongst low-density and uninhabited areas.The patchiness is driven, in part, by environmental heterogeneity, particularly of seabed substrates, but also by, amongst others, inter-and intra-specific interactions such as competition and predation (Ovaskainen et al., 2017).Especially for sessile taxa, connectivity amongst patches depends on exchanges of larvae, with persistence across inter-generational time being dependent on supply from upstream (Fontoura et al., 2022;Taylor et al., 1993).In offshore marine systems, larval transport is primarily passive, dependent on large-scale ocean currents, though some larvae are capable of controlling their vertical movements in the water column.The duration of planktonic stages, before settlement to the seabed, differs amongst species and, in combination with water velocities, constrains connectivity distances (Álvarez-Noriega et al., 2020;Cowen & Sponaugle, 2009).Thus, certain high-density source patches can be critical for the persistence of others downstream, while maintenance of the spatial configuration of distributions and preservation of connectivity pathways are important precautionary conservation objectives (Fontoura et al., 2022).
The loss or severe degradation of particular patches, sufficient to disrupt connectivity networks and isolate downstream areas, can arise through various processes, including natural events such as submarine slumping (Paull et al., 2021), heat stress (Piazzi et al., 2021) and epizootic infections (Hamilton et al., 2021).The most pervasive threat, however, is from bottom-contact fishing, which is often conducted across broad swaths of seabed (e.g., Koen-Alonso et al., 2018), and continued for long periods-increasing the potential for extirpation (Thompson et al., 2019).Bottom trawls, in particular, can remove a high proportion of the emergent epibenthos in their paths (Pham et al., 2019;Wassenberg et al., 2002), with detectable impacts on habitat quality and species presence (e.g., Clark et al., 2016;Downie et al., 2021;Murillo, Kenchington, et al., 2020;Murillo, Weigel, et al., 2020).This potential for harm has generated international conservation concern, leading to United Nations resolutions calling for the protection of vulnerable marine ecosystems (VMEs) in the high seas from significant adverse impacts of bottom-contact fishing (FAO, 2009(FAO, , 2016)).VMEs are identified by the abundance of indicator taxa (FAO, 2009), many of which create biogenic habitats for other species (e.g., Beazley et al., 2015;Hawkes et al., 2019;Henry & Roberts, 2017), increasing their importance for biodiversity conservation.
In the northwest Atlantic, responsibility for implementation of the UN resolutions lies with the Northwest Atlantic Fisheries Organization (NAFO), whose mandate is to effectively manage and conserve the high seas fishery resources and to safeguard the ecosystems in which those resources are found.NAFO has identified VMEs associated with seven groups of indicator taxa: large sponges, sea pens, small and large gorgonian corals, erect bryozoans, tunicates and black corals (NAFO, 2023).The component species in each group (Table 1) share morphological, phylogenetic and functional traits, many of which are consistent with susceptibility to habitat fragmentation (Yeager et al., 2020).Sponge VMEs are characterized by high biomass of large structure-forming demosponges (Order Tetractinellida, Suborder Astrophorina) that reach sizes of more than 25 cm diameter and constitute more than 99% of the total invertebrate biomass over extensive areas (Murillo et al., 2012).Sea pens are found in association with mud bottoms (Murillo, Durán Muñoz, et al., 2011), while gorgonian corals (order Alcyonacea) are found in various habitats in this region, with smaller species on soft bottoms and larger ones on rock (Murillo et al., 2016;Murillo, Durán Muñoz, et al., 2011).Five species of black coral have been documented (Kenchington, Lirette, et al., 2019).A single species dominates the bryozoan turf (Murillo, Kenchington, et al., 2011), while the tunicate beds comprise primarily Boltenia ovifera, a habitat-forming stalked species (Francis et al., 2014).All seven groups form biogenic habitats, though none include reef-forming species.
In all, 84 patches of VME are known within the fishing 'footprint' in the NAFO Regulatory Area (Kenchington, Lirette, et al., 2019;Kenchington, Wang, et al., 2019).Beginning in 2008, NAFO has incrementally implemented 15 area closures targeted to protect some of those VMEs from bottom-contact fishing (NAFO, 2023), but not all known patches have yet been protected, and only three are entirely so.There have been no targeted closures protecting bryozoan or tunicate patches.Thus, there are residual concerns over the vulnerability of the connectivity networks for these taxa to the degradation of unprotected patches, the loss of which might undermine existing conservation efforts.Identification of critical nodes in those networks would aid in guiding additional protection measures.
The field of landscape ecology has suites of methods for characterizing connectivity networks and assessing the impacts of habitat fragmentation.Once a network has been described, uniting patch distribution and dispersal distances (Calabrese & Fagan, 2004), graph theory and network analysis can together identify critical nodes (or patches) with high centrality, meaning connections to multiple other patches.The average number of connections per node, termed a network's 'degree', is one of its most important attributes (Arancibia & Morin, 2022) and is readily calculated.
Additionally, the contributions of individual patches to landscapelevel connectivity can be examined through changes in network metrics following the simulated removal of particular nodes (Bodin & Saura, 2010;Keitt et al., 1997).Various metrics are available for that experimental approach, each of which considers such properties of patch configuration as total habitat area, patch size, numbers of patches, lengths of patch edge and patch isolation (Andrén, 1994;Calabrese & Fagan, 2004;Gustafson, 1998;Hargis et al., 1998;Keeley et al., 2021;McGarigal & Marks, 1995;Rutledge, 2003).
Indices that are responsive to the extent of the areas of patches, within viable dispersal distances of others, best capture patch connectivity (Bender et al., 2003;Fahrig, 2013;Fortin & Dale, 2005;Rutledge, 2003;Taylor et al., 1993).Some of those approaches have been applied to marine ecosystems (e.g., Abecasis et al., 2023;David et al., 2022; Ospina-Alvarez TA B L E 1 Extent of suitable habitat and network characteristics for each of seven groups of deep-sea benthic invertebrate taxa forming vulnerable marine ecosystems., 2020;Thomas et al., 2014;Treml et al., 2008), but their application to the deep-sea benthos over spatial scales of hundreds of kilometres remains challenging, requiring trade-offs between the information content of available metrics and their data requirements (Calabrese & Fagan, 2004).Habitat patches of VMEs in the deep sea can neither be precisely localized nor well characterized.Sampling coverage is inevitably poor, hence unrecognized 'ghost' patches may be present, especially where seabed conditions prevent deployment of standard survey gears.Moreover, the biology of the planktonic larval stages of most species is poorly known (Kenchington, Wang, et al., 2019).Conversely, dispersal vectors can be modelled with Lagrangian particle tracking (LPT) methods, in which virtual particles are released into the flow fields extracted from numerical ocean models (Cowen & Sponaugle, 2009;Lange & van Sebille, 2017).That approach has successfully identified connectivity pathways in the deep sea, including their sometimes complex curvilinearities (e.g., David et al., 2022;Wang et al., 2020Wang et al., , 2022;;Wang, Kenchington, Wang, et al., 2021;Xu et al., 2018;Zeng et al., 2019).It can also be tailored to specific spawning seasons and larval depth distributions (Kenchington, Wang, et al., 2019;Wang et al., 2020).While the constraints prevent the application of more advanced metrics, where suitable data are available, simple connectivity measures can provide a coarse-grained estimate of network coherence (Calabrese & Fagan, 2004;David et al., 2022).One such measure, the proximity index PX (Gustafson & Parker, 1994), represents the spatial context of habitat patches in relation to their neighbours.It only requires data on patch area and minimum connection distance, and has been shown to be responsive to the isolation of patches within the same spatial extent (Gustafson & Parker, 1994), suggesting that it is a suitable metric for detecting network fragmentation (Rutledge, 2003).
Application of PX to the connectivity networks of VME in the northwest Atlantic is facilitated by prior studies of the benthos in the NAFO Regulatory Area, which have largely been based on data from routine random-stratified bottom-trawl surveys.The species present have been identified (Murillo, Durán Muñoz, et al., 2011;Murillo et al., 2012Murillo et al., , 2016;;Murillo, Kenchington, et al., 2011) and the patches of VME indicator taxa mapped, using kernel density estimation coupled to a spatial-expansion evaluation (Kenchington et al., 2014).Habitat suitability models, predicting the distributions of the indicator taxa from a suite of oceanographic and terrain variables (Kenchington, Lirette, et al., 2019), allow partitioning of the region into suitable and unsuitable habitats.Connectivity amongst the network of closed areas has been examined with a 3-D particle tracking algorithm (Wang et al., 2020) coupled to oceanographic models validated by regional observations (Wang et al., 2018;Wang, Brickman, et al., 2019).Finally, the intensity of bottom-contact fishing activity during 2010-18 has been spatially resolved from vessel monitoring system data (NAFO, 2019) and can be used to assess matrix condition (Ramírez-Delgado et al., 2022).
Here, we build on that foundation by quantifying the relative

| Study area
The study area (Figure 1), in international waters off Newfoundland,

| Characterization of connectivity networks
For each of the seven groups of VME indicator taxa, the study area was partitioned into subareas of suitable and unsuitable habitat (Supplementary Text, Figures S1-S7).For each taxon group other than the small gorgonian corals, suitability was defined as predicted presence of the indicator taxa derived by random forest models, as previously reported by Kenchington, Lirette, et al. (2019).For the small gorgonians, the spatial extent of biomass modelled from kernel density analysis (Kenchington, Lirette, et al., 2019) was taken as representing occupied, and therefore suitable, habitat, with the remainder of the study area assumed unsuitable.
The locations and extents of extant patches of each VME type were drawn from previous kernel density analyses of biomass records of catches of their indicator taxa taken by routine researchvessel bottom-trawl surveys (Kenchington, Lirette, et al., 2019).
Delineation of the habitats was determined through an aerialexpansion technique (Kenchington et al., 2014) that identified the catch-weight threshold distinguishing high-density patches from the presence of isolated individuals and small patches in the habitat matrix (Supplementary Text).
The potential for disturbance of the seabed by bottom-contact fishing, during 2010-18, was determined for the area of suitable habitat for each taxon group, for each extant patch, and for a 20 km buffer area surrounding each patch.In each case, data on fishing activity drawn from vessel monitoring system records were mapped

| Particle tracking
Minimum inter-patch distances and connectedness degree were estimated using 3-D LPT biophysical models to represent larval transport pathways (Wang et al., 2020;Wang, Kenchington, Murillo, et al., 2021).Climatological monthly average current velocities, for 1990-2015, were extracted from the Bedford Institute of Oceanography North Atlantic Model (BNAM; Wang et al., 2018;Wang, Brickman, et al., 2019), which has a nominal resolution of 1/12° in latitude and longitude (Figure 2a).For those taxon groups for which information on spawning season was available (Kenchington, Wang, et al., 2019;Lacalli, 1980;Powell, 1968), transport pathways were modelled using either the average currents for individual spawning months or else grand means calculated from monthly averages for relevant months.Thus, spawning of various sea pen species (Table 1) was represented by three seasons (spring: April, May and June; summer: July, August and September; winter: January, February and March).Bryozoan larval transport was modelled separately for each of July, August and September, as that of tunicates was for each of January and February.Spawning of sponges was separately modelled for each month of the year.For the other three taxon groups, larval transport was modelled using the grand mean of velocities calculated from all 12 monthly averages.
Following the optimization work of Wang, Wang, et al. (2019), Wang et al. (2020), for each of the seven VME taxon groups, virtual particles were seeded into the modelled water flows at the seabed, within each of that taxon group's patches (Figure 2b, Tables S1-S7).
In general, a single particle was seeded every 1 km across each patch, following a uniform spacing, ensuring full coverage of the patch, as edge-to-edge connections may not provide minimum distances in a 3-D flow field.For some smaller patches, the spacing was reduced to ensure that at least 45 particles were seeded in each patch in each spawning season.Numbers seeded varied from 46 to 9253 particles per patch, per model run (Tables S1-S7).
Inter-patch distances were calculated as summations of the distances moved in each 10-min time step of the model from particle release until reaching another patch of the same taxon group (Figure 2d).A single minimum downstream inter-patch distance for each connected pair of patches was extracted from amongst the suite of models, for the relevant taxon group (Figure 2e).Those distances were compiled into connectivity matrices, which were examined and the number of connections for each habitat patch was determined.
Network connectivity degree, for each taxon group, was calculated Flow chart illustrating the steps involved in performing the Lagrangian particle tracking modelling.
as the average number of connections per patch (Arancibia & Morin, 2022), while the indegree (incoming connections) and outdegree (outgoing connections) were summarized for each patch.The proportions of released particles reaching another patch within each taxon network were averaged across all model runs.The networks were visualized with VOSviewer v 1.6.19(https:// www.vosvi ewer. com/ ), with nodes drawn in proportion to the patch area and positioned using the geographic coordinates of the patch centres.

| Habitat configuration index
PX is defined as: where S i is the area of patch i, in a network of n patches, and z i is the distance from patch i to its nearest neighbour (Gustafson & Parker, 1994).
Thus, PX has linear dimension and was here quantified in units of kilometres.It was evaluated for the networks of each of the seven groups of VME indicator taxon groups.

| Simulation of patch loss
For each taxon group, simulated networks were generated by systematic removal of one patch per simulation, with replacement.PX was recalculated for each simulation, with its values presented as percentage declines from the corresponding baseline value (intact network).For each taxon group, the mean decline in PX, averaged over all simulations, was calculated as an indicator of the vulnerability of the connectivity network to fragmentation.
Anomalous relationships between patch area and the mean decline in PX were explored through linear regression and analyses of residuals, performed in JMP 15.1.0(SAS Institute Inc., Cary, NC, 1989-2021).Clusters of responses to the simulations in the networks were identified with Jenks natural breaks classification (Jenks, 1967), applied to the mean per cent PX declines.Goodness of variance fit (GVF) was used to evaluate results with two to eight classifications, a range deemed useful for interpreting our results for management actions.Selection of the best scheme was based on GVF and an arbitrary minimum of >6 samples in a class.

| Network characterization
Full results from the LPT modelling are provided in Tables S1-S7, which present matrices of z and S for each taxon group and the proportion of modelled larvae showing connectance; in Figures S1-S7, which are maps of the patches and of the extent of suitable habitat within the study area; and in Figures S8-S14, showing the shortest particle trajectories connecting the patches.The seven taxon groups differed in the amount of suitable habitat available within the study area, from 14% for tunicates to 49% for bryozoans, as well as in the size and number of their patches, with only eight patches for black corals but 18 for tunicates (Table 1).Across all taxa, most of the 84 recognized patches were small, half being 25 km 2 or less in area, and only a quarter greater than 490 km 2 (Tables S1-S7).
Sea pens had the highest degree of connectivity (average number of connections per patch: 3.90), while black corals had the least connected network (average 0.75 connections per patch) (Table 1, Figure 3).
Patches serving as source populations to multiple other patches (outdegree > 0) were prevalent in the sea pen network, in which every patch was a source to at least one other, and SP1 had downstream connections to all other patches (outdegree = 10; Figure 3, Table 2, Table S2).
Important source patches, especially SGC2, SGC6 and SGC7, were also identified in the small-gorgonian coral network (Table 2, Table S4).
Some patches were supported from multiple upstream sources, creating redundancy that builds resilience and stability into the connectivity network.For example, SP3 and SP4 each received particles from 8 of the 11 patches in the sea pen network (Table 2, Table S2).The least redundancy is seen in the black corals, in which very few of the eight patches were connected to others and only BC8 received particles from more than a single upstream patch (Figure 3, Table 2, Table S7).

| Habitat disturbance
Bottom-contact fishing activity was widespread within the extents of suitable habitat for each taxon group; tunicates were the most affected with 68% of the 1 km grid-cells in their habitat impacted during 2010-18, whereas sponge habitats, being deeper, saw the least activity (Table 3; Figure S15).There has also been bottom fishing within the patches of each taxon group, the prevalence again highest for tunicates, with 79% of the cells within their patch area impacted during 2010-18, and lowest for sponge patches (Table 3).
The latter will have seen a decrease in fishing over time, as more of their area has been protected through closures but, to date, the patches of small gorgonian corals, tunicates and bryozoans have no targeted protections (NAFO, 2019(NAFO, , 2023)).The 20-km buffers around the patches also saw considerable fishing activity (Table 3).

| Habitat configuration
PX was largest for the sponges, at 155 km, and high also for tunicates, but very low for black corals and small gorgonian corals (Table 1).
Aside from the nine isolated patches, simulated removals necessarily led to declines in the index, though the reduction was very small in many cases (Figures S16 and S17).Conversely, the removal of some individual patches could have large effects; three taxon groups (large gorgonian corals, sea pens and tunicates) each showing more than an 85% drop in PX from the deletion of their largest patch (Table 2; Figures S16 and S17).
The positive relationship between the response of PX to the removal of a patch and the area of the patch removed, was statistically significant (R 2 Adj = .260,p < .0001; Figure 4), as expected from the relationship between area and distance in the PX formulation.This relationship was used to examine the residuals from the linear model fit to identify patches that, when removed, caused a disproportionately large decline in PX.Six 'outlier' patches were so identified.Two were patches of large gorgonian corals (LGC1 and LGC9), two were of small gorgonian corals (SGC3 and SGC4) and one each were bryozoan (BR1) and tunicate (TU1) patches.
Conversely, the largest sponge patch (S1) had less influence than expected for its size.

After examination of the partitioning of declines in PX using
Jenks natural breaks classification (Table S15), a three-class scheme was chosen.The placement of each patch into its class is shown in Table 2.For the sponges, sea pens and black corals, all of the patches in Classes 1 and 2 currently have some degree of protection from disturbance by bottom-contact fishing by area closures (Table 2).

| DISCUSS ION
Persistence of the sessile benthos over the long term depends on larval supply, and hence on inter-patch connections.Habitat fragmentation has the potential to alter connectivity, affecting population dynamics and ecosystem functioning (Ewers & Didham, 2006;Tischendorf & Fahrig, 2000;Valanko, 2012;Yeager et al., 2020), and may lead to a loss of biodiversity (Crooks et al., 2017;Fahrig, 2003;Thompson et al., 2019).Moreover, climate change is expected to have synergistic effects with habitat fragmentation, compounding disruption to connectivity and potentially limiting species' adaptive responses (Alaerts et al., 2022;Fourcade et al., 2021;Opdam & Wascher, 2004).Hence, effective conservation measures should maintain approximately natural network configurations, in order to preserve their functioning, at least as a precautionary step pending better knowledge of the ecology of the vulnerable taxa.Although much has been done to protect VMEs in the High Seas, including by NAFO in the northwest Atlantic, the focus has been on amounts of habitat protected and, to date, network configuration and connectivity have not been considered.
The present study has been a first attempt to address northwest Atlantic VME network connectivity, in that context.As such, it cannot provide comprehensive, conclusive answers, but it has advanced understanding.Aside from the black corals, with their short larval durations resulting in multiple isolated patches, the connectivity networks all averaged more than one connection per patch, indicating a degree of redundancy in their networks.Across the taxon groups, the majority of patches, 67 of 84, were sources supporting other, downstream patches-indeed, for the sea pens and F I G U R E 3 Stylized connectivity networks produced with VOSviewer using default settings for (a) sea pen habitats and (b) black coral habitats, illustrating the most and least connected networks, respectively.Nodes are labelled by patch code and their size is proportional to patch area, within the taxon (Tables S1 and S2).Node position represents the patch centroid in geographic space.The unconnected black coral patch, BC5 (Figure S14), is not shown.Colours aid in visualizing connections.(They represent cluster groups based on association amongst nodes but are not interpreted.)Stylized networks for the sponges, large and small gorgonian corals, bryozoans and tunicates are shown in Figures S8, S10-S13, respectively.
TA B L E 2 Percentage declines in PX from simulated removal of individual patches, averaged across various particle tracking models, ordered by magnitude (see also Tables S8-S14) and taxon-specific grand mean (bold).LGC2 a
tunicates especially, multiple other patches.Conversely, few patches (and none amongst the sea pens or tunicates) had no downstream connections.In short, the existing networks are generally well connected and, by inference, those connections are likely important to the persistence of VME within the study area.The high values of the PX index for sponges and tunicates (155 and 73 km, respectively) suggest that the extant networks of those taxa are intensively connected, whereas those of small gorgonian corals and black corals (PX 3.3 and 9.8 km, respectively) were much less so.Whether the persistence of the former groups requires higher connectivity remains to be explored, but the latter pair appear particularly vulnerable to future network fragmentation.
Simulated removal of single patches found very considerable differences in their effects on network connectivity.Removal of many patches, 28 of the 84, had little effect on PX (declines of <0.05%), but in every taxon group, there was at least one patch that, when removed, reduced the index by more than 45% and, in some cases, by more than 85%.While PX cannot capture every important aspect of network configuration, it is clear that some patches, especially those that are here grouped into Class 1, are much more important to achievement of overall conservation goals than are others.
Thus, further explorations of connectivity, potentially also applying genetic methods (e.g., Taboada et al., 2023;Xuereb et al., 2018), and considering multiple generations and time scales (Boulanger et al., 2020), may provide a foundation for increased efficiency in management measures.
All Class 1 and Class 2 patches of sponges, sea pens and black corals are currently at least partly closed to bottom-contact fishing,  S8-S14).(c) Residuals ordered by taxon group (see colour codes in the legend) and by decreasing patch area within each taxon.
TA B L E 3 Proportions of 1 km 2 squares that experienced bottom-contact fishing during the period 2010-18, within the suitable habitat of the habitat matrix, the patches and areas of suitable habitat in 20 km buffer zones surrounding the patches for each taxon group, expressed as percentages.However, before greater reliance is placed on the present analyses, while advancing conservation measures, it must be emphasized that this is only a first approach towards examination of the connectivity of VME networks in the northwest Atlantic and its results carry several substantial uncertainties.The delimitation of the known VME patches is well founded and has been supported by additional surveys with underwater cameras (Kenchington et al., 2014) but despite a very high sampling intensity for a deepsea ecosystem (and indeed for many other ecosystems), the possibility remains that other, smaller patches that are important to network connectivity exist in areas of untrawlable bottom.There are particular data gaps along the southern, southeastern and eastern slopes of Flemish Cap (Kenchington, Lirette, et al., 2019).Also, scattered individuals and colonies outside any denser patches might contribute significantly to connectivity, providing 'stepping stone' links undetected by our approach.However, the intensity of the bottom-contact fishing (NAFO, 2019), which has occurred over more than a quarter of the suitable habitat, excepting that of the sponges, and in about 50% or more of the immediate area around the known patches, indicates a highly disturbed environment that likely has a compromised ability for larval subsidies.
Matrix condition (measured as the extent of high human footprint levels) is an important aspect exacerbating the impact of habitat fragmentation, and may be critical to habitat restoration (Ramírez-Delgado et al., 2022).Furthermore, the species addressed here are not endemic to the study area.All are found upstream, along the Labrador Shelf and Slope (Wang et al., 2020), and some also downstream to the west and southwest (e.g., Kenchington et al., 2022).
Should there be significant larval subsidies from outside the study area, it would alter the consequences of simulated removal of patches, especially the more northerly ones.Comparing the dispersal paths of those habitats at the edges of the study area (Figures S8-S14) with the position of downstream habitats in Canadian waters, only LGC4 and SGC1 are within reach of habitats outside of the spatial extent of this study and therefore may not be sinks but important source populations for downstream habitats (Kenchington et al., 2022).
In terrestrial systems, when simple distance-based configuration indices like PX are applied, inconsistent correlations of the distance between patches and the degree of connectivity have been reported (Ewers & Didham, 2006;Kindlmann & Burel, 2008;Moilanen & Nieminen, 2002).For some species, this is attributed to the choice in habitat selection conferred by active mobility, while others may be constrained in their movements by a combination of traits such as body size, trophic level, rarity, and the distances, nature and quality of the habitat matrix within the landscape (Ewers & Didham, 2006;Ramírez-Delgado et al., 2022).In the marine systems modelled here, connectivity is heavily dependent on water velocities, and genetic isolation by distance is commonly seen in marine benthic invertebrate populations (e.g., Wright et al., 2015).However, while LPT modelling better represents effective connectivity mediated through larval drifts, when compared to uninformed nearest neighbour distances between habitat patches (Wang, Kenchington, Murillo, et al., 2021), it still only generates approximate indications of the strength of those connections.Many uncertainties remain in, for example, the use of average water movements when modelling the rare and exceptional larval-settlement events typical of long-lived benthic species, while the spawning seasons and the durations of pelagic stages remain only poorly known for the taxa of concern here.
In the simulations presented, habitat fragmentation was modelled only by the removal of whole patches.The available fishing effort data at km 2 resolution lack sufficient spatial resolution for exploration of more subtle changes, such as quantitative depletion of the species within patches or erosion around their edges from fishing pressure, both of which are likely to occur (NAFO, 2020).
Access to vessel monitoring system tracks would facilitate such work in future.Finally, apart from the tunicates and bryozoans, the taxon groups recognized in this study each comprised multiple species.Although phylogenetically related and sharing similar biological traits (Murillo, Kenchington, et al., 2020), the members of each group are not identical and will respond differently to habitat fragmentation (Thompson et al., 2019).
These uncertainties do not negate our conclusion that the patches of VMEs within our study area comprise inter-connected networks, such that maintenance of connectivity should be assumed essential to the persistence of the patches and hence of the VMEs.Nor do they alter our observation that the extant patches in each network differ very considerably in their relative importance to connectivity.
Our identifications of which patches are of greatest importance to conservation goals are less than fully certain but they can serve as a starting point for focusing attention where it is most needed.

ACK N OWLED G EM ENTS
The authors are grateful to the members of the Northwest Atlantic importance of individual patches to the connectivity of each of the seven groups of VME indicator taxa.Connectivity networks are constructed using dispersal vectors from LPT modelling and characterized by their degrees, while relative patch importance is evaluated through sequential simulated removal and replacement of each patch, with the effects quantified by changes in the PX index.We relate our results to the distribution of fishing effort which may indicate whether the patches have been reduced in size and/or fragmented into small, closely spaced habitats and whether the edges have been modified.The implications of our results for effective area-based conservation measures for the protection of VME are considered.
delimited by the outer boundary of Canada's Exclusive Economic Zone (EEZ) and the 2000 m bathymetric contour.That depth marks the limit of the 'fishing footprint' recognized by NAFO, outside of which different regulations apply (NAFO, 2023).The area includes portions of fishery-management units denoted as NAFO Divisions 3LMNO.Within it, Flemish Cap, an isolated bank with a minimum depth of 122 m, is considered a discrete bioregion and ecosystem production unit (Koen-Alonso et al., 2019).Flemish Pass, a 1200 m deep channel, separates the Cap from Grand Bank.The latter, which comprises a separate production unit, extends into the Canadian EEZ, but the Nose and Tail of the Bank lie in the study area (NAFO, 2015).Three principal currents influence the region(Colbourne & Foote, 2000).Flowing from the north, the Labrador Current follows the shelf break around Grand Bank, while the deeper flow of Labrador Sea Water divides at the Nose of the Bank, most passing eastwards to the north of Flemish Cap and the rest southwards, through the Flemish Pass towards the Tail of the Bank, where it turns away to the eastward.The Gulf Stream and its accompanying Warm Slope Water flow eastwards south of the Tail, thence heading to the east of Flemish Cap, where they feed the North Atlantic Current.Near-bottom velocities, forced by steep bathymetric gradients, can reach 10 cm s −1 , even at depths of 2500 m(Wang & Greenan, 2014).
onto a 1 km grid(NAFO, 2019).The resulting distribution was united with map layers representing the various areas of interest, and the proportion of 1 km cells that had experienced fishing activity was determined, using geoprocessing tools in ArcGIS Desktop v.10.8.1.F I G U R E 1 Map of the study area showing the probability of occurrence of large sponges, drawn from a habitat suitability model (Knudby et al., 2013), overlain by the locations of patches of sponge VME (S1 to S9, outlined in red).A probability of .042based on data prevalence (Kenchington, Lirette, et al., 2019) was used to create a binary map of presence and absence (Figure S1) identifying the suitable and unsuitable habitat.The red curve indicates the boundary of the Canadian Exclusive Economic Zone.Black lines demarcate NAFO Divisions 3L, 3M, 3N and 3O.Depth contours are shown at 200, 500, 1000, 2000, 3000 and 4000 m.Map projection: NAD83 UTM 23.The inset shows the position of the mapped area relative to the North Atlantic.
-S7).However, some of those 'sinks' (patches S6, LGC4, SGC1, BC5 and TU5) are near the downstream boundary of the study area and may supply larvae to patches within the Canadian EEZ, whereKenchington et al. (2022) identified significant concentrations of the same taxa.Finally, nine of the 84 patches, including members of the large gorgonian coral, bryozoan, tunicate and black coral networks, were completely isolated (outdegree = indegree = 0; Results are colour-coded by classes defined using Jenks natural breaks classification.Class 1 is indicated in dark blue and signifies declines in PX of >58.81%;Class 2 is indicated in light blue and signifies declines in PX of <48.98% and >18.57%.All others are Class 3 (see Supplementary

F
I G U R E 4 (a) Relationship between patch area and the percentage decline in PX (averaged across particle tracking models) when the patch was removed in simulation modelling, including points for all 84 VME patches and a linear regression line.The area between the 95% confidence limits for the fitted line is shaded, with the 95% confidence curves for an individual predictor value shown by the dashed lines.(b) Plot of the residuals from the linear model fit against the predicted values for the decline in PX, with a histogram of the frequency of residuals shown to the right.The red lines bound outliers identified by studentized residuals greater than |2.0|.Patches falling outside those are labelled (Tables but one of the two Class 1 patches of large gorgonian corals remains fully open (LGC9), while there is little or no existing protection for the patches of small gorgonian corals, bryozoans or tunicates(NAFO, 2019).The analyses presented here indicate that priority should be given to protecting patches LGC9, SGC1, 3, 4 and 5, BR1 and 2, and TU1, conservation of which may be crucial for the persistence of their respective networks.As each of those three networks has already experienced extensive fishing activity, their habitat quality is likely already degraded, increasing the urgency of protection of what remains in their priority patches.

Fisheries
Organization (NAFO) Scientific Council Working Group on Ecosystem Science and Assessment (WG-ESA) for the discussions that led to the development of this manuscript.Funding for this project was provided by Fisheries & Oceans Canada (DFO) through the Competitive Science Research Fund project 'Ecological Assessment of Significant Adverse Impacts of Fishing in NAFO'.EU (Portugal and Spain) surveys were co-funded by the EU through the European Maritime, Fisheries and Aquaculture Fund (EMFAF) within the Spain Work Plan for data collection in the fisheries and aquaculture sectors regarding the Common Fisheries Policy.We thank T. Kenchington, M. King and R. Daigle (DFO) for providing comments on an earlier draft of this manuscript.