Introduction

Cold-water coral (CWC) reefs are important marine ecosystems, as they are hotspots of biomass and biodiversity that provide a wide range of ecosystem processes and services (Armstrong et al. 2012) through carbon (C) cycling, C burial, fisheries and new opportunities for pharmaceutical development (Foley et al. 2010; Skropeta 2008; van Oevelen et al. 2009). Since CWC ecosystems occur in the aphotic zone, they depend on the supply of particulate organic matter (POM) through oceanographic processes such as tidal downwelling and internal waves (Roberts et al. 2006). Climate change models predict global increases in ocean temperatures, as well as associated increases in ocean acidification and decreases in ocean oxygen levels (IPCC 2018). These changes might affect ocean currents, organisms’ metabolic demands and the export flux of POM, ultimately affecting the amount of biomass deep-sea ecosystems can support and consequently affecting ecosystem processes such as C cycling (Sweetman et al. 2017). Quantifications of the current status of ecosystems in terms of biomass and C turnover are thus needed to better comprehend future changes in ecosystem functions.

Calculating the amount of biomass in a benthic reef ecosystem is important in a number of ways. Firstly, by taking respiration or oxygen consumption measurements, biomass can be used as a proxy for calculating C turnover and thus ecosystem functioning (van Oevelen et al. 2009). From C turnover, one can quantify the importance of the different supply mechanisms of POM from surface water primary productivity (PP) to the reef, ranging from natural deposition to flow-driven upwelling and advection. Finally, biomass values can also be used to calculate C standing stock, which could feed into assessments of C budgets and help evaluate the importance of CWC reefs as carbonate sinks or sources (Burrows et al. 2014; Titschack et al. 2015).

Existing marine biomass assessments can be based on allometric (unequal growth rate) or isometric (equal growth rate) relationships (Dornelas et al. 2017; Benoist et al. 2019; Madin et al. 2020). Compared to tropical corals, cold-water corals have low metabolic rates and allometric data are currently not available. Biomass data can also be measured with small-scale localised samples collected with box cores, Van Veen grabs or from fishing activities (Bohnsack and Harper 1988; de Froe et al. 2019; Kazanidis and Witte 2016; van Oevelen et al. 2009). These single spot measurements, however, do not reflect the spatial heterogeneity of natural ecosystems (de Froe et al. 2019). While terrestrial studies can exploit remote sensing (e.g. satellite imagery) in combination with field measurements to map biomass at the ecosystem scale (Baccini et al. 2008), this approach is not applicable to deep-sea habitats. Despite increased availability of continuous photographic seafloor maps (Bodenmann and Thornton 2017; Thornton et al. 2016) wide-scale applications remain scarce, with no maps available for CWC reefs biomass to date.

The reef-forming coral Lophelia pertusa and the massive sponge Spongosorites coralliophaga are two important ecosystem engineers (sensu Jones et al. 1997) as they are both key to sustaining healthy CWC reef ecosystems (Roberts 2005; Wheeler et al. 2007; Kazanidis et al. 2016, 2018a). The three-dimensional complex framework of live L. pertusa alters the chemistry and currents of its surrounding environment, but it is mostly the dead coral rubble and framework that provide microhabitats for a diverse fauna (Henry and Roberts 2007; Buhl-Mortensen et al. 2010). Similarly, the sponge S. coralliopaga is an ecosystem engineer that hosts a diverse and rich epifaunal community (Kazanidis et al. 2016).

A novel approach is proposed here to map whole-reef biomass of L. pertusa and S. coralliophaga using surface area data from high-definition (HD) videos in combination with field measurements and predictive modelling techniques based on the CWC Mingulay Reef (MR) Complex, one of the best studied reefs in the world (Davies et al. 2009; De Clippele et al. 2017; Douarin et al. 2014; Duineveld et al. 2012; Findlay et al. 2014; Henry et al. 2010; Henry and Roberts 2014; Kazanidis and Witte 2016; Moreno-Navas et al. 2014; Roberts et al. 2009, 2005). Carbon stock and turnover were then assessed from the predicted biomass values using oxygen consumption and respiration rates available through the literature.

Methodology

Location

Our study area, known as Mingulay Reef area 01 (MR1), lies on the east–west oriented ridge of the MR Complex off the west coast of Scotland (Fig. 1). The MR1 compromises of over 500 small reef mounds, referred to as “minimounds” (De Clippele et al. 2017). These “minimounds” are on average 8 ± 5 m (SD) high, 44 ± 6 m (SD) long and 30 ± 11 (SD) m wide (De Clippele et al. 2017, 2019a). The shallowest point on the ridge rises to less than 100 m water depth, and the deepest point reaches 260 m. Live coral can be found between 120 and 190 m depth (Roberts et al. 2009; De Clippele et al. 2017). The MR Complex is a hotspot of marine biodiversity which was designated as a marine protected area in 2010 (Marine Scotland Science 2016). The majority of the MR is covered with coral rubble (Roberts et al. 2005) which hosts a rich associated fauna of anthozoans, ophiuroids and sponges (Henry and Roberts 2007, Kazanidis et al. 2016). Even though the MR complex could, in the future, be affected by ocean warming, due to its relatively shallow location, there is no immediate evidence of ocean acidification at the MR Complex (Findlay et al. 2013, 2014; Thresher et al. 2015; Morato et al. 2020) and it might therefore act as a refuge (De Clippele et al. 2017). There are two hydrographic mechanisms that supply the MR Complex with nutrients. Firstly, hydraulic flow and internal waves cause rapid tidal downwelling (Davies et al. 2009) supplying warmer and nutrient-rich water from the surface. Secondly, POM is also supplied via re-suspended particles during peak tidal flows (Davies et al. 2009).

Fig. 1
figure 1

a Location of the Mingulay Reef Complex in the Hebridean sea off the west coast of Scotland. b Mingulay reef 01 with the red square showing the location of the high-resolution map (c). c Middle section of the Mingulay reef with a resolution of 0.35 × 0.35 m where the minimounds are visible (De Clippele et al. 2017). The black star indicates the location of the aquatic eddy co-variance that was deployed in 2012 during the Changing Oceans 2012 (JC073) expedition

Full size image

Data

Biological data

HD video data were collected during the Changing Oceans 2012 expedition, RRS James Cook cruise 073 (BODC 2015), using the remotely operated vehicle (ROV) Holland-1 (more details in De Clippele et al. 2017). Using the software VLC v.3.0.7, video frames were extracted whenever the coral L. pertusa or the sponge S. coralliophaga was present (Table 1, Fig. 2).

Table 1 ROV video transect information
Full size table
Fig. 2
figure 2

Graphic overview of the novel methodological approach used in this study

Full size image

Spongosorites coralliophaga and L. pertusa specimens were collected during the 2012 expedition. Spongosorites coralliophaga specimens were initially collected for sponge feeding ecology experiments (Kazanidis and Witte 2016) for which their wet and dry weight and respiration measurements were collected and used in this study (Table 2) (Kazanidis et al. 2018a,b).

Table 2 Information on location and time of sample collection together with dry weight biomass, surface area (SA) from the extracted video frames and the calculated SA to dry weight conversion value of S. coralliophaga. For details on the ratio calculation, see “Biomass estimation” section.
Full size table

Lophelia pertusa fragments were collected from 141 to 167 m depth (Hennige et al. 2014, 2015) and used for skeletal + tissue dry weight measurements (Fig. 2). The average skeletal + tissue dry weight of a single coral polyp was found to be 0.78 ± 0.55 g (SD), based on a sample size of 192 polyp branches.

Environmental data

Bathymetric data were gathered using a Simrad EM2000 ship-mounted multibeam echosounder (MBES), using the RV Lough Foyle, between 28 June and 5 July 2003 (cruise LF26, Roberts et al. 2004). Data were processed to a resolution of 2 × 2 m. The system has an angular coverage sector of 120°, 111 beams per ping and a 1.5° beam width across the track. A Seapath 200 GPS system provided real-time heading, altitude, position and velocity solutions (Roberts et al. 2004). Additional multibeam backscatter data were collected in 2012 (cruise JC073) using the shipboard Simrad EM710, with a resolution of 2.5 × 2.5 m. Physical habitat characteristics were derived from multibeam data including depth, slope, Bathymetric Positioning Index (BPI) (inner and outer radius of 8 and 24 cells respectively, see De Clippele et al. 2017), rugosity (3 × 3 cells) and backscatter using the ArcGIS 10.6.1 software.

Oxygen consumption and respiration data

In this study, the ex-situ measured O2 consumption of 7.2 ± 1.2 (SD) µmol O2 skeletal dry weight g−1 d−1 for live L. pertusa was used as it was collected from the Mingulay Reef Complex (Dodds et al. 2007). In a laboratory, the flow and feeding conditions can be different from the natural environment and can therefore vary (Table 3). However, the value provided by Dodds et al. (2007) was favoured as it is very similar to the in situ measured incubations by Khripounoff et al. (2014) from the submarine canyons of the Bay of Biscay, where the temperature range is similar to that of the MR Complex.

Table 3 Overview of L. pertusa oxygen consumption rates available in the literature from the North East Atlantic, normalised for skeletal + tissue dry weight in gram per day. Skeletal + tissue dry weight is reported here, as this unit is the most commonly chosen one to calculate O2 consumption of L. pertusa
Full size table

At the MR1, an in situ O2 consumption rate of the community associated with rubble was measured with the non-invasive aquatic eddy co-variance (AEC) technique and provided an average value of 27.8 O2 ± 2.3 (SE) mmol m−2 d−1 (Rovelli et al. 2015). The respiration rates of S. coralliophaga of 47.52 ± 16.21 (SD) µmol CO2 g−1 d−1 were derived from ex-situ oxygen consumption rate values from chamber incubations by Kazanidis et al. (2018a) (Fig. 2h).

Lophelia reef habitat

Since our study seeks to better understand the ecological functioning of the biogenic reef habitat L. pertusa provides through its coral framework and rubble, a Lophelia reef habitat map was produced at MR1. This map is based on habitat type data (Lophelia reef habitat vs non-Lophelia reef habitat) (i.e. mud, coarse and fine sediments) from videos from the JC073 and MINCH 2003 expeditions (Roberts et al. 2005) and the environmental variables detailed in the previous section. The Lophelia reef habitat was produced by creating a composite layer in ArcGIS, which consists of a single raster data set produced from multiple raster bands of environmental data. The rasters used were depth, slope, BPI (inner and outer radius of 8 and 24 cells, respectively, see De Clippele et al. 2017), rugosity (3 × 3 cells) and backscatter. The composite layer was then overlaid with data on habitat type. The “Training Sampler Manager” tool was then used to manually classify the different habitat types with the “polygon delineation” tool. The ArcGIS 10.6.1 Interactive Supervised Classification tool was subsequently used to produce a MR1 Lophelia reef habitat map. One-third of the habitat data set was excluded from the analyses and used to visually assess the accuracy of the habitat map.

Biomass estimation

Live biomass at MR1 was calculated from HD videos, using a combination of video surface area and coral and sponge dry weight measurements. Data on the skeletal + tissue dry weight of specimens (Hennige et al. 2014; Kazanidis et al. 2016; Kazanidis and Witte 2016) and on methods to estimate coral polyp density (Cathalot et al. 2015) were further compiled from laboratory measurements and the literature. First, the surface area of L. pertusa and S. coralliophaga was calculated for each extracted video frame. Based on these measurements, the biomass of live L. pertusa, the skeletal + tissue dry weight of dead L. pertusa framework and the biomass of S. coralliophaga were calculated. The steps are described in detail in Fig. 2 and in the Online Resource 1. Biomass is here defined as the live tissue of S. coralliophaga and L. pertusa. These data were then used to create a predictive biomass map for the Lophelia reef habitat area, using random forest modelling (see “step 4” section). Data extracted from the predictive map could then be used to calculate C stock and turnover (see “step 3 and 5” section). The following steps introduce uncertainties across the different measurements. To account for these uncertainties, the minimum, mean and maximum biomass were calculated and used in each of the five steps (see the Results section and Online Resource 2). However, due to the novelty of the approach utilised in this study, more emphasis will be given to the discussion of the obtained mean estimates.

Step 1

After extracting the video frames from the HD videos (Fig. 2a), the scaled surface area of L. pertusa and S. coralliophaga was calculated using the software Image J2 (Rueden et al. 2017). Next, by using a Wacom graphics pen and tablet, the coral or sponge contours were delineated in 179 and 537 images, respectively, and by using the Image J2 “measure” tool, the organism’s surface area was calculated (Fig. 2b). From twenty extracted video frames, following the approach of Cathalot et al. (2015), an average of 3102 ± 1614 (SD) visible surface polyps per square meter was counted. Following Cathalot et al. (2015), the “visible” polyp density was then multiplied by a factor of 5 to estimate the true areal density of living L. pertusa polyps (Fig. 2c).

Step 2

Dry weight measurements of specimens were used to convert surface area to biomass. First, collected sponge dry weight measurements were related to the surface area of the same specimens seen in the video (Table 2). An average ratio (dry weight of collected sponge: video surface area of collected sponge) of 6595 ± 2917 (SD) (Table 2) was used to convert other sponge surface areas measurements from the videos to dry weight. Dry weight measurements can be used to calculate C turnover rates (see Biological data section). To convert to biomass, the average ratio of 5.234 ± 1.4 (SD) was calculated and used to convert sponge dry weight to wet weight biomass (Table 2). To calculate the live biomass of L. pertusa, first, the surface area was converted to mass using the average skeletal + tissue dry weight of a single coral polyp. Then, the tissue wet weight (TWW) of L. pertusa was calculated from the skeletal dry weight using the linear relationship between tissue dry weight and tissue and skeletal weight with the following equation derived from Hennige et al. (2014): TDW = 0.0415 (TWW + SDW) + 0.0849. With TDW = tissue dry weight (g) = SDW* 5%; TWW = tissue wet weight (g); SDW = skeletal dry weight (g) (Fig. 2d, e, Online Resource 1, Table 1).

Carbon stock

Step 3

The carbon stock of S. coralliophaga, live and dead framework was calculated (Fig. 2C). However, due to low visibility, the surface area of the dead coral framework of the coral colonies was challenging to delineate compared to the white and orange live branches. By following Vad et al. (2017), it was possible to estimate the dead skeletal mass from live skeletal mass assuming that at least 73% of L. pertusa colonies are composed of dead coral framework. The value 73% is the minimum proportion of dead skeleton in the L. pertusa colonies and was chosen here as advised by Vad et al. 2017. The average proportion is 0.82% ± 0.05 (SD). Windholz et al. 1983 found a ratio of 0.12 between the mass of CaCO3 and mass of inorganic C. The skeletal mass of L. pertusa framework (Fig. 2c–e) equalling CaCO3 mass, the amount of inorganic C stores was therefore calculated by multiplying the CaCO3 mass by 0.12. For non-calcifying organisms (e.g. sponges, anthozoans, tunicates), it is estimated that the average C content amounts to 18% ± 3.47 (SD) (Kazanidis and Witte 2016; Kazanidis et al. 2018a, b). This value was used to calculate C stock of S. coralliophaga from its dry weight measurements (Kazanidis and Witte 2016; Kazanidis et al. 2018a, b a,b).

Predictive mapping

Step 4

Based on video frame data points (Fig. 2e) and environmental variables (i.e. depth, BPI, slope, backscatter, rugosity) (Fig. 2f), a predictive biomass map was created using a random forest approach (Fig. 2g) with the random forest package in R (Breiman 2001). Correlated environmental variables (> 0.5) were removed prior to analyses. The random forest algorithm fits multiple decision trees to a training data set using randomly selected subsets of the predictor variables. Here, the training data set contained one-third of the total data points. To improve the prediction, randomly selected coral and sponge biomass absence data points were added to the data set from the JC073 and MINCH 2003 expeditions (De Clippele et al. 2017).

The importance of the environmental variables was assessed by plotting the mean decrease in accuracy for each variable to indicate their contribution to the model’s performance. Based on Rowden et al. (2017), we used a bootstrap technique to produce estimates of model uncertainty. In a process repeated 100 times, a random sample of the data was drawn with replacement each time and a model equal to the original constructed. Predictions were then made, resulting in 100 estimates from which the coefficient of variation (CV) was calculated to examine the output stability, giving a range the random forest model output might vary. This is measured as the standard deviation/mean × 100 (Wei et al. 2010).

Carbon turnover

Step 5

From the predictive Lophelia reef biomass map (see step 4 section), the total amount of biomass was extracted. This is used to calculate the yearly C turnover from O2 consumption and respiration measurements, assuming a 1:1 C to O2 ratio (Glud 2008). Shifts in the O2/CO2 ratio over time, as well as spatially, have been widely documented in tropical coral reef habitats and other benthic marine ecosystems (Therkildsen and Lomstein 1993; Glud 2008; Takeshita et al. 2018; Bolden et al. 2019). Although the discussion of such dynamics for deep-sea CWC reefs settings goes behind the remit of this study, it should be noted that shifts in the O2/CO2 ratio would proportionally offset our conservative estimates of the yearly C turnover rate by the reef. Carbon turnover is here defined as the conversion of ingested food into biomass and loss by respiration (Fig. 2i). Since the community associated with the coral rubble is similar to the community associated to dead exposed coral framework (Kazanidis et al. 2016), it was assumed that the respiration rate of the community associated with coral rubble and the dead coral framework is the same at the MR1. Gross activity such as biomass build up followed by subsequent degradation was not accounted for this study. However, in deep-sea settings net rates will be very close to the gross rates (Glud 2008). From an ecosystem-wide budget assessment, it is the net rates that are relevant for evaluating the overall importance of cold-water coral reefs in oceanic C budgets and were therefore used here.

Results

Predictive maps

The ArcGIS Interactive Supervised Classification procedure resulted in a Lophelia reef map covering an area 1.7 km2 (Fig. 3). From the results in “Stock and turn-over of carbon” section, it was calculated that live and dead L. pertusa framework cover 0.3 km2, S. coralliophaga covers 0.009 km2 resulting in L. pertusa derived rubble covering a total of 1.3 km2 of the Lophelia reef habitat.

Fig. 3
figure 3

Lophelia reef habitat predicted using the ArcGIS Interactive Supervised Classification

Full size image

The live L. pertusa biomass random forest model, based on the mean biomass input data, explained 16.8% of the variation in data. The random forest models based on the minimum and maximum biomass input data can be found in the Online Resource 2. The environmental variables that contributed most to explaining the spatial variability in the amount of live coral biomass were rugosity, depth and slope (Fig. 4). The L. pertusa predictive biomass map (Fig. 5) illustrated that the highest biomass is mostly located on the eastern ridge of MR1, with hotspots of biomass visible on the minimounds. The predictive maps indicate a total mean skeletal mass of 1046 T (range 137–2525 T skeletal mass) and live L. pertusa mean biomass of 209 T (range 26–515 T biomass). This was converted to a mean of 2828 T of dead exposed coral skeleton framework (range 370–6827 T skeletal mass) (see Methodology step 2, Table 4). The CV map provided an indication for the possible range where this random forest prediction might vary (Fig. 6). The CV map indicates a high uncertainty on the south-western part of the reef where video data are missing.

Fig. 4
figure 4

Mean decrease in accuracy plots of the mean Lophelia pertusa and Spongosorites coralliophaga random forest model indicating what the contribution of each variable is to the model performance. When the mean decrease accuracy value is higher for a certain variable, the removal of this variable from the model will decrease the model’s performance

Full size image
Fig. 5
figure 5

Modelled amount of the mean biomass of live Lophelia pertusa in the Lophelia reef habitat area of the Mingulay reef

Full size image
Table 4 Overview of the range Lophelia reef habitat predicted absolute minimum, mean and maximum dry, wet weight and carbon (C) stock masses, together with the mass of oxygen (O2) and C turned over
Full size table
Fig. 6
figure 6

The coefficient of variation computed as the standard deviation/mean × 100 for the random forest model of the mean biomass of L. pertusa in the Lophelia reef habitat area of the Mingulay reef

Full size image

The S. coralliophaga biomass random forest model explained 5.88% of the variation in biomass. The environmental variables that contributed most to explaining the spatial variability in the sponge biomass were rugosity and slope (Fig. 4). The S. coralliophaga predictive map (Fig. 7) showed the highest biomass on the eastern ridge of MR1. The CV map gives an indication for the possible range where this random forest prediction might vary (Fig. 8). The CV map indicates a high uncertainty on the north-eastern part of the reef where video data are missing.

Fig. 7
figure 7

Modelled amount of the mean biomass of Spongosorites coralliophaga in the Lophelia reef habitat area of the Mingulay reef

Full size image
Fig. 8
figure 8

The coefficient of variation computed as the standard deviation/mean × 100 for the random forest model of the mean biomass of S. coralliophaga in the Lophelia reef habitat area of the Mingulay reef

Full size image

The results of the minimum and maximum (range) of biomass of L. pertusa and S. coralliophaga random forest models can be found in Online Resource 2.

Stock and turnover of carbon

A total mean of 126 T C (range 16–303 T C) and 339 T C (range 44–819 T C) was calculated to be stored within the live and dead L. pertusa framework, respectively, for the total Lophelia reef habitat area of 1.7 km2. These values largely exceeded the mean C stock of 10 T (range 5–18 T C) of S. coralliophaga, reaching a total mean of 475 T C (range 65–1140 T C). The average live mass (skeletal weight + biomass) of a L. pertusa colony is 0.74 kg m−2 (range 0.09–1.66 kg m−2) with the average dead skeletal weight of L. pertusa colony being 1.6 kg m−2 (range 0.2–4 kg m−2). The average biomass of S. coralliophaga is 213 g m−2 (range 118–308 g m−2). The community associated (e.g. anthozoans, ophiuroids, sponges) with rubble was responsible for the largest turnover of organic material, reaching an annual rate of 163 T yr−1 (range 149–176 T C yr−1) (Table 4, Fig. 9). The community associated with dead L. pertusa framework and live L. pertusa was responsible for a mean of 44 (range 6–139 T C yr−1) and 32 T yr−1 (range 4–93 T C yr−1) of C mineralisation, respectively (Table 4, Fig. 9). The sponge S. coralliophaga only contributed 1 T C yr−1 (range 1–2 T C yr−1). This resulted in a total C turnover of the Lophelia reef area with a mean of 241 T C year−1 (range 160–400 T yr−1), corresponding to an O2 consumption of 32.37 mmol m−2 d−1 (range 21.48–53.72 mmol O2 m−2 d−1). If the Lophelia reef area was to consist entirely of soft-sediments, for which depth-based turnover rates of oxygen/carbon are available in the literature (see Glud 2008), than the area would just turnover 57 C T yr−1. An overview of the range of the Lophelia reef habitat predicted absolute minimum and maximum dry, wet weight and carbon (C) stock masses, together with the mass of oxygen (O2) and C turned over are given in Table 4.

Fig. 9
figure 9

Diagram representing the Mingulay cold-water coral reef, the mean amount L. pertusa rubble, live and dead framework and the yellow encrusting sponge S. coralliophaga contribute to the carbon turnover. The orange circles and stars represent the fauna (including dense Parazoanthus sp. and ophiuroids) and microbes associated with the rubble and dead coral framework. The dark green circles represent the organic matter that is transported from the surface water to the reef. The thickness of the upward facing arrows indicates the relative importance of the contribution of the coral habitats and the sponge mean annual carbon turnover. The diagram also shows an estimate of the amount of primary production above the reef and the contribution of two supply pathways to the reef, natural deposition and tidal downwelling. The given range refers to the deeper and shallower area of the reef, respectively

Full size image

Discussion

This study developed a new methodology, to successfully map and estimate cold-water coral reef biomass and from that to make the first extrapolations of reef-scale C turnover and standing stock. These calculations were performed for one of the best-studied CWC reefs in the world and provide a much needed baseline assessment at a time of rapid climatic change when marine food webs are predicted to undergo rapid transitions. In addition, such biomass maps can guide sampling and monitoring expeditions, help identify areas that should be protected from human activities and even potentially predict areas that could have marine biotechnological potential. Furthermore, this work advances our nascent knowledge of C storage in CWC habitats and the significance of their secondary productivity, one of the criteria used to define ecologically or biologically significant areas (EBSAs) (Burrows et al. 2014; Titschack et al. 2015; Johnson et al. 2018).

This study further emphasises the importance of CWC reefs as hotspots for benthic mineralisation (Cathalot et al. 2015; Rovelli et al. 2015). Carbon turnover at the MR1 was found to be three to seven (with a mean of four) times higher than the global average for soft-sediment at the same depth (Glud 2008). By taking the spatial heterogeneity of L. pertusa and S. coralliophaga into account, the mean amount of oxygen consumption found here (32.37 mmol m−2 d−1) was estimated to be slightly higher than Rovelli et al.’s (2015) finding at MR1 (27.8 mmol m−2 d−1). However, the value reported here falls within the range that de Froe et al. (2019) found for CWC communities at the deeper Logachev Mounds and that White et al. (2012) found at the shallower Tisler Reef. While complex to undertake, the AEC technique has many merits since it measures community oxygen consumption, thereby inherently accounting for the complex and highly variable benthic activity associated with biogenic reef communities (Cathalot et al. 2015; de Froe et al. 2019; Rovelli et al. 2015). However, the size of an AEC footprint (i.e. the area of the benthic habitat integrated within the AEC flux measurements) varies with local conditions such as bottom roughness and deployment strategies (i.e. measurement height above the reef). For example, at Mingulay, the estimated AEC footprint size was 15 m2, while at the Logachev Mound Province, it was estimated to be 500 m2 (Rovelli et al. 2015; de Froe et al. 2019). The Mingulay AEC site was dominated by rubble, with little to no biomass of S. coralliophaga and live L. pertusa (see Fig. 1c in Rovelli et al. 2015). If the O2 consumption value from the AEC technique at the rubble-dominated site was to be considered alone, the annual organic C turnover of the reef would be as low as 204 T C yr−1. In our case, the colonies of L. pertusa and S. coralliophaga were under-represented under the AEC footprint, and their inclusion in this study added an additional 37 T C yr−1 to the reef’s total mean C turnover. This demonstrates that accounting for the heterogeneity of biomass can help to (a) improve the assessment of the role a CWC reef plays in the regional C cycling; (b) ground truth the upscaling of AEC measurements and (c) help place AEC instrumentation, strategically targeting locations that best represent average community compositions or locations dominated by a specific community.

While this study provides a novel approach to mapping biomass using data derived from HD videos, the resulted C turnover budget is affected by the uncertainties propagated across all of the measurements and assumptions present in each of the five steps (Table 4). The ability of automated underwater vehicles (AUVs) to collect more continuous photographic seafloor maps (Bodenmann and Thornton 2017; Thornton et al. 2016) has the potential to provide a more accurate method to convert surface area measurements to biomass. In addition, an increase in the availability of oxygen consumption and respiration rates could also affect the accuracy of the output values. For example, S. coralliophaga is typically covered by a species-rich community of sessile fauna (Kazanidis et al. 2016), which could not be included in the respiration rate measurements due to technical restrictions (i.e. the size of the respiration chambers available). The sponge’s biomass, however, accounts for ~ 90% (or more) of the total sponge plus epifauna biomass; therefore, a ~ 10% increase to the total respiration should be considered but remains to be confirmed. Also, this study assumed that the contribution to the C turnover of bacteria, infauna and epifauna growing on rubble and dead coral framework is equal, however, in reality, they might be different (Rowe et al. 1997; van Oevelen et al. 2009). It is also important to note that only a small proportion of the predictive models (5.88% for S. coralliophaga and 16.8% for L. pertusa) is explained by the environmental variables that we were able to take into account. The prediction will be improved with more relevant and higher resolution variables. It is an imperative to obtain such variables to produce predictive maps, especially if there is a clear relationship between biomass and particulate organic carbon (POC), for example. These considerations highlight that with more respiration and environmental data becoming available, C budgets will be further refined and improved (Stratmann et al. 2019).

Our calculations suggest that communities associated with dead coral branches (rubble and dead coral framework) contribute 79–97% (average 85%) of the total benthic C turnover at the MR1. Our estimate is slightly higher than the estimates from the deeper (500–800 m) cold-water coral carbonate mounds at the Logachev Mound Province (LMP) where dead framework has been estimated to contribute between 10 and 75% of the total benthic C turnover (de Froe et al. 2019). This range was calculated based on six box cores, sampled from two mounds, for which the dry weight of dead and live coral framework was calculated per square metre (de Froe et al. 2019).

There was 27 times more dead than living coral framework at the LMP, whereas at MR1, there was only three times more dead than living coral framework. Since dead coral framework C turnover is estimated to be 30 times lower than that of live coral framework (de Froe et al. 2019), the far higher proportion of live coral reef at MR1 likely explains the higher proportion of C turnover ascribed to live coral reef at MR1.

Furthermore, it is important to note that MR1 is located at much shallower depths where surface productivity is tightly coupled to the benthos through tidal downwelling (Davies et al. 2009), further explaining the high total C turnover compared to the LMP. This is also supported by the findings of Kazanidis et al. (2016), who found that S. coralliophaga at the MR1 hosted a much higher epifaunal biomass compared to S. coralliophaga found in the LMP.

In addition, there are gross differences in Lophelia pertusa colony morphology between LMP and MR1 (e.g. Vad et al. 2017; De Clippele et al. 2019b). Cold-water coral colony morphology is influenced by many interacting actors including settlement substrate stability and colony growth space constraints (Zibrowius 1980, 1984), local hydrodynamics, competition for food and space (Freiwald et al. 1997; De Clippele et al. 2018) and the presence of the non-obligate symbiotic polychaete Eunice norvegica (Linnaeus 1767) (Freiwald and Wilson 1998, Roberts 2005). Given the differences in depth, hydrodynamics and food supply between the two sites, we also see differences in skeletal thickening and polyp density. Indeed, visual observations indicate that the corals at the MRC have far more slender and denser packing of polyps compared to colonies at the LMP (Vad et al. 2017; De Clippele et al. 2019b). However, to quantify these differences, there is a need to measure the contribution of live and dead framework within and between sites to the total benthic C turnover. This is especially important in relation to some deeper reefs, where ocean acidification threatens to dissolve dead coral framework of which the majority of some CWC reefs consist (Hennige et al. 2015). Interestingly, sponges generally do not seem to be affected much by ocean acidification, suggesting that they might have a competitive advantage (Bell et al. 2018). However, the dominant growing substrate for S. coralliophaga could decrease owing to the dissolution of dead coral framework. The resil

ience to the effects of ocean warming appears to be species specific. For sponges, both positive and adverse effects have been listed in the literature (e.g. Bell et al. 2018; Stevenson et al. 2020). For L. pertusa, an increase in metabolic rate has been found when increasing seawater temperatures (Dodds et al. 2007). However, climate change-driven increases in ocean temperatures can cause secondary physical changes in the water column such as stratification and changes in the mixed layer depth resulting in lower nutrients and net primary production. A reduced or altered food supply, especially in combination with an increased metabolism could cause the benthic fauna to starve (Sweetman et al. 2017; Lesser and Slattery 2020).

One of the advantages of estimating the annual amount of C turnover is that one can quantify the importance of the different C supply pathways (e.g. by natural deposition and tidal downwelling) of C from surface water primary productivity (PP) to the reef. An average primary productivity in the water column above the MR1 of 0.048043 g m−3 day−1 can be calculated using an open-source marine data layer on mean PP from the Bio-Oracle website (Tyberghein et al. 2012; Assis et al. 2018). This equals to a yearly PP above the Lophelia reef habitat of 294 T C year−1 (Fig. 9), a value that is broadly similar to the here estimated area’s mean C turnover by the CWC community (241 T C year−1). Using the parametrization by Suess (1980), the amount of C reaching the seafloor from the sea surface via natural deposition was estimated to be 65–124 T C year−1, calculated for the shallowest (91 m) and deepest point (180 m) of Lophelia reef habitat area. To sustain the annual reefs’ mean C turnover rate reported in this study, an additional 117–176 T C year−1 (49–73% of the reef turnover) would have to be supplied through tidal downwelling and/or deep-water advection (see Fig. 9). When considering the calculated range in the reef turnover rate, an additional minimum 36–95 T C year−1 (23–59% of the reef turnover) or maximum 276–335 T C year−1 (69–84% of the reef turnover) would have to be supplied to the reef through tidal downwelling and/or deep-water advection. The estimated maximum amount of C supplied from PP through natural deposition (124 T C year−1) is close to the minimum amount of C reef turnover (160 T C year−1). This illustrates that monitoring and/or managing surface PP would be a key consideration for any conservation efforts of this CWC reef ecosystem. Isotopical analyses (δ13C, δ15N) of sponge and coral tissue samples from the MR support this finding, since they indicate L. pertusa, S. coralliophaga and associated benthic macrofauna primarily feed on fresh algal matter vertically transported from the water surface rather than on decayed organic matter being laterally advected (Duineveld et al. 2012; Kazanidis and Witte 2016). The lipid signature of L. pertusa at the MR also indicates an input of lipid-rich prey species (i.e. zooplankton prey) and a strong link to surface production (Dodds et al. 2009). Both the quality and the quantity of food resources have an essential role in the functioning of the coral reef (Duineveld et al. 2012), therefore changes in C supply processes and primary production could have a strong impact on the functioning of the reef.

Lastly, this study demonstrates the value of combining regional acoustic mapping, high-resolution video imagery and AEC measurements, allowing radically improved assessments of the dynamics and functioning of CWC communities. Automating surface area measurements from imagery using machine learning techniques and developing 3D visual imaging technology for seafloor mapping will provide new and more efficient pathways to assess deep-sea biomass and its contribution to the C cycle (Elawady 2015; Thornton et al. 2016; Bodenmann and Thornton 2017; Conti et al. 2019). The methodology developed here has the potential to be applied to other underwater habitats. By closing these significant gaps in our current knowledge on biomass and cold-water coral reef C turnover, we can increase the capacity to understand anthropogenic and climate change pressures and ensure the long-term health and resilience of these ecosystems.