Climate‐induced changes in the suitable habitat of cold‐water corals and commercially important deep‐sea fishes in the North Atlantic

Abstract The deep sea plays a critical role in global climate regulation through uptake and storage of heat and carbon dioxide. However, this regulating service causes warming, acidification and deoxygenation of deep waters, leading to decreased food availability at the seafloor. These changes and their projections are likely to affect productivity, biodiversity and distributions of deep‐sea fauna, thereby compromising key ecosystem services. Understanding how climate change can lead to shifts in deep‐sea species distributions is critically important in developing management measures. We used environmental niche modelling along with the best available species occurrence data and environmental parameters to model habitat suitability for key cold‐water coral and commercially important deep‐sea fish species under present‐day (1951–2000) environmental conditions and to project changes under severe, high emissions future (2081–2100) climate projections (RCP8.5 scenario) for the North Atlantic Ocean. Our models projected a decrease of 28%–100% in suitable habitat for cold‐water corals and a shift in suitable habitat for deep‐sea fishes of 2.0°–9.9° towards higher latitudes. The largest reductions in suitable habitat were projected for the scleractinian coral Lophelia pertusa and the octocoral Paragorgia arborea, with declines of at least 79% and 99% respectively. We projected the expansion of suitable habitat by 2100 only for the fishes Helicolenus dactylopterus and Sebastes mentella (20%–30%), mostly through northern latitudinal range expansion. Our results projected limited climate refugia locations in the North Atlantic by 2100 for scleractinian corals (30%–42% of present‐day suitable habitat), even smaller refugia locations for the octocorals Acanella arbuscula and Acanthogorgia armata (6%–14%), and almost no refugia for P. arborea. Our results emphasize the need to understand how anticipated climate change will affect the distribution of deep‐sea species including commercially important fishes and foundation species, and highlight the importance of identifying and preserving climate refugia for a range of area‐based planning and management tools.


| INTRODUC TI ON
The deep sea represents at least 95% of the ocean and plays a critical role in climate regulation through uptake and storage of heat and carbon dioxide (Purkey & Johnson, 2010;Sabine et al., 2004).
However, changes linked to these regulating services have consequences for the health of the ocean including warming, acidification, and deoxygenation of deep waters, leading to decrease in food availability at the seafloor (Bindoff et al., 2019;Chen et al., 2017;Gehlen et al., 2014;Mora et al., 2013;Perez et al., 2018;Sulpis et al., 2018;Sweetman et al., 2017). Recent projections of deep water mass properties suggested that portions of the seafloor will experience average temperature increases in excess of 1°C, pH decreases greater than 0.3 units, dissolved oxygen decreases up to 3.7%, and a 40%-55% decrease in particulate organic matter flux to the seafloor by 2100 Sweetman et al., 2017). These projected changes may severely affect productivity, biodiversity, and distribution of deep-sea fauna, including species that underpin vulnerable marine ecosystems (VMEs) as well as commercially important deep-sea fishes, thereby compromising key ecosystem services (Johnson, Ferreira, & Kenchington, 2018;Jones et al., 2014;Levin & Le Bris, 2015;Pecl et al., 2017;Thurber et al., 2014).
Along with climate regulation, provisioning of food from fish stocks is one of the most critical ecosystem services provided by the deep sea (Thurber et al., 2014); these stocks are increasingly important to global food security (Victorero, Watling, Palomares, & Nouvian, 2018;Watson & Morato, 2013). However, warming and deoxygenation will simultaneously affect fishes by increasing metabolic rates and oxygen demand while limiting supply of oxygen to their tissues to meet increased demand for oxygen in low-oxygen environments (Holt & Jørgensen, 2015;Pörtner, Bock, & Mark, 2017;Pörtner & Knust, 2007). Decreasing food availability will indirectly exacerbate stress imposed by increased metabolism in warmer waters (Woodworth-Jefcoats, Polovina, & Drazen, 2017).
With few exceptions (Tittensor et al., 2010), the lack of reliable projections of future environmental conditions close to the seabed has constrained efforts to project shifts in distributions of deep-sea bottom-dwelling species.
The recent modelling of global-scale scenarios for future deep ocean environmental conditions (e.g. Sweetman et al., 2017), in tandem with increased understanding of the ecology and distribution of key deep-sea benthic species (e.g. Orejas & Jiménez, 2019;Priede, 2017;Rossi, Bramanti, Gori, & Orejas, 2017), enabled projections of distributional changes in deep-sea species (FAO, 2019). Utilizing the best available curated species occurrence data obtained from multiple public and restricted sources, along with a set of static (depth, slope, among others) and near-bottom dynamic environmental parameters (particulate organic carbon flux to the seabed, near seafloor pH, dissolved oxygen concentration and temperature, and near seafloor aragonite and calcite saturation state), we modelled habitat suitability for six cold-water coral and six deep-sea fish species under current conditions and projected changes under future projected high emission climate conditions for the whole North Atlantic Ocean. With this study, we asked how much suitable habitat we expect will be lost, gained or sustained as refugia areas under the business-as-usual emissions trajectory RCP8.5for indicators of VMEs and commercially important deep-sea fishes at an ocean basin scale, to support climate change adaptive management.

| Study area
Habitat suitability models of VME indicator taxa and commercially important deep-sea fish species were developed for the deep waters of the North Atlantic, from 18°N to 76°N and 36°E to 98°W. This region encompasses one of the best-studied deep-water regions in the world with respect to species distribution, environmental conditions and deep-sea species responses to environmental variability.
Additionally, the North Atlantic Ocean contains two well-established Regional Fisheries Management Organisations, increasing the relevance of these analyses for fisheries management, for conserving and protecting VMEs, and for the designation of OECMs. This basinscale focus also enhances model performance because it accounts for a wide range of environmental variability and species' ecological niches.

| Species selection and presence data
Six VME indicator taxa and six commercially important deep-sea fish species representative of both Eastern and Western North Atlantic deep-sea habitats were selected based on their wide spatial distribution, ecological significance or catch relevance in deep-sea fisheries, and on the availability and spatial coverage of existing occurrence records ( Table 1). The VME indicator taxa included three scleractinian corals that form aragonite skeletons (Lophelia pertusa, 1 Madrepora oculata and Desmophyllum dianthus) and three octocorals forming calcitic axial skeletons (Acanella arbuscula), and with sclerites in their axis or coenenchyme and polyps (Acanthogorgia armata, and Paragorgia arborea). Despite the widespread occurrence of these two groups of VME indicators in the North Atlantic (FAO, 2019), they are expected to respond differently to future water mass conditions properties. The six deep-sea fish species selected were the commercially harvested round nose grenadier (Coryphaenoides rupestris), Atlantic cod (Gadus morhua), blackbelly rosefish (Helicolenus dactylopterus), American plaice (Hippoglossoides platessoides), Greenland halibut (Reinhardtius hippoglossoides) and beaked redfish (Sebastes mentella). Although many of these fishes are not strictly considered deep-sea species, they all occur beyond 200 m depth (Table 1) and are relevant to deep-sea fisheries including in areas beyond national jurisdictions.
Georeferenced presence-only records were obtained from institutional databases of partners participating in this work (see Supporting Information Appendix S1, Table S1) as well as from public databases (Table 1)

such as the Ocean Biogeographic Information
System portal 2 (OBIS), the NOAA Deep Sea Coral Data Portal, 3 and the ICES Vulnerable Marine Ecosystems data portal. 4 In order to reduce potential errors in the spatial position of the occurrence records, we compared the depth values given by OBIS and NOAA with depth values extracted from the depth raster layer, excluding any occurrence records with no depth information or with depths that differed more than 30% and more than 50 m in absolute depth.
In the case of the ICES VMEs database, those records with a position accuracy lower than 5000 m of linear distance were excluded. Species occurrence provided directly by co-authors was crosschecked for accuracy of reported depth prior to submission of the data and were considered accurate, and took priority over OBIS data from the same institutional sources. Maps of the presence records used in the models are provided as supporting information ( Figure S1).
Most HSM approaches require information on the location of both species presence and absence. However, existing biological datasets rarely include information on species absence, despite its importance for model performance and precision (Iturbide, Bedia, & Gutiérrez, 2018;Iturbide et al., 2015;Wisz & Guisan, 2009 To overcome this obstacle, we generated pseudo-absence data (a.k.a. background points) by adapting the methodology in Iturbide et al. (2015) to our specific data. In the first step, we used the function OCSVMprofiling from the R package MOPA (Iturbide et al., 2018) to limit the geographic region for pseudo-absence data generation using environmental profiling based on presence data. In the second step, pseudo-absence data were randomly generated in the region defined above using the function pseudoAbsences from the MOPA package, but excluding a buffer distance to presence records of 6 km. The number of pseudo-absence data points generated differed between cold-water corals (10,000) and deepsea fishes (100,000) because of the different numbers of presence records. Finally, the pseudo-absence data were randomly stratified subsampled by depth strata to match the proportion of the presence records distribution by depth of all cold-water corals and fish species in public databases. Environmental variables of present-day and future conditions, including particulate organic carbon (POC) flux at 100 m depth (epc100, mg C m −2 day −1 ), bottom water dissolved oxygen concentration (µmol/kg), pH and potential temperature (°C) were downloaded from the Earth System Grid Federation (ESGF) Peer-to-Peer (P2P) enterprise system. 5 The epc100 was converted to export 5 https ://esgf-node.llnl.gov TA B L E 1 Number of grid cells with occurrence data obtained from multiple sources for the six cold-water corals and six deep-sea fishes used to model the suitable habitat in the North Atlantic. Depth ranges of occurrence records, and mean and standard deviation (±) for slope, Bathymetric Position Index (BPI), temperature at seafloor (Temp), POC flux to seafloor, and Aragonite (Ωar) and Calcite (Ωcal) saturation state at seafloor are also shown. Depth ranges for scleractinian corals and octocorals are shown for reference purposes only since depth was not considered for these species  Collinearity between all candidate predictor variables was evaluated using Spearman's coefficient of correlation and the variation inflation factor (VIF, Zuur, Ieno, & Elphick, 2010

| Modelling approach
We used an ensemble modelling approach to predict habitat suitability under present-day    Hastie & Tibshirani, 1986) and the random forest machine learning algorithms (RF, Breiman, 2001). Maxent models were developed using the function maxent from the R package Dismo (Hijmans, Phillips, Leathwick, & Elith, 2017), with prevalence set as the proportion of presences over the pseudo-absences generated. GAMs were fitted with the function gam from the R package mgcv (Wood, 2018) using a binomial error distribution with logit link function, constraining the smooth curves to four knots to avoid overfitting; we used three knots for temperature and aragonite in cold-water coral models. Finally, we computed RF models using the function randomForest from the R package of the same name (Liaw & Wiener, 2002).
Variable selection in GAMs used the Akaike information criterion and the function dredge from the R package MuMIn (Barton, 2018), whereas the other two models (Maxent and Random Forest) were fitted with the original set of variables. To assess the contribution of each variable to the final predictions, we used a randomization procedure adapted from Thuiller, Lafourcade, Engler, and Araújo (2009) dictors and predicted habitat suitability was analysed using response curves, described by Elith et al. (2006).
Performance for the present-day model was evaluated using a cross-validation method based on a random 'block' selection of training and testing data (Guinotte & Davies, 2014). Past work suggests this method of partitioning a unique dataset provides the best spatial independence between training and testing datasets (Fourcade et al., 2018 Fielding and Bell (1997) provide a complete description of these statistics.
Each model was then used to predict a relative index of habitat suitability (HSI) across the study area under present-day  conditions and to project the HSI for the period 2081-2100 by projecting the present-day niche onto the environmental layers of projected future conditions. The modelled logistic outputs consisted of an HSI that ranked grid cells according to their predicted suitability for a particular species, rather than the probability of presence (Elith et al., 2011;Greathead et al., 2015

| RE SULTS
In general, the three families of modelling approaches (Table S2) and the ensemble model predictions ( Ensemble models for all cold-water coral species also achieved good accuracy, although models for scleractinian species per- The habitat suitability models developed here included seven different predictors which contributed differently to the different modelled species (Table 3; Table S4). In general, POC flux to the seafloor, bottom water temperature, and aragonite or calcite saturation were the most important predictors for scleractinians and octocorals, in contrast to depth, POC flux, and temperature for deep-sea fishes (Table 3). We also note, however, the impor-    (Table 4). However, refugia for L. pertusa estimated with the F I G U R E 1 Habitat suitability index predicted under present-day  and projected under future (2081-2100; RCP8.5 or business-as-usual scenario) environmental conditions for cold-water corals in the North Atlantic Ocean using an ensemble modelling approach F I G U R E 2 Habitat suitability index predicted under presentday  and projected under future (2081-2100; RCP8.5 or business-as-usual scenario) environmental conditions for commercially important deep-sea fishes in the North Atlantic Ocean, using an ensemble modelling approach 10th percentile threshold was only about 1.5% of the North Atlantic present-day habitat (Table 4) (Table 4; Figure 6).
The ensemble model projected a 30%-50% reduction of suitable habitat for the fish species G. morhua and Hipoglossoides platessoides, between 10% and 15% for R. hippoglossoides, and between 2% and 25% for C. rupestris, mostly in their lower latitudinal limit (Figure 3;

F I G U R E 3
Climate-induced projected changes (RCP8.5 or business-as-usual scenario) in the suitable habitat for cold water corals and deep-sea fishes in the North Atlantic Ocean with an ensemble modelling approach. The extension of the habitat was calculated from binary maps built with two thresholds: 10 percentile training presence logistic (10th percentile) and maximum sensitivity and specificity (MSS)

F I G U R E 4
Climate-induced projected changes (RCP8.5 or business-as-usual scenario) in the latitudinal distribution of cold water corals and deep-sea fishes in the North Atlantic Ocean with an ensemble modelling approach. The extension of the habitat was calculated from binary maps built with two thresholds: 10 percentile training presence logistic (10th percentile) and maximum sensitivity and specificity (MSS). The black line indicates the median latitudinal F I G U R E 5 Climate-induced projected changes (RCP8.5 or business-as-usual scenario) in the depth distribution of cold water corals and deep-sea fishes in the North Atlantic Ocean with an ensemble modelling approach. The extension of the habitat was calculated from binary maps built with two thresholds: 10 percentile training presence logistic (10th percentile) and maximum sensitivity and specificity (MSS). The black line indicates the median depth Figure S9a,b). The decrease in suitable habitat for G. morhua included important shallower water fishing grounds, such as Georges Bank, the Irish Sea, the Norwegian Sea and the southern North Sea ( Figure   S9a,b). Of those species examined, only H. dactylopterus and S. mentella increased in projected total suitable habitat by 2100 (by about 20%-30%), mostly by expanding their northern latitudinal limit (Figure 3; Figure S9a,b). Therefore, we observed a clear northern shift in the median latitude of suitable habitat for most fishes by 2100 (from 2.0° to 9.9°; Figure 4), but with no clear trend in depth distribution ( Figure 5).
Projected climate refugia for deep-water fishes were comparatively large compared to corals, averaging between 51% and 63% of present-day habitat, depending on the threshold used (Table 4). These projected climate refugia occur mostly on both sides of the northern part of the North Atlantic (Figure 7).
Our results strongly suggest that warming, acidification, and decreasing food availability ( Figure S3) will act additively or synergistically to alter the availability of suitable habitat for deep-sea species (Figures 6 and 7

| D ISCUSS I ON
Our model predictions and projections showed that North Atlantic deep-sea species with the best-studied distributions could experience a significant reduction in suitable habitat by 2100 as a result of climate change. Indeed, our results suggest that the suitable habitat of scleractinian corals in the North Atlantic may be reduced by more than 50%, potentially extirpating all three octocorals studied (A. arbuscula, A. armata and P. arborea). This reduction could be of particular concern for A. armata, a species limited in distribution to the North Atlantic Ocean. Our projection also suggested a northward shift of suitable habitat for many commercially important deep-sea fishes, a finding consistent with the hypothesis of a poleward expansion in response to climate change (Jones et al., 2013;Perry et al., 2005;Poloczanska et al., 2013). Our study projected very limited, discrete climate-change refugia for cold-water corals, and especially octocorals, highlighting the need for accurate fine-scale climate grids and methodologies to properly identify climate refugia (Ashcroft, 2010;Kavousi & Keppel, 2018;Valdimarsson, Astthorsson, & Palsson, 2012).
Occupancy of the future suitable habitats will depend on connectivity pathways and will differ greatly between deep-sea fishes which have juvenile and adult mobility and the cold-water corals which can only disperse as larvae (Andrello, Mouillot, Somot, Thuiller, & Manel, 2015;Baco et al., 2016;Hilário et al., 2015). The latter will be much more dependent on downstream connectivity paths that have not been evaluated in our projections but which are likely to further impose F I G U R E 6 Projected present-day suitable habitat loss, gain, and acting as climate refugia areas (sensu ) under future (2081-2100; RCP8.5 or business-as-usual scenario) environmental conditions for cold-water corals in the North Atlantic Ocean. Areas were identified from binary maps built with an ensemble modelling approach and two thresholds: 10 percentile training presence logistic threshold (10th) and maximum sensitivity and specificity (MSS) fragmentation of populations through loss of source populations (Fox, Henry, Corne, and Roberts (2016). Fox et al. (2016) have shown that past changes in the NAO significantly altered network connectivity and source-sink dynamics for L. pertusa in the northeast Atlantic, and there is every reason to anticipate similar impacts associated with the future climate scenarios shown here.
Previous studies suggested some of the general changes in distribution patterns identified here for both cold-water corals and deep-sea fishes, based on inferences about changes in distribution resulting from the likely effects of climate change on the marine environment. Specifically, several studies projected significant loss of suitable habitat for cold-water coral reefs globally (Guinotte et al., 2006;Tittensor et al., 2010;Zheng & Cao, 2014)  and reduced abundance of Greenland halibut in the Barents Sea (Fossheim et al., 2015). Such northward shifting of ranges is already being documented for some western North Atlantic shelf and slope fishes (Møller et al., 2010;Nye et al., 2009), and although Ross, Rhode, Viada, and Mather (2016) also reported northward range extensions, they cautioned that lack of historical baseline surveys limits interpretation of distributional data. Our study presents another example of the potential value of environmental niche modelling for projecting changes in suitable habitat for deep-sea species under future climate scenarios, and extends the more localized or species-specific studies already reported above, to much larger spatial scales and more consistent and rigorous analytical methods.
With climate change potentially affecting all species in the ecosystem its influence on community assembly processes remains a major knowledge gap. Field surveys suggest an association between some deep-sea fishes (e.g. H. dactylopterus) and live cold-water corals reefs (e.g. L. pertusa; D' Onghia et al., 2012;Pham et al., 2015). Fish species use such habitats as both spawning and nursery grounds (Corbera et al., 2019), calling into question whether the projected range expansion of the blackbelly rosefish is likely to occur given projected declines of a species forming one of its prime habitats, for example, L. pertusa. However, insufficient evidence exists to infer that blackbelly rosefish can occupy transitional and noncoral habitats in many regions (Biber et al., 2014;Milligan, Spence, Roberts, & Bailey, 2016;Ross & Quattrini, 2007), suggesting that declines in scleractinian suitable habitat may not translate into loss of habitat for this deep-sea fish species.
Temperature and depth were important predictors of habitat suitability for cold-water coral and deep-sea fishes respectively. This result corresponds to similar studies on distributions of other deep-sea TA B L E 4 Proportion of the present-day suitable habitat acting as refugia areas  under future (2081-2100; RCP8.5 or business-as-usual scenario) environmental conditions for cold-water corals and commercially important deep-sea fishes in the North Atlantic Ocean. Refugia areas were identified from binary maps built with an ensemble modelling approach and two thresholds: 10 percentile training presence logistic threshold (10th) and maximum sensitivity and specificity (MSS  & Guinotte, 2011;Georgian, Shedd, & Cordes, 2014;Guinotte & Davies, 2014;Lauria et al., 2017;Tittensor et al., 2010) and deep-sea fishes (Gomez et al., 2015;Parra et al., 2017;Ross & Quattrini, 2007).
However, the strong autocorrelation between depth and temperature and significant correlation with other environmental and biological factors complicates efforts to elucidate the environmental parameters primarily responsible for the observed patterns. Slope and other terrain attributes also help shape distributions of some cold-water coral species (Rengstorf, Yesson, Brown, & Grehan, 2013) and are linked to the higher suitability of high geomorphological relief habitats that promote stronger near-bed currents and enhanced food supply (Genin, Dayton, Lonsdale, & Spiess, 1986;Hebbeln, Van Rooij, & Wienberg, 2016;Soetaert Mohn, Rengstorf, Grehan, & Van Oevelen, 2016). Habitat slope and rugosity are also important elements influencing distributions of some deep-sea fishes (Quattrini, Ross, Carlson, & Nizinski, 2012;Ross & Quattrini, 2009;Ross, Rhode, & Quattrini, 2015). For those fishes intimately tied to complex habitat, loss of corals may tend to disperse (as they search for remaining habitat) or concentrate (as they utilize shrinking habitats) fish communities.
Multiple studies document the importance of aragonite and calcite saturation state in determining cold-water coral habitat suitability (Davies & Guinotte, 2011;Thresher et al., 2015;Tittensor et al., 2010;Yesson et al., 2012), because waters supersaturated in carbonate enable coral skeleton bio-calcification. The chemical dissolution and biological erosion of the unprotected skeleton exposed to corrosive waters will impair the long-term survival of cold-water coral reefs (Hennige et al., 2015;Schönberg, Fang, Carreiro-Silva, Tribollet, & Wisshak, 2017;Thresher, Tilbrook, Fallon, Wilson, & Adkins, 2011). However, cold-water corals may occur in undersaturated waters of high productivity, leading to the hypothesis that increased food supply may compensate to some degree for undersaturation by providing the additional energy necessary to survive (Baco et al., 2017;Ross, unpublished data;Thresher et al., 2011).
Elevated food supply may also compensate for low dissolved oxygen concentrations (Hanz et al., 2019). However, in an environment with consistently scarce food or low oxygen concentration, the metabolic costs of calcifying in extremely low carbonate conditions may become prohibitively expensive, thus compromising coral survival (Carreiro-Silva et al., 2014;Hennige et al., 2015;Maier et al., 2016).
In fact, food availability measured as POC flux to the seafloor was also an important predictor of suitable habitat for most cold-water corals and deep-sea fishes in our study. This finding corroborates reports of abundant L. pertusa in regions of elevated POC flux, both in recent times (Davies & Guinotte, 2011;White, Mohn, Stigter, & Mottram, 2005) and since the last glacial events (Boavida et al., 2019;Henry et al., 2014;Matos et al., 2015;Wienberg et al., 2010). Indeed, multiple studies link reduced food availability to reduced physiological performance (e.g. calcification and respiratory metabolism) and condition of cold-water corals (Büscher et al., 2017;Larsson, Lundälv, & van Oevelen, 2013;Naumann, Orejas, Wild, & Ferrier-Pagès, 2011), as well as their ability to cope with ocean change (Büscher et al., 2017;Georgian et al., 2016;Gomez et al., 2019;Maier et al., 2016;Wood, Spicer, & Widdicombe, 2008). In contrast, the direct link between POC flux and deep-sea fish abundances has proven difficult to demonstrate (Bailey, Ruhl, & Smith, 2006), despite some evidence that increased surface production may fuel key fish prey taxa such as benthic invertebrates (Bailey et al., 2006;Drazen, Bailey, Ruhl, & Smith, 2012;Ruhl & Smith, 2004). Therefore, the projected decrease in food availability by 2100 in the North Atlantic Gomez et al., 2019;Sweetman et al., 2017) may exacerbate the likely negative effects of other environmental changes.
Inferring the capacity of deep-sea species, and corals in particular, to adapt to changes in water chemistry projected by climatic models is challenging. For example, experimental (Keller & Os'kina, 2008) and palaeoecological studies (Wienberg et al., 2010) suggest that some M. oculata populations can tolerate elevated seawater F I G U R E 7 Projected present-day suitable habitat loss, gain, and acting as climate refugia areas (sensu ) under future (2081-2100; RCP8.5 or business-as-usual scenario) environmental conditions for commercially important deep-sea fishes in the North Atlantic Ocean. Areas were identified from binary maps built with an ensemble modelling approach and two thresholds: 10 percentile training presence logistic threshold (10th) and maximum sensitivity and specificity (MSS) temperature; this tolerance may explain their prevalence at shallower depths (180-360 m) in the Mediterranean Sea (Chimienti, Bo, & Mastrototaro, 2018;Chimienti et al., 2019;Freiwald et al., 2009;Gori et al., 2013). L. pertusa also may occur in regions (e.g. beneath the Florida Current, Gulf Stream) that experience periodic high temperatures (12-15°C) and rapid water property fluctuations (Brooke, Ross, Bane, Seim, & Young, 2013), but the impact of these conditions is unclear. However, D. dianthus, which may also tolerate high temperatures (Naumann, Orejas, & Ferrier-Pagès, 2013) and survive in waters undersaturated in aragonite (Jantzen et al., 2013;Rodolfo-Metalpa et al., 2015;Thresher et al., 2011), may experience reduced metabolism that compromise survival when exposed to the combined effects of increased temperature and reduced aragonite saturation . Additionally, although other scleractinians and octocorals may calcify and grow under low or undersaturated conditions (Büscher et al., 2017;Form & Riebesell, 2012;Hennige et al., 2014Hennige et al., , 2015Maier et al., 2012Maier et al., , 2013Movilla et al., 2014;Thresher et al., 2011), their capacity to sustain calcification and other physiological processes under unfavourable conditions remains unclear, given studies that show effects of low carbonate concentrations on coral metabolism (Hennige et al., 2014) and increased energy demand required to maintain pH homeostasis at calcification sites (McCulloch et al., 2012;Raybaud et al., 2017). Despite the many uncertainties regarding potential acclimation and adaptation of cold-water coral species to changes in climate, along with interspecific genetic variability (Kurman et al., 2017) and potential for local adaptation (Georgian et al., 2016), growing evidence points to limited long-term capacity for adaptation to multiple stressors associated with climate change (Kurman et al., 2017).
Habitat suitability modelling approaches come with some caveats, and we, therefore, acknowledge multiple common and well-known limitations that may be particularly pronounced when modelling deep-sea taxa. For example, cold-water coral and fish distributions respond to small-scale variation in terrain, such as substrate type and seabed rugosity, as well as local oceanographic conditions such as food availability (Bennecke & Metaxas, 2017;De Clippele et al., 2017;Drazen et al., 2012;Rengstorf et al., 2013;Ross et al., 2015;White et al., 2005). We also recognize some limitations from the quantity, quality and spatial coverage of occurrence data, availability of absence records as well as some uncertainty in deepsea species identification (mostly for cold-water corals). For example, two new species of Acanella were recently described from the Gulf of Mexico and Norfolk Canyon off the coast of eastern United States (Saucier, Sajjadi, & France, 2017). We, therefore, cannot state whether previous records from several databases include some of these new species. The deep sea remains one of the least studied and sampled areas on the planet, with many undescribed species and unresolved taxonomy that constrain determination of the full spatial distributions of many species. Extensive exploration of the deepsea environment may eventually reduce this uncertainty, but it will take time. Further integration of species-level biogeochemical and physical data, as well as results of the ecophysiological performance of deep-sea organisms from ex situ experimental work, will improve suitability and distribution mapping, but noting the need for additional mechanistic (experimentally derived) understanding of how climate drivers elicit ecological responses.
Finally, future climate scenarios always hinge on assumptions; indeed, the RCP8.5 or business-as-usual scenario projections for 2081-2100 assume specific future greenhouse gas emissions, world population growth and technology development Van Vuuren et al., 2011), and also encompass climate model errors shared by all scenarios. Consequently, our projected future models for cold-water coral and fish species potentially represent worstcase scenarios, with a high degree of uncertainty. Furthermore, future climate projections may not capture localized effects that may influence benthic organisms. Nevertheless, our projections of distribution changes in key species across the North Atlantic offer critical, best available information for decision makers to develop long-term sustainable management plans (Rheuban, Doney, Cooley, & Hart, 2018), highlighting the utility of enhanced international dialogue on basin-scale management.
Following sufficient ground-truthing, habitat suitability models can become valuable tools to inform environmental management and conservation policy (Robinson, Nelson, Costello, Sutherland, & Lundquist, 2017)  Ocean Assessment) can help set such science agendas.
In summary, we have shown that despite all the caveats, habitat suitability models can produce potentially useful projections of future changes in the distribution of deep-water fish and invertebrate species, and areas where foundation species could be impacted by climate change and may be used to inform management decisions.
This application is especially relevant for dramatic changes such as those projected here. Although ocean-basin scale models provide useful coarse, directional information regarding climate change impacts on deep-sea fauna, regional models could help to resolve changes of distribution and identification of refugia for monitoring purposes. We hope our study offers a suitable template for, and will stimulate, similar analyses on other taxa or regions.

ACK N OWLED G EM ENTS
This work contributes to the European Union's Horizon 2020 re-

DATA AVA I L A B I L I T Y S TAT E M E N T
The bathymetry data supporting the analyses are publicly avail-