Towards a scientific community consensus on designating Vulnerable Marine Ecosystems from imagery

Survey methodology

At the 15^th Deep-Sea Biology Symposium in 2018 in Monterey, California, multiple presentations based on image data, as well as discussions among a subset of the attendees, indicated there was a critical need for consensus criteria for identifying VMEs from imagery data. A working group was formed under the auspices of the Deep-Ocean Stewardship Initiative (DOSI) to focus on developing these criteria. To reduce their carbon footprint, the working group met remotely during 2019–2021 and collated global datasets to discuss their expert opinions on what they considered to be a VME. Discussions covered a comparison of VME indicators among regions, how presence of a VME can be recognized in a single image, the number of images needed to determine whether a site is a VME, and areal extent and thresholds. Individual scientists shared images of the seafloor in areas they were working and discussed attributes of the images that led them to the conclusion that an image did or did not show a VME (details below). Incorporating opinions from deep-sea experts and managers from 15 countries and images from around the globe, this study represents the consensus of the discussions undertaken and reviews the remaining open questions and challenges for using imagery to determine whether sites are VMEs.

Question 1: Which taxa are considered VME indicators?

First, a list of taxa already designated as VME indicators by RFMO/As was compiled. VME indicators are defined as those taxa that meet at least one of the FAO criteria (uniqueness or rarity, functional significance, fragility, life history traits that contribute to slow recovery, or areas of structural complexity-see Table 1 for more detail) and are therefore proxies for the possible presence of a VME (UNGA Resolution 64/72, 2009; FAO, 2009). It is important to recognize here that a VME is an ecosystem, not simply the taxa that provide habitat structure, and that using VME indicators as a proxy to determine areas to protect as VMEs provides a mechanism for protecting biodiversity overall.

The list of VME indicators currently identified by RFMO/As indicates that what is considered a VME indicator taxon varies considerably by region (Table 2). The cnidarian orders Alcyonacea, Scleractinia, and Antipatharia are the only taxa with members that are considered VME indicators across all RFMO/A regions. Additionally, many RFMO/As only include subsets of a given taxon, for example, a specific family or order within a class, while other RFMO/As include the whole class. For example, within the order Alcyonacea, there is some disparity on which taxa are included. Some RFMO/As (e.g., General Fisheries Commission for the Mediterranean (GCFM) and Convention on the Conservation of Antarctic Marine Living Resources (CCAMLR)) include all Alcyonacea, while NAFO only includes the alcyonaceans formerly known as Gorgonacea. Similarly, within the Scleractinia, the EU recognizes cup corals as VME indicators whereas other RFMO/As do not (e.g., NAFO; ICES, 2020).

In some cases, RFMO/As omit a taxon from a region because that VME indicator taxon has not been found in that region. While it is understandable that VME indicators are drawn from inventories of known taxa, large portions of many RFMO/A convention areas are unexplored, and further surveys may reveal missing taxa that are considered VME indicators in other areas. An example of this was the discovery of deep-sea scleractinian reefs on seamounts in the North Pacific (Baco et al., 2017), after over two decades of exploration in the region and speculation that seawater chemistry in the region would prevent reef formation (Guinotte et al., 2006). In other cases, taxa are present and known to be structure forming in that region and yet are not included as VME indicators, even though they should be based on the FAO criteria (Table 1). Examples include sponges, xenophyophores, bryozoans, and chemosynthetic ecosystem taxa.

Sponges are very abundant in the North Pacific (e.g., Krautter et al., 2001; Campbell et al., 2009; Parrish et al., 2015; Kennedy et al., 2019; Downey, Fuchs & Janussen, 2020), occur within heavily fished areas (Du Preez, Swan & Curtis, 2020), can be caught in the North Pacific Fisheries Commision (NPFC) convention area by bottom trawls with moderate frequency (Miyamoto & Kiyota, 2017), and at the time of submission of this manuscript are listed as VME indicators in every single RFMO/A region except the NPFC. This discrepancy has been identified by global experts (FAO, 2019) and contradicts the known ecological importance of sponges within the NPFC region (e.g., Convention on Biological Diversity (CBD), 2016) and the inclusion of sponges in the management plans of surrounding EEZ MPAs (Marine Conservation Institute, 2021).

Among the more unusual VME indicators are giant, sediment-agglutinating protozoans called xenophyophores. Recognized as VME indicators by NEAFC, NAFO, and CCAMLR, these large foraminifera can attain high densities (>1/m²) on hard and soft substrates on seamounts, continental margins, and abyssal plains (Levin & Thomas, 1988; Gooday, Aranda da Silva & Pawlowski, 2011; Amon et al., 2016) and are known to host snailfish embryos (Levin & Rouse, 2019) and diverse assemblages of invertebrates (Levin et al., 1986; Levin, 1991). They are extremely fragile and vulnerable to disturbance, dominate in the CCZ area targeted for polymetallic nodule mining (Gooday et al., 2017; Gooday, Durden & Smith, 2020), and occur within fishing depths (e.g., Levin et al., 1986). Thus, they meet several of the FAO criteria, but are not yet listed by most RFMO/As, despite being found globally (Ashford, Davies & Jones, 2014).

Bryozoans are another taxon that meet multiple FAO criteria, and while included by SPRFMO, NEAFC, and CCAMLR, they are overlooked by NPFC, and only one species (Eucatea loricata) is currently considered as a VME indicator by NAFO. Habitat-forming bryozoans can provide habitat for diverse assemblages at the centimeter-to-meter scale, with associated assemblages comprised of more than 130 non-bryozoan species including Mollusca, Annelida, Arthropoda, Cnidaria, Porifera, and Echinodermata (Schlacher et al., 2010; Wood et al., 2012; Lombardi, Taylor & Cocito, 2020). Biogenic structures formed by bryozoans can attain significant sizes (up to several meters high) and can extend over 1,000 km (Wood et al., 2012; Lombardi, Taylor & Cocito, 2020). They are common in the Southern Ocean, New Zealand, Australia, the North Pacific around Japan, the northern Mediterranean, Bahamas, North-East Atlantic, and the North Sea (Lombardi, Taylor & Cocito, 2020).

Similarly, structure-forming chemosynthetic ecosystem taxa (e.g., mussels, tubeworms, clams) are not recognized as VME indicators in all regions in which they are known to occur, e.g., SPRFMO currently does not recognize them despite the known presence of hydrothermal vents on the East Pacific Rise and in ABNJ sections of the Kermadec Ridge (reference map in Menini & Van Dover (2019)). (However, bottom trawling by SPRFMO nations does not currently occur in these areas of the high seas).

Further issues with VME indicator lists used for move-on rules result from them being based on fauna caught as trawl or longline bycatch, which may result in exclusion of the many smaller or more fragile taxa that are destroyed or washed out of trawls (Wassenberg, Dews & Cook, 2002; Heifetz, Stone & Shotwell, 2009; Jones & Lockhart, 2011; Pitcher, Williams & Georgeson, 2019). Ecosystems are typically considered on the scale of epibenthic megafauna, particularly those with well-understood life histories, while smaller scales and poorly understood taxa are currently overlooked. Infauna, all size classes of which include ecosystem engineers, are also not considered. Furthermore, most RFMO/As list higher taxonomic levels to simplify taxonomic identification for fishers or fisheries observers, which can result in both inclusion and exclusion of taxa unintentionally.

A final issue is geographic bias in contributions to VME assessments. Seafloor imagery being used to augment fishery-based designation of VMEs is more common in some regions such as the North Atlantic or NE Pacific, and extremely limited in much of the ABNJ in the Indian Ocean, South Pacific and South Atlantic (Howell et al., 2020). Additionally, not all ocean regions have RFMO/As with competence to manage seafloor impacts of fishing. The tropical East Pacific, SW Atlantic, tropical North Atlantic, the NE Indian Ocean (including the Bay of Bengal) are examples of large international seabed areas without such RFMO/As in place (Bell, Guijarro-Garcia & Kenny, 2019), and thus have not undertaken VME designation. Having a consistent set of VME indicators across regions could help to simplify the designation process in all areas, especially those where RFMO/A programs are still developing. Beyond fisheries regulation, having such a consensus could benefit spatial management efforts in other sectors, including regulation of deep seabed mining, and oil and gas exploration. Consensus lists could also broadly and proactively inform biodiversity conservation efforts in ABNJ.

Thus, based on these observations and caveats, the first recommendation is that there is a need to establish a consensus list of the key VME indicators across regions that is continually updated when new taxa and communities are encountered, or more scientific information is gained that allows new assessments against the FAO criteria, and that benefits from observations made using imagery.

Question 2: Can a VME be identified from a single frame?

In some RFMO/As (e.g., the NPFC), VME imaging surveys to inform fisheries are conducted with drop cameras, which may only provide a single image per lowering. Additionally, even in multi-image and video surveys, single images or single frames from video are often the sampling unit, and therefore the starting point for analyses (e.g., Rowden et al., 2017; Williams et al., 2020a). Thus, the next question to address was: can a VME be identified from a single image or single video frame? Even without a consensus on which taxa are VME indicators across regions, to resolve this question we used the list of VME indicators in at least one region (Table 2) as a starting point, in concert with the FAO criteria (Table 1). For this task, scientists from different regions each shared 3–5 images that they considered to be a VME, to determine whether others agreed with this assessment. From this qualitative exercise, it was concluded that in some cases, all could agree that a single image showed a VME. Examples of single image VMEs are included in Figs. 2–8. Common themes to the agreed frames were the presence of reef or of an octocoral or antipatharian garden (Figs. 2 and 3). It is known that both live and dead corals can be important habitat (Mortensen & Fosså, 2006) and that deep-sea scleractinian reefs are fragile and comprised of species with life history characteristics that make them vulnerable (reviewed in Clark et al., 2016; Rogers et al., 2018). For example, Fig. 2A depicts a well–developed scleractinian reef, that has a commercial fish sitting on it, while several other invertebrate species can be seen in the image, all in clear association with the coral structure. This image meets all the FAO criteria, and the group consensus was that this is a VME.

Other examples where there was agreement included:

• Figure 3A, an image of an octocoral and antipatharian garden, with high density of individuals, a diversity of species, and encompassing a relatively large area within the field of view.

• Sponge reefs and gardens (Figs. 4, 6C).

• A well-developed chemosynthetic ecosystem (Fig. 5).

• A high density of a single VME indicator (Figs. 2, 3B, 3C, 4A, 4B, 5A–5C, 6).

• Multiple VME indicators in the same image (Figs. 3A, 3D–3F, 4D, 5B, 5D–5F).

• Images that demonstrated an association of other megafauna with the VME indicator(s) (Figs. 2 A, 2D–2F, 3C, 5B and 5F).

• Visible evidence of spawning or use of the VME indicator(s) as a nursery habitat (Figs. 7A and 7B).

Assessments were not always this simple, however. For example, in the image in Fig. 8A, there is an antipatharian coral that is clearly acting as a habitat for other taxa, but it is just one large coral colony. The question of whether this image was sufficient to represent a VME generated considerable debate. Similarly, assigning an image or video frame of four Hyalonema sponges as a VME was debatable, without the contextual knowledge of this frame being part of a continual patch of these sponges (Fig. 8B). It is noted that larger fields of view at an oblique angle to the seafloor can help imply this context within a single image (e.g., Figs. 6A–6C).

Question 3: What criteria can we use to identify a VME from a single image?

Given the consensus that certain VMEs could be discerned from a single image, the next step was to assess which qualities led to the conclusion that it was an image of a VME? And relatedly, which of the FAO criteria (Table 1) can be captured in a single image to help make the determination? The images with agreed VME presence were reviewed to assess which aspects of the image contribute to this conclusion. Building on the points of consensus outlined in Question 2, a simple flow chart was constructed for designating VMEs from single images (Fig. 9). The first step in this chart is to assess whether there are VME indicators present, if not then the conclusion is that the image does not have evidence of a VME and more images from the site need to be evaluated. If VME indicators are present, the FAO guidelines state that “merely detecting the presence of an element itself is not sufficient to identify a VME” (FAO, 2009). Thus, the first step of the flow chart is the “element detection” step, and the second step and all subsequent steps introduce additional factors and decisions that aid in identifying a VME.

From this point, if any of the given steps lead to a “Yes”, then the image can be considered to represent a VME and no further steps need to be tested. For example, the next step is answering the question, “is there reef present”? If the answer is Yes, then the area in the image can be considered a VME. This can be a scleractinian reef or a sponge reef, as reefs meet most or all of the FAO criteria in Table 1 (see Question 2 above). The reef can also be alive or dead, because dead coral reef has been shown to host as much or more diversity as live reef in some areas (Mortensen & Fosså, 2006). Similarly, skeletal remains of sponges (dead reef, spicule mats, and body stalks) have been shown to provide suitable substrate for settlement of sponge juveniles and other benthic epifaunal or infaunal organisms, increasing local diversity (Bett & Rice, 1992; Beaulieu, 2001; Dunham et al., 2018).

If there is no reef present, then the next question to evaluate is, are there chemosynthetic ecosystem taxa present? What are considered VME indicators for chemosynthetic ecosystems varies regionally, some include endemic taxa, some specifically list bivalves, others decapods, and still others specifically list polychaetes (Table 2); however, any taxa that create structure would count. Like reefs, chemosynthetic ecosystems meet many of the FAO criteria including Structural Complexity—tubeworms and molluscs often harbor significant epifauna (Van Dover & Trask, 2000; Van Dover, 2002; Van Dover et al., 2003; Guillon et al., 2017; Gollner et al., 2021); Functional Significance—both vents and seeps have been observed to act as nurseries for chemosynthetic and non-chemosynthetic taxa (Gollner et al., 2021; Salinas-de-León et al., 2018; Sen et al., 2019; Turner et al., 2020; Fig. 7B); Uniqueness or Rarity, as they generally occur in discrete areas and exhibit regional endemism of fauna (Gollner et al., 2021); Fragility and Life History characters–at seeps, tubeworms have been documented to live for 200–800 years, fitting with the characteristics of being slow-growing and long-lived (Durkin, Fisher & Cordes, 2017).

If neither of the previous two conditions are met, the next step of Large/Old Individuals may capture one or several of the FAO criteria of: Uniqueness or Rarity, Functional Significance, and Fragility, and came from a discussion of size. More individuals of a smaller size fit into a given space than of larger individuals, hence single individuals alone might not meet other criteria in the flow chart. However, the large/old individuals may be unique or rare for the given location. From a functional perspective, large individuals may contribute disproportionately to reproductive success (e.g., Beiring & Lasker, 2000; Fountain, Waller & Auster, 2019; Beazley & Kenchington, 2012). It is also often the case that large individuals provide a habitat for many associated fauna (Buhl-Mortensen & Mortensen, 2004; Wagner, Luck & Toonen, 2012). In terms of fragility, it has been shown that most (or many) deep-sea corals have maximum longevities of tens to hundreds of years, and some corals may live for over 1,000 to 4,000 years (Roark et al., 2006, 2009; Prouty et al., 2017), which would make recovery impossible on a 5–20 year timescale, (which is the time frame for recovery established in the FAO Guidelines (paragraph 19)). Relatedly, if trawled, large individuals might make a significant contribution to meeting the weight thresholds for fisheries move-on rules. Thus, large or old individuals in an image should also result in a VME designation. As ‘large’ is a relative concept we suggest a benchmark of an individual (or discrete colony) of sufficient size to make it likely to be more than 100 years old. We note that in many cases, determination of what constitutes an “old” individual for a species will require previous information.

If none of these previous criteria are met, but there is a visible functional role, e.g., as a nursery, then the image can also be considered a VME (Fig. 7). The FAO criteria (Table 1) refer to VMEs as “areas or habitats” necessary for the survival, function, reproduction, or recovery of fish stocks…”. It has been documented for several species of corals and sponges that, for example, elasmobranchs and other commercially important species may attach or hide their eggs in them (e.g., Etnoyer & Warrenchuk, 2007; Busby et al., 2012; Baillon et al., 2012). Other examples include the Bathyraja deep-sea skate spawning ground described on Shiribeshi Seamount (Hunt, Lindsay & Shahalemi, 2011), which led to the designation of this seamount as an MPA, and the Muusoctopus octopus spawning site off Costa Rica (Hartwell, Voight & Wheat, 2018), and more recently on Davidson Seamount off California (King & Brown, 2019). Similarly, if juvenile or newly recruited fishes or other taxa are present in a site (Fig. 7), it could warrant VME designation. However, it should be noted that in many cases the nursery role of VME indicators is not apparent from survey images and requires finer-scale examination of specimens (e.g., Figs. 7C–7E; Baillon et al., 2012).

Also related to the Functional Significance criterion, and the Uniqueness and Rarity criterion, for the flow chart box of “Threatened Taxa”, the FAO Guidelines in paragraph 42 (Table 1) refer to VMEs as “areas or habitats … necessary for the survival, function, … reproduction, or recovery of … rare, threatened, endangered or endemic species.” “In the case of confirmed or likely rare or endemic species the presence of these species should be sufficient grounds to identify the area as a VME.” This guideline implies that areas where rare or endemic species have been found or are likely to occur should be designated as VMEs, irrespective of whether biogenic habitat or listed VME indicators are present. Many hydrothermal vent molluscs would fit into this decision criterion, with 72% of all vent mollusc species globally, listed as critically endangered, endangered, or threatened on the IUCN Red List (Thomas et al., 2021). Many of these species are not themselves listed as VME indicators, but instead occur as epifauna on the other larger vent taxa and chimney surfaces. Observations of these listed species would warrant a VME designation of an area regardless of the presence of other VME indicators.

Of course, the VME indicators themselves may be rare or threatened, e.g., the octocoral Isidella elongata Esper, 1788 is on the IUCN Global Red List as critically endangered along with nine other deep-sea coral species that are categorized as endangered and seven that are listed as vulnerable (IUCN, 2016). Thus, regardless of whether the threatened species is a VME indicator or not, or whether other VME indicators are present or not, according to the FAO criteria, the presence of a threatened species is enough to designate a VME area.

If none of the previous conditions are met, the last three flowchart criteria focus on the diversity and density of taxa in an image. The first criterion is whether there is a threshold number of VME indicators present in the image. The next applies to monotypic stands of VME indicators that meet a minimum density threshold. And the last step looks at the density of all taxa together to allow for the fact that there may be one or two individuals of one VME indicator and one or two of another, so the image would not meet either of the previous criteria, but combining them together can still meet the FAO criteria. The final step of the flow chart is that if the image meets none of these criteria, then that single image does not depict a VME. To be certain about the implied absence of VMEs in that area though, more images should be evaluated.

Question 4: What are the thresholds (density or diversity) that need to be met to characterize a single image as a VME?

The last three steps of the flow chart each include a placeholder “threshold” value. Within the FAO criteria, the definition of “Structural Complexity” for designating a VME is given as: “v. Structural Complexity–an ecosystem that is characterized by complex physical structures created by significant concentrations of biotic and abiotic features” (FAO, 2009). The term “significant concentrations” implies the need for a threshold value that qualifies as “significant”. For example, what is a high enough density of a VME indicator or diversity of species for a site to be considered a VME?

Ideally, defined thresholds would be based on in situ measurements to determine the functional significance of different densities of each taxon and the spatial extent of each species, since densities are taxon dependent and vary among regions due to abiotic conditions (e.g., depth, productivity regime) (OSPAR Commission, 2010a, 2010c, 2009). Currently however, only a few studies have started to quantify this in specific geographic regions and only for specific species (see section on “Towards Density Thresholds Related to Ecosystem Function”). Until more data are available, the next best approach is to address three questions: (1) What is the natural range of densities that VME taxa occur in? (2) What is the natural range of taxa richness that is observed in a single image? And (3) What portion of these ranges should be considered a VME? To address these questions, authors with available images each took 50 images at random from each of their study sites. These were selected from among images that included benthic megafauna. Using the list of VME taxa in Table 2, the number of each taxon in each image was counted. Then for each image, the expert gave their opinion on whether this image was a VME, in the format of ‘Yes’, ‘No’, or ‘Maybe’. Additional metadata collected include site name, depth, image area, % hard substrate, camera type, and camera resolution. Results were compiled from ten laboratories for 27 sites encompassing the North Atlantic, the North Pacific and the South Pacific (Table 3, Fig. 10 and Table S1), giving a total of 1,273 images. Work in the Papahānaumokuākea Marine National Monument was permitted under permit #PMNM-2014-028 and #PMNM-2016-021.

Data were analyzed as density and taxon richness per unit area using the image area to convert counts to density values. The range of image areas was large, 0.25–50 m² but with 38 images having an area <1 m². Analyses were therefore caried out on the 1,235 images of >1 m² area to avoid inappropriate density extrapolations. R code used for the analyses can be found at https://github.com/bexeross/Baco-et-al-Anova-plots.git.

As abundance generally decreases with increasing depth (Rex & Etter, 2010), it might be expected that density would also decrease with depth and could confound the data analyses. A regression of observed density with depth (image range 100–4,176 m) showed no relationship (p = 0.6678), so depth was not included in further analyses. Similarly, densities might vary on hard substrates vs soft substrates. A regression of percent soft substrate vs overall density showed there was also no difference in the observed values by substrate type (p = 0.3432).

With these potential confounding factors accounted for, the first threshold in Fig. 9 is for number of taxa. The observed range of VME taxon richness per image was 1–9 taxa, giving a range of 0.02–6.53 taxa per m² with a mean of 0.54 (see Fig. 11A for taxa per m² visualization and Table S2 for summary statistics for all tested categories). The range of observed taxon richness values per m² were overlapping for all of three VME designation categories. Regardless, a simple one-way analysis of variance (see Table S3 for ANOVA results for all analyses) showed that there is a statistically significant difference between VME designation choices (p < 0.001, Fig. 11A, although p < 0.01 for Yes vs Maybe, and Yes vs No see Table S4 for Tukey HSD results for all analyses). No had a mean of 0.45 ± 0.7 taxa per m², Maybe a mean of 0.98 ± 1.4 taxa per m², and Yes a mean of 0.69 ± 1.1 taxa per m². The disparity between Yes and Maybe values possibly reflect other factors such as which taxa are co-occurring, the size and density of the taxa in question, and whether all taxa are agreed upon across all RFMO/As.

Working at a coarse taxonomic level, for most taxa, there was a statistically significant difference in densities for Yes vs No. For example, alcyonaceans had a significant difference for all three choices (p < 0.0001, Fig. 11B) with the mean for Yes being 1.47 ± 2.7 individuals per m² and No being 0.29 ± 1.2 individuals per m².

For analytical simplicity, Scleractinian corals that form a reef structure were annotated as counts where feasible by detecting distinct coral heads or isolated colonies. However, percentage cover (not considered here) is also an important metric when quantifying cold-water coral reef VMEs. Scleractinians were further complicated by the fact that some regions consider cup corals to be VME indicators while others do not. As a result, values in the No category ranged up to 25.4 individuals per m². Despite the broad range of values, Scleractinians showed significant differences between Yes, with a mean of 3.59 ± 9.3 individuals per m², and 0.48 ± 2.18 individuals per m² for No (p < 0.0001, Fig. 11C, Tables S3 and S4).

Pairwise tests for sponges (Porifera) showed a significant difference between Yes and No (p = 0.0175, Fig. 11D, Tables S3 and S4), with Yes having a mean of 1.29 ± 3.7 individuals per m², and No having a mean of 0.58 ± 0.9 individuals per m². However, there was no significant difference between Yes and Maybe (mean of 1.48 ± 2.21). Overlap may have arisen because experts from different regions have different opinions on encrusting vs upright sponges, or on the critical density of sponges that constitute a VME. Sponges are also very diverse in shape and size with larger individuals potentially more likely to be associated with VMEs (e.g., NAFO lists large-sized sponges specifically, NAFO, 2021), however, larger taxa will also occur in lower densities which may further confound this assessment. Figures and analyses for additional taxa are provided in Fig. S1 and Tables S2–S4.

The final threshold in Fig. 9 is for the density of all VME taxa combined. The general test was significant (p < 0.0001, Fig. 11E and Table S3). The mean for Yes was 6.00 ± 11.5 individuals per m². The mean for No was 2.32 ± 8.6 individuals per m². However again there was no difference between Yes and Maybe (mean 5.81 ± 11.7, Table S4). This result may reflect the issues pointed out above where small individuals can be highly numerous but may not be considered a clear VME indicator to some experts and regions.

This is a coarse approach, but is a start for developing threshold density and diversity metrics for designating VMEs, and will require significant refinement before realistic or consistent thresholds can be established. However, from this exercise we can infer several points. The first is that there is generally a difference between the means for deciding Yes a site is a VME and No a site is not a VME, indicating that there is the potential to develop consistent thresholds across regions based on expert consensus. The density values for a Yes vs No for individual taxa, however, vary widely among groups, making it unrealistic to set a threshold density that will work across all taxa. Thus, taxon thresholds will need to be calculated on a taxon-specific and potentially a regional basis, with any calculations ideally taking taxon size into account. Our study sites were clustered in the North Atlantic and North Pacific, additional information from the South Atlantic, South Pacific, Southern Ocean and Indian Ocean is of paramount importance in order to form robust, evidence-based thresholds on both a global and regional scale (Fig. 10).

Areal extent

This effort is a work in progress and represents the first step in the process of determining an objective, quantitative method for identifying VMEs from images. However, many more questions need to be addressed to advance this process e.g., advancing beyond single images. While an image of ~10 m² (range of study images 0.25–50 m²) may show high densities of taxa, this tiny area relative to the total management area makes it challenging to use a single image to justify management actions such as area closures. Furthermore, some VMEs cannot be discerned from a single image alone due to the dominance of larger taxa or having lower natural densities (e.g., Figs. 8A and 8B). A greater number of images would have a better chance of capturing natural variability in distribution and density to both initiate management actions and ensure that areas dominated by large taxa or low-density VMEs are also identified.

On the positive side, single images and video frames are rarely taken in isolation, generally even drop cameras are deployed multiple times at a survey location. If multiple single images from a location are determined to depict VMEs, then there would be great support for management action. However, the issue remains that there is a need to determine how many images over what area need to be taken and what proportion of those need to depict VMEs. Thus, the next step proposed for this process is to develop standards for multiple image/video assessments to capture the areal extent of VMEs.

This is not straightforward, since setting thresholds for VME areal extent could be based on: VME community patch sizes (e.g., the average size of a reef), average densities of lower density VME indicators (e.g., Umbellula encrinus Linnaeus, 1758 giant sea pen communities in Norway may only reach densities of 6.4 individuals every 100 m²; Gonzalez-Mirelis & Buhl-Mortensen, 2015), minimum viable community patch sizes (e.g., >25 m² is suggested by OSPAR as the minimum areal extent for a biotope, OSPAR, 2008a, 2008b), management practicalities (e.g., utilizing current RFMO/A move-on rule distances, trawl haul areas, or minimum viable MPA sizes as buffer areas of search), or identifying geomorphological features that host VMEs (e.g., seamounts, which could be considered as VMEs themselves, sensu Watling & Auster, 2017).

Many of these approaches have limitations. A VME community patch size-based approach would need to recognize that the natural range of patch sizes appears to be both dominant-taxon- and location-dependent. For example, cold-water coral reefs dominated by Solenosmilia variabilis have been predicted to be between 625 m²–0.425 km² in size in New Zealand (Rowden et al., 2017) but 0.02–1.16 km² in Tasmania (Williams et al., 2020a). Boreal Ostur sponge aggregations in Norway have been measured at >50 km² with empty gaps of <30 m (Kutti, Bannister & Fosså, 2013), while glass sponge reefs in Canada may be only 35–72 m in diameter (Chu & Leys, 2010). Should a review of patch sizes be undertaken, potentially a minimum viable distance could be determined to identify the minimum number of images at X meters apart needed to be sure that management action could be worth initiating. However, the low-density VMEs that could not be identified from a single image may continue to be overlooked by failing to survey a wide enough area. Meanwhile, move-on rule distances or minimum viable MPA sizes are RFMO/A dependent, and geomorphological features may capture some VMEs, but soft-bottom VME communities cannot be delimited in this fashion (e.g., xenophyophore fields and sea pen aggregations). It is therefore likely that any VME areal extent standard that could be developed would need to be flexible, describing multiple search techniques and listing multiple criteria.

Confidence index

Another useful step would be to develop a Confidence Index e.g., tied to the number of images in an area that represent a VME. Confidence is commonly used to assess the accuracy or uncertainty of a method or product and can be assessed based on a range of factors, including data quality and data deficiency (Wallace et al., 2011). Morato et al. (2018) developed a multi-criteria assessment to evaluate the likelihood of VME presence in the North Atlantic. This approach was based on available VME indicator and habitat data from the ICES VME database (ICES, 2016), with outputs mapped on a 0.05° × 0.05° grid cell scale. As part of this method, a measure of confidence was included based on four criteria: the survey method (with visual surveys scoring higher than trawl surveys or other survey methods such as acoustic data); the number of surveys in the area (grid cell); the survey time period; and the age of the last survey in the area. Final scores were assigned to grid cells as either ‘High’, ‘Medium’ or ‘Low’ confidence, and could be mapped alongside the likelihood of VME presence to present a visual representation of outputs. Similar methods, along with an evaluation of the existing confidence index approaches, could be considered within a global standard for any VME assessment method.

Towards density thresholds related to ecosystem function

A key next step in developing VME indicator density thresholds is continuing the study of ecosystem function relative to VME indicator density. Links between structure forming VME indicator taxa and enhanced biodiversity (Beazley et al., 2013; Jonsson et al., 2004; Henry & Roberts, 2007; Price et al., 2019), fisheries species (Foley et al., 2010; Stone, 2014; Rooper, Goddard & Wilborn, 2019; Price et al., 2019; Henderson, Huff & Yoklavich, 2020) and ecosystem functioning (Kutti, Bannister & Fosså, 2013; de Clippele et al., 2021) have been well documented. However, few studies have quantified VME indicator density with associated diversity and ecosystem functioning. Understanding when VME indicator become dense enough to form an influential habitat (and presumably a VME), can help underpin marine spatial management solutions through objective definitions of habitats (Bullimore, Foster & Howell, 2013), and predicting the spatial extent of VME indicator taxa/habitat (Rowden et al., 2017; Williams et al., 2020a).

Most studies defining VME indicator density thresholds focus on natural density ranges and omit quantitative analyses of associated biodiversity or ecosystem function. For example, Vertino et al. (2010) used percent coverage to distinguish mound (40–66%) and inter-mound (6–9%) reef habitats with 20–40% coverage of live or dead coral delimiting “coral framework” in the Mediterranean Sea (Vertino et al., 2010). Rogers et al. (2013) suggested coral colony densities should reach >10 times background densities, and usually >0.1 colonies per m². Using this approach, Bullimore, Foster & Howell (2013) assessed the background density in their area of study and found a value of >0.47 colonies m² would be required to achieve >10 times the background density. However, this value is relative and therefore area specific.

Some studies have started to link ecological function and density thresholds though. Henry & Roberts (2014a, 2014b) in reviewing OSPAR Threatened and/or declining habitat definitions for coral gardens and deep sea sponge aggregations, assessed published data against a series of criteria including density and ecological function. Coral densities required that VME indicators were at least “frequent” on the SACFOR scale (Strong & Johnson, 2020) in an image, video, or sample, while ecological function was considered high when other species co-occurred in high frequencies, or non-coral taxa characterized at least 50% of the assemblages. Sponge aggregations, used OSPAR density thresholds, the SACFOR scale, or the NEAFC move-on threshold of 400 kg (NEAFC, 2014), while ecological function required presence of listed associated fauna (as outlined in OSPAR Commission, 2010b) or that a SIMPER analysis (Clarke & Warwick, 2001) highlighted other taxa as characteristic of the assemblage.

Additional literature has focused on connecting the density of VME indicators and the diversity of associated fauna. For example, Beazley et al. (2015) used imagery to identify that the largest turnover in megafaunal community composition in NW Atlantic sponge grounds occurred when the sponges reached 15 individuals m². Price et al. (2019) used 3D photogrammetry on a scale of tens of meters to link structural complexity and biodiversity, finding areas of high structural complexity and coral coverage above 30% harbored distinct and more diverse communities. While Rowden et al. (2020) posited thresholds for “significant concentrations” supporting “high diversity” in Solenosmilia variabilis reefs near New Zealand as 24.5–28% cover of framework-building coral or a density of “live coral heads” of 0.11–0.14 (over areas of 50, 25 m² in video) or 0.85 coral heads per m² (in 2 m² still images).

Imagery data is starting to play a critical role in assessing ecosystem function, but a greater number of studies need to be completed before threshold guidelines can be developed that directly tie species composition, abundance, or density to ecosystem function for most VME indicators. In the meantime, imagery data provides a more accurate picture of species composition, density, and functional importance than trawling surveys do, so images will be a more accurate way to determine VME ecosystem function-density thresholds.

Habitat suitability modeling

Another tool used for the designation of VMEs is habitat suitability modeling. Obtaining images or bycatch samples of VME indicator taxa may confirm their presence, but the proportion of the seafloor that has been observed or sampled to date is <0.001% (Stel, 2021). Habitat Suitability Modelling (HSM, aka Ecological Niche Modeling) is one way to fill in the gaps and provide an objective prediction of where VMEs may exist in the unexplored regions of the world’s oceans. HSM refers to the use of computer algorithms to model the mathematical relationship between occurrences of a species/habitat and its preferred environmental conditions such that its spatial distribution can be predicted in unsampled areas with environmental data (Vierod, Guinotte & Davies, 2014).

The use of HSM as evidence for the management of VMEs was endorsed in 2016 by UNGA resolution 71/123 (§180-181) and guidelines for their use remain under development by management bodies (ICES, 2021). To date, HSM has been used inter alia to provide a basis for spatial management planning (Rowden et al., 2019), estimate MPA effectiveness against percentage targets (Ross & Howell, 2013), cross-reference VME distribution and fishery activity (Jackson et al., 2014), locate potential higher density VME thresholds and hotspots (Rowden et al., 2017, Gonzalez-Mirelis et al., 2021), predict pre-fishing baseline densities of VME indicator taxa (Downie et al., 2021), predict potential changes in VME indicator distribution under climate projections (Morato et al., 2020), and to combine with dispersal estimates to identify isolated VME populations (Ross, Wort & Howell, 2019).

Most commonly, HSM is applied to species (aka Species Distribution Models, SDMs, e.g., for a VME indicator taxon) but these predictions do not necessarily capture the distribution of the community that taxon is associated with, nor the specific areas where a species may form more complex habitat, e.g., deep-sea scleractinian coral reefs (Howell et al., 2011). However, there are growing efforts to apply HSM to communities/biotopes (Ferrier & Guisan, 2006; Howell et al., 2016), to density of structure-forming taxa, (e.g., Rooper et al., 2016, 2018), or traits (e.g., Murillo et al., 2020), multiple taxa simultaneously (e.g., Joint Species Distribution Modelling, Warton et al., 2015), or to intersect multiple stacked SDMs (e.g., Lyons et al., 2020): approaches which may be better suited to capturing VME extent and distribution.

It is important to note that the modelling approaches described above are fallible, with issues stemming from sampling bias, positioning errors, combining datasets with different qualities, coverage and resolution of environmental data, taxonomic resolution, unaccounted for drivers/limiters of distribution, etc. all contributing to potential errors in model predictions (Vierod, Guinotte & Davies, 2014). Indeed, field validation studies have indicated deep-sea HSMs based on low resolution global bathymetry performed poorly (Anderson et al., 2016), highlighting the need for suitable high-resolution underpinning predictor datasets. Key elements of models to be useful for VME identification are to quantify uncertainty in available predictor variables, ensure variables are ecologically significant, use abundance or biomass data rather than presence/absence, and increase taxonomic resolution to ecologically relevant levels (Bowden et al., 2021). It is therefore necessary to request and utilize any information on model uncertainty to temper model interpretations. Ground-truthing of models using imaging surveys will also improve their accuracy (Rooper et al., 2016, 2018; Winship et al., 2020), or lead to their rejection for VME designation use (e.g., Anderson et al., 2016). A close interaction between modelers and stakeholders can help to ensure that HSM and other forms of modelling are properly applied and interpreted to benefit marine management needs (Villero et al., 2016).

Detailed analyses results from threshold investigations

All analyses were performed in R with the results forming the basis of threshold conclusions relating to Question 4 (What are the thresholds (density or diversity) that need to be met to characterize a single image as a VME?) R code used for the analyses can be found at https://github.com/bexeross/Baco-et-al-Anova-plots.git.

Supplemental Figure 1. Boxplots of YMN (Yes, Maybe, No) for Taxa per m² and additional taxa.

Supplemental Table 1. Raw data used for threshold investigations in Question 4.

Supplemental Table 2. Summary statistics for YMN (Yes, Maybe, No) decisions per category discussed in the text in Question 4 across all images with >1 m² area.

Supplemental Table 3. Anova (stats::aov) results tables for YMN (Yes, Maybe, No) decisions per category discussed in the Question 4 across all images with >1 m² area.

Supplemental Table 4. Tukey HSD (rstatix::tukey_hsd) results tables for YMN (Yes, Maybe, No) decisions per category discussed in Question 4 across all images with >1 m² area.