Cold-water coral reefs and adjacent sponge grounds: hotspots of benthic respiration and organic carbon cycling in the deep sea

the seaﬂoor as detritus. It remains elusive how these hotspots of biodiversity can thrive in such a food-limited environment, as data on energy ﬂow and organic carbon utilization are critically lacking. Here we report in situ community respiration rates for cold-water coral and sponge ecosystems obtained by the non-invasive aquatic Eddy Correlation technique. Oxygen uptake rates over coral reefs and adjacent sponge grounds in the Træna Coral Field (Norway) were 9–20 times higher than those of the surrounding soft sediments. These high respiration rates indicate strong organic matter consumption, and hence suggest a local focusing onto these ecosystems of the downward ﬂux of organic matter that is exported from the surface ocean. Overall, our results show that coral reefs and adjacent sponge grounds are hotspots of carbon processing in the food-limited deep ocean, and that these deep-sea ecosystems play a more prominent role in marine biogeochemical cycles than previously recognized.


Introduction
Cold-water corals and sponges form complex reef structures on the seafloor (Figure 1), which support a rich community of suspension-feeding fauna and play a crucial role as a refuge, feeding ground and nursery for various commercial fishes (Miller et al., 2012).These reef ecosystems are widespread across the deep ocean (Klitgaard and Tendal, 2004;Roberts et al., 2006), but they face an uncertain future, as the human imprint on the deep-sea rapidly increases (Pusceddu et al., 2014).As oil and gas exploration are moving into deeper waters, seafloor installations, accidental oil spilling and sediment plumes are increasingly impacting the reef structures (Purser and Thomsen, 2012;Fisher et al., 2014;Roberts and Cairns, 2014), while physical disturbance by intensified bottom trawling significantly damages the slow growing reefs (Fossa et al., 2002;Althaus et al., 2009;Miller et al., 2012).Cold-water coral ecosystems are also considered to be strongly sensitive to various impacts of climate change.Foremost, ocean acidification is considered a major threat to their future existence, due to the anticipated shoaling of aragonite saturation horizon and the associated reduction in calcification rates (Roberts and Cairns, 2014).Energy availability is considered a prime factor in the sensitivity of cold-water corals to ocean acidification, and therefore, climate-induced shifts in the primary production and export from the surface ocean should also be considered.As rising sea-surface temperatures can reduce water column mixing and intensify the recycling of organic matter in the surface ocean, climate change can decrease the carbon export flux to the deep sea (Bopp et al., 2001).Moreover, the energy demands and respiration rates of the reef-forming species are predicted to increase in a warmer ocean (Dodds et al., 2007;Roberts and Cairns, 2014), thus aggravating the food limitation problem.While the sensitivity to ocean acidification is increasingly addressed in experimental and modeling studies, it is unclear how these climate-induced changes in food supply and respiration will affect the ecosystem functioning of cold-water coral and sponge communities.To resolve this, data on the energy flow and organic carbon (OC) processing rates within these deep-sea ecosystems are critically needed.
A recent food web model analysis from the Logachev Mound complex (NE Atlantic) suggests that OC processing rates in Cold-Water Coral Reefs (CWCR) largely exceed those of the surrounding soft sediments (Van Oevelen et al., 2009).Moreover, Kutti et al. (2013) estimated that reef-associated sponge grounds can also process considerable amounts of OC, based on sponge abundance data and laboratory-based respiration rates.Yet, until now, there has been no experimental in situ validation of this "high OC processing" hypothesis.Community respiration rates in reef environments are difficult to quantify, as the diversity of fauna, habitat complexity and up-scaling issues prevent the use of traditional techniques (e.g., incubation chambers, microsensor profiling).Here we report in situ measurements of community respiration rates for CWCR and adjacent sponge grounds, obtained by Aquatic Eddy Correlation (AqEC), which is a novel technique that allows the in situ determination of community respiration rates in topographically complex seafloor environments (Berg et al., 2003;Lorrai et al., 2010;Long et al., 2013;Attard et al., 2014).The AqEC technique is non-invasive, and integrates the benthic oxygen flux over a large footprint area (>10 m 2 , Berg et al., 2007).Our data show that carbon processing in CWCR and associated sponge grounds is much more intense than previously recognized: these ecosystems not only act as havens of biodiversity, but are true hotspots of organic carbon cycling in a food-limited deep sea.

Study Site and Sampling
Data were collected in May 2011 on board R/V Håkon Mosby, on a cruise to the Traena Marine Protected Area (MPA), which covers 300 km 2 of the Norwegian continental margin.The seafloor in the Traena Deep Coral reef field (Norwegian continental shelf) consists of three main habitat types (Buhl-Mortensen et al., 2010;Kutti et al., 2013): (i) elongated, cigarshaped reefs (100-150 m long, 25-55 m wide and on average 7 m high) formed by the reef-building coral Lophelia pertusa, (ii) dense sponge grounds surrounding the coral reefs, consisting of aggregations of large demosponges, and (iii) stretches of bare sediment, containing typical soft-sediment infauna, which are located further away from the reefs and sponge grounds.
We performed three types of data collection (Table 1): (i) Aquatic Eddy Correlation (AqEC) lander deployments to obtain in situ Community Respiration (CR) rates, (ii) Smøgen boxcorer deployments to retrieve soft sediment, living coral branches and dead coral branches for shipboard incubations, and (iii) dives with the Remotely Operated Vehicle (ROV) Aglantha to collect branches of living coral and samples of living sponges, and to obtain video footage for seafloor habitat mapping and estimates of coral and sponge densities.Deployments were done at five different locations: two reefs (R1 and R2), their associated surrounding sponge grounds (S1 and S2), and one control site with bare soft sediments (C).The control site did not host corals or sponges, and was located at a similar water depth, east of the Traena Coral MPA (Figure 1B).
CWCR in the Traena area are aligned along the main westward current direction (Laberg et al., 2002) and are typically composed of three distinct parts: a "head" part with living coral, a "middle" section with dead coral framework, and a long "tail" part, which consists primarily of coral rubble (Figure 1C).The reef 's head part faces the current and consists of several meters wide vertically-oriented lobes, primarily composed of living Lophelia pertusa, (Buhl-Mortensen et al., 2010, Figure 2D).The sponge ground sites were located in the vicinity of the reef sites and were dominated by three sponge types: Mycale lingua, Oceanapia spp., and Geodiidea (Kutti et al., 2013).

Biological Data on Corals and Sponges
We determined the coral polyp density from ROV video transects at site R1 (Table 2) along three different lobes of live Lophelia.The number of "visible" polyps per square meter was estimated on calibrated (i.e., corrected for the view angle of the camera) still images taken from the video.To this end, we delineated 3-4 squares of 10 cm by 10 cm in a single video image and counted the number of "visible" polyps within these reference squares.
By comparing the number of living polyps on intact coral branches collected by the ROV with the number of living polyps  that were visible on single branches in calibrated ROV video images, we estimated that 80 ± 10% of the living polyps were hidden from view on the ROV video images (Figure 1E).We hence multiplied the "visible" polyp density estimated from the ROV video images by a factor of 5 to arrive at the true areal density of living L. pertusa polyps.
We estimated the abundance of the four main sponges (Geodia barretti, G. atlantica, G. macandrewii, and other Porifera) by analyzing images of an ROV survey (HL5-5) carried out in 2009 at site S1 (Kutti et al., 2013).In each image, individual sponges were taxonomically identified and their density (ind m −2 ) and individual size (length, diameter) was estimated.
In addition, a scaling relation was derived between the volume and biomass of Geodia barretti (Table 3) to estimate the total sponge biomass.During the cruise, five specimen of G. barretti were collected with the ROV, and their individual volume (through water displacement) and wet weight was measured.Dry weight biomass was calculated from wet weight using a 0.2 conversion factor determined from previous measurements on Geodia sponges (Tjensvoll et al., 2013;Kutti et al., 2014).

Eddy Correlation in situ Community Respiration Measurements
Five deployments were performed within the three habitat types (1 in soft-sediment, 2 in sponge ground and 2 in coral reef habitat) using a tripod lander, which carried the Aquatic Eddy Covariance (AqEC) instrument and which was launched from the research vessel (Table 1).The AqEC flux technique is based on the simultaneous high-frequency measurement of the current velocity vector u and the oxygen concentration C at a given height above the seabed, from which the turbulent fluctuations in the oxygen concentration C ′ and in vertical current velocity u' z are calculated.The oxygen flux toward the seabed is subsequently obtained as the temporal average of the covariance u ′ z C ′ (Berg et al., 2003(Berg et al., , 2009)).The AqEC instrument consisted of an Acoustic Doppler Velocimeter (ADV, Vector Nortek R ) and a fast-responding Clark-type O 2 microelectrode directly connected to an auto-zeroing amplifier (Unisense R , Denmark).Both the ADV and the O 2 sensor were interfaced to a controller unit (Unisense R , Denmark), which took care of power supply, sensor synchronization and data storage.The O 2 sensors (Unisense R , Denmark) had an outer tip diameter of ∼5 µm, and a response time of ∼0.3 s.The O 2 sensor was polarized for at least 6 h prior to use and a two-point calibrated was carried out using saturated bottom water (100% reference) and an oxygen-free solution of 1 M Na-ascorbate and 0.5 M NaOH (0% reference).
The AqEC instrument was mounted on a tripod bottom lander equipped with a vane attached to one leg of the lander, which ensured that the O 2 sensor faced the direction of the main current (Figure 1).Visual inspection by the ROV confirmed the correct positioning of the AqEC lander on the reef framework.In all deployments, the sensor volume of the ADV was positioned 35-45 cm above the sediment-water interface.AqEC deployments typically lasted 2 h, except for the deployment in the bare sediment control area, which lasted 24 h.During this 24 h deployment, we selected 2 h (8 bursts) of data that passed quality control criteria (see details below).Data collection always proceeded in 15 min bursts, each consisting of a 14 min period of data recording at 64 Hz, followed by a 1 min pause.AqEC data time series were pre-processed using a dedicated R script developed at NIOZ, which involved de-spiking of the oxygen and velocity data, followed by a two-axis rotation of the velocity data (Mcginnis et al., 2008).In case of the reef deployments, we applied a planar fit rotation to the velocity data in order not to make any further assumptions on the flow field, and to examine the occurrence of any flow divergence that might generate spurious fluxes.A low-pass filter (3 Hz) was applied to the vertical velocity data to remove high-frequency noise.In oceanic environments, flux contributions are typically below 1 Hz (Lorrai et al., 2010), and we verified that our low-pass filtering did not affect the calculated fluxes.
In AqEC studies, three methods have been used to obtain the fluctuating components of the vertical velocity and oxygen (mean removal, linear detrending, and running mean filtering).In this study, we implemented running mean filtering because it gave the best linear cumulative fluxes (please see below).To this end, the mean velocity and oxygen concentration were calculated as the running means obtained by applying a 1/60 Hz high-pass filter to the 3 Hz filtered data in each burst.Referring to these running means as u z and C z , values of u ′ z and C ′ were then calculated as u zu z and C -C z where u z and C represent the 3 Hz filtered data.The AqEC flux was then calculated as the mean covariance over the time domain of each burst together with the associated cumulative flux.The cumulative flux is simply the value of the instantaneous AqEc flux (u ′ z C ′ ), which is integrated over time.If u ′ z C ′ is constant in time, this cumulative flux should increase in a linear way throughout each burst.Hence, the cumulative flux was visually inspected to reflect a steady linear uptake of O 2 by the seafloor during the respective bursts.If the cumulative flux did not display a sufficiently linear trend (r 2 < 0.6), the burst was discarded.A fast Fourier transformation of u ′ z and C ′ provides the individual spectra as well as the co-spectrum, which expresses the flux contribution as a function of the eddy frequency (Berg et al., 2003).After careful examination of the variance preserving co-spectra of each burst, the use of a 60 s averaging time period was selected (i.e., 1/60 Hz high-pass filter  as noted above) ensuring that the lowest frequencies contributing to the flux are accounted for.Quality control of the AqEC time series data is crucial for accurate flux extraction.Each burst was examined separately and the burst data were discarded if (1) ADV data were of insufficient quality (beam correlation values below 70%), (2) the current velocity and oxygen concentration showed any sign of statistical non-stationarity (by investigating changes in the mean and standard deviation of C and u z over time) or (3) storage effects were noted (by monitoring the oxygen inventory between sediment and the O 2 sensor position) (4) the cumulative flux was not sufficiently linear (Reba et al., 2009;Hume et al., 2011;Lorke et al., 2013).Hence, the entire 2 h data-sets for the reefs and sponges deployments, and the entire 24 h datasets for the sediment control deployment, were visually inspected for subsets with stationary conditions: we ensure that burst averaged velocity and concentrations showed only limited changes over time.For instance, on reef 1, burst averaged horizontal velocities vx in those stationary subsets ranged from −0.06 to 0 m s −1 and burst averaged oxygen concentrations ranged from 268 to 281 µM.Burst-to-burst changes of averaged oxygen concentrations were maximum 4 µM, which allowed us to disregard potential storage effects (Lorke et al., 2013).Additionally, each 15 min burst was examined separately.Bursts with clear artifacts in data (such as abrupt changes in averages of subsections) where discarded, as well as bursts where the cross correlation function did not show a clear minimum or maximum but was either entirely erratic or showed multiple extrema over the −100 -+100 s time lag.When the cumulative fluxes showed a random-walk like behavior rather than a linear trend, we co-examined the co-spectra, ogives, crosscorrelation function and discarded the burst if they all presented significant anomalies.
We used the cross correlation among the horizontal u x and u y and vertical u z velocity components to calculate the friction velocity (Kim et al., 2000), and this way, we could assess a rough estimate for the footprint area of our AqEC measurements (Berg et al., 2007).This calculation revealed that footprints covered an area about 10-25 m long and 2.3-2.9 m wide in the various deployments.

Incubation-Based Community Respiration Rates
Living coral branches (Lophelia pertusa) and dead coral branches (Lophelia pertusa) were collected by the ROV and their oxygen consumption rates were determined by closed chamber incubations (Table 1).
Coral fragments were placed in acrylic chambers (I.D. 10 cm) filled with filtered (50 µm pore size) bottom water collected by a CTD rosette sampler.Chambers were then closed with a lid equipped with a magnetic stirrer, temperature and oxygen sensors (FIBOX oxygen optode, PreSens Gmbh).Incubations were performed in the dark at in situ temperature in a temperature-controlled shipboard incubator.Temperature and oxygen concentrations were continuously logged during incubations.Oxygen concentrations in the chamber were additionally determined at the beginning and end of each incubation via Winkler titration (Grasshoff, 1976).Oxygen consumption rates were computed by linear regression of the oxygen concentration data over the first period of the incubations (total O 2 decrease < 20%).The O 2 consumption rate of intact bare sediment substrate was determined in a similar way, using intact sediment cores subsampled from the boxcore.The duration of the incubations was 7.9 ± 2.6 h, and the O 2 concentrations had by then dropped on average by 13%.On-board respiration rates for sponges (Geodia barretti) were obtained in a similar way, using larger incubation chambers, yet during earlier cruises (transect HL5-5, in Kutti et al., 2013).
To validate the AqEC measurements, whole ecosystem respiration rates were estimated in a "bottom-up approach" by suitably combining the respiration rates of separate habitat components and upscaling these within the footprint area of the AqEC instrument.This procedure was applied to the 3 habitats (coral reef, sponge ground, and soft-sediment) where the AqEC lander was deployed.
Soft-sediment community respiration (CR) rates were considered as directly equal to the rates obtained by incubating the sediment substrate.
Sponge CR rates were calculated as the soft-sediment CR rates summed with the sponges incubation-based respiration rates.Indeed, we considered the sponge ground ecosystem as the sum "sponges + sediment." Coral reef community respiration (CR) rates were calculated as the area-weighted mean of the four main habitat types that covered the AqEC footprint area: (1) live Lophelia, (2) mixed and dead Lophelia framework, (3) dead coral framework, and (4) coral rubble.To determine the contribution of each habitat type, ROV video transects were carried out over site R1 from the AqEC deployment location up to the head part of the reef (thus mapping the habitat variation in the footprint area of the AqEC instrument along the longitudinal axis of the reef).To each habitat type, we subsequently assigned a specific community respiration rate.Respiration rates of dead coral framework and coral rubble were obtained as described above for the coral branches collected by the ROV.To estimate respiration rate of the live Lophelia habitat, we accounted for the 3D structure of the reef and the whole polyp activity: we counted the polyps on the incubated living coral branches and normalized the oxygen consumption rates by the number of polyps.Live Lophelia respiration rates were hence calculated as the polyp normalized live respiration rate (i.e., oxygen consumption rate per polyp) multiplied by the polyp density of the reef estimated through the ROV transect (i.e., number of polyp per square meter) and corrected from the 5 layers of polyps (multiplied by 5, see above).

Upscaling Procedures of the CR Rates
Both of our AqEC and bottom up CR rates were upscaled over the entire coral reefs by considering the cigar shape of the reefs and based on 3D reconstruction of reef 1 from the ROV images and the positioning of the EC lander (see Supplementary Materials, length: 125 m, width: 45 m).The reefs were modeled as two joint domains (see Supplementary Materials): one dominated by live corals (area A1: 1767 m 2 ,), and the other one dominated by dead coral framework and coral rubble sediment (area A2: 1688 m 2 ).Reef CR rates, calculated above, were taken as representative of the domain A1, whereas sediment CR rates were taken as representative for domain A2.Upscaled CR rates were then calculated as the area weighted mean using the following formula: For larger scale considerations and regional estimates of CR rates over reef and sponge beds, we chose to consider only the rates obtained by the AqEC technique, in order to take a more integrated value of the ecosystem.For the sediment control area however, we chose to average the rates obtained by both techniques, since AqEC and incubation would give a representative estimate of the CR rates.

Error Propagation and Statistical Differences
Statistical differences between the AqEC and bottom-up techniques were tested among the three habitats using the Wilcoxon rank signed sum-test.
In the error propagation, we accounted for the standard deviation of fluxes using standard basic rules (please refer for instance to http://www.rit.edu/cos/uphysics/uncertainties/Uncertaintiespart2.html).
Hence, taking into account the associated standard deviation of the fluxes, we calculated the average between (i) AqEC fluxes measured over identical habitats (e.g., reefs R1 and R2, sponge grounds S1 and S2), (ii) incubations based CR of different coral compartments, and iii) AqEC and incubation based CR over sediment area.

Reef Habitat Composition
The seafloor in the Traena Deep Coral reef field (Norwegian continental shelf) comprises three main habitat types (Buhl-Mortensen et al., 2010;Kutti et al., 2013): (i) elongated, cigarshaped reefs (100-150 m long, 25-55 m wide and on average 7 m high) formed by the reef-building coral Lophelia pertusa, (ii) dense sponge grounds surrounding the coral reefs, consisting of aggregations of large demosponges, and (iii) stretches of bare sediment away from the reefs, containing typical soft-sediment infauna.Video transects in the footprint area of the AqEC lander at the Reef site R1 revealed a mosaic of four main reef habitat types: live coral framework (covering 53% of the transect), mixed live and dead coral framework (8%), dead coral framework (12%), and coral rubble (27%) (Table 4).

Community Respiration (CR) Rates
Representative data time series obtained during AqEC deployments for the three habitat types (reef R1, sponge ground S1 and control bare sediment C) are shown in Figure 2. Community Respiration (CR) rates, obtained by AqEC, differed by two orders of magnitude between the three habitats, revealing strong differences in carbon processing rates (Figure 3).The AqEC-based CR of the bare sediment site was 6 ± 2 mmol O 2 m −2 d −1 and compared favorably to CR obtained from shipboard incubations of sediment cores (3 ± 1 mmol O 2 m −2 d −1 ).These low CR values are also consistent with rates (≈1 mmolO 2 m −2 d −1 ) previously reported for the Norwegian continental shelf at comparable water depths (Graf et al., 1995).AqEC-based CR rates within the sponge grounds were ∼10 times higher (54 ± 5 mmol O 2 m −2 d −1 ) than the bare sediment rates and compared well with the bottom-up estimate at the sponge ground site (48.7 ± 21.5 mmol O 2 m −2 d −1 ), which was obtained by combining and integrating the individual respiration contributions of different habitat types within the AqEC footprint area (Table 3).CR rates over the coral reef were systematically higher than in the adjacent bare sediments and sponge grounds (Figure 3).The highest AqEC-based CR rates (246 ± 18 and 177 ± 40 mmol O 2 m −2 d −1 ) were obtained during the deployments on the head part of the reef, which is densely covered with living Lophelia pertusa (Figures 1B,D).The bottom-up CR estimates over CWCR (82 ± 10 mmol O 2 m −2 d −1 ; Table 4) are about 40% of these AqEC fluxes.However, comparing the two AqEC and bottom-up methods over the 5 deployments (i.e., all 3 habitats), CR rates were not statistically different (p = 0.25, n = 5).Upscaling our AqEC measurements to the whole reef structure (including the downstream tail part that is less metabolically active) provided an average CR rate of 122 ± 10 mmol O 2 m −2 d −1 (Table 5), whereas upscaled bottom-up CR rates gave a value of 43 ± 7 mmol O 2 m −2 d −1 .Ratios between the reef and the sediment CR rates were 20 and 14, respectively for the AqEC and bottom-up technique.For the sponge habitat, ratios were of 9 and 16, respectively for the AqEC and bottom-up estimates.

Discussion
Despite the complex topography of the reef structure and the substantial uncertainty associated with the "bottom-up" CR rates, the AqEC and bottom-up values align very well, and are not significantly different for the sponge ground and bare sediment.However, AqEC technique did provide significantly higher CR rates than the "bottom-up" incubation method over the reef habitats (Figure 3).This difference can be either due to an overestimation on behalf of the AqEC technique, or equally, an underestimation associated with the bottom up incubation approach.Clearly, the bottom-up approach requires a whole series of upscaling steps, where each upscaling step introduces a new set of parameters (e.g., polyp density, local habitat cover), which each are associated with a given uncertainty.These uncertainties eventually propagate through the upscaling process, which creates the possibility for a substantial uncertainty on the final community respiration rate.On the other hand, the application of the AqEC technique over a topographically complex habitat, such as a cold-water coral reef, is also challenging, and one should be vigilant that the basal assumptions underlying the AqEC technique are still satisfied.One critical assumption is that flow convergence or flow divergence are negligible, which is not necessarily the case in an environment with complex and rough topography, such as coral reefs and sponge grounds.When present, flow convergence/divergence may induce vertical velocity variations that can induce higher, spurious fluxes (Reimers et al., 2012; Each habitat CR rates is calculated as the sum of its different compartments (Formula).Areal habitat CR rates are each individual habitat CR rates scaled-up to the entire reef using their respective surface area contribution.Lorke et al., 2013).We specifically assessed the importance of flow divergence at the reef site, by implementing a planar fit rotation to the ADV velocity data.This coordinate rotation aligns the mean horizontal velocity along the main streamline and hence one does not need to invoke any assumption regarding the flow field (Lorke et al., 2013).We found that the AqEC fluxes did not change after the application of the planar fit rotation, which hence suggests a negligible impact of flow divergence on the fluxes.Moreover, hydraulic studies examining the influence of forward facing steps on the flow pattern (Ren and Wu, 2011), have shown that for similar Reynolds numbers (Re ∼400) as over the reefs, the effect of flow divergence becomes negligible after ∼1.5 times the length of the step (in our case about 7 * 1.5 = 10.5 m, while the AqEC lander here was positioned 20 meters behind the 7 meter step increase).Accordingly, we believe that flux divergence was most likely negligible in our reef measurement, but future studies could however consider a more detailed and more extensive evaluation of the AqEC technique in topographically complex environments like coral reefs (e.g., by placing multiple instruments in parallel, or examining whether the flux at different sensor heights).
Both the bottom-up approach and the AqEC methods show that the CR rates on the reef and sponge grounds are substantially elevated compared to the bare sediment control site.These high CR rates evidently suggest intensified organic processing within the CWCR and sponge grounds of the Traena MPA.Based on our AqEC rates (which we consider more reliable than the corresponding bottom up estimates), community respiration in CWCR and sponge grounds are, respectively, ∼20 and ∼9 times higher than those of the surrounding soft sediments.The observation that CWCR and sponge grounds function as hotspots of respiration is supported by indirect estimates of community respiration at other locations in the North Atlantic (Van Oevelen et al., 2009;White et al., 2012).Recently, Rovelli et al. (2015) reported respiration rates up to 46 mmol O 2 m −2 d −1 in CWCR habitats in the Stjernsund area on the Norwegian shelf, which were also significantly higher than within adjacent bare sediment areas.If we compare the few available data against a global compilation of soft sediment data, CWCR and sponge grounds clearly stand out by their high community respiration rates (Figure 4), although it is clear that there is variability among the few CR rates available in the literature.Indeed, CWCR and sponge ground sites can differ significantly both in terms in density and structure, impacting the absolute values of the CR rates between different sampling sites.However, the common feature of CWCR and sponge field among all studies is their carbon processing rates significantly higher than the surrounding environments (Figure 4).
The high metabolic activity of CWCR and sponge grounds has important implications for carbon cycling and benthic-pelagic coupling at the regional scale of the Traena MPA, but also the wider scale of the Norwegian continental shelf (Table 5).Using the CR rates obtained here, and considering a ∼22% seafloor coverage (Lindberg, 2004;Kutti et al., 2013), CWCR and sponge grounds account for 78% of the total benthic respiration, and consume 63% of the total OC export flux in the Traena MPA (Slagstad et al., 1999).CWCR habitats cover ∼1671 km 2 of the Norwegian continental shelf (150.000km 2 , Fossa et al., 2002), while recent ROV observations suggest at least a similar coverage for sponges (Cruise report, R/V Håkon Mosby-cruise #2009615).Using these estimates of seafloor coverage, CWCR and sponge grounds are jointly responsible for 36% of the total benthic respiration on the Norwegian continental shelf, and process 5% of the total primary production in the area (Schluter et al., 2000).Conventional carbon budgets typically only account for soft-sediment respiration and hence, these budgets may significantly underestimate total seafloor carbon cycling and the intensity of the benthicpelagic coupling on continental shelves with coral and sponge reefs.
The intense metabolic activity of CWCR and sponge grounds can only be maintained through a local focusing of the downward export flux of detritus from the surface ocean.CWCR are frequently reported from sites with locally accelerated currents, along margins with enhanced down-slope particle transport, or where internal tidal waves facilitate and increase the seabed food supply (Roberts and Cairns, 2014).The carbonate framework baffles currents and enhances organic matter trapping, thus providing a detritus-focusing mechanism.This way, CWCR act as ecosystem engineers, directly modulating the availability of resources to their surrounding environment (Mienis et al., 2009;Van Oevelen et al., 2009).Compared to corals, sponges form smaller, decimeter-sized, structures on the seafloor, but are known for their impressive filtration capacity.In the Traena MPA, the local sponge community has been estimated to filter about 1 m of the benthic boundary layer each day (Kutti et al., 2013).We hypothesize that the focusing of the OC export flux onto CWCR and sponges, through organic matter trapping and filtration (Kutti et al., 2013), will reduce the detritus input to the peripheral soft-sediment communities, impacting the local biodiversity and carbon cycling.By dominating benthic OC processing in Traena MPA, CWCR and sponge grounds will also influence regional ecosystem functions, such as carbon storage and nutrient recycling.As a consequence, future seafloor carbon budgets should account for the intense biogeochemical cycling in CWCR and sponge grounds.Moreover, the interplay between the high respiration activity described here, and the anticipated Comparison between the AqEC respiration rates and the Organic Carbon (OC) export flux and Primary Production (PP).*Benthic respiration rates based on our AqEC measurements.Benthic oxygen uptake for bare sediment was however calculated as the average of our AqEC fluxes and bottom-up rates taking into account the variance of each (see the error propagation section).+ Respiratory quotient: 0.77 (Froelich et al., 1979).† Total export flux: 57 gC m −2 y −1 (Slagstad et al., 1999).‡ Primary Production: 112.7 gC m −2 y −1 (Schluter et al., 2000).a From (Fossa et al., 2002).b Based on (Klitgaard and Tendal, 2004), we assumed a similar coverage for sponges than for corals.c From (Schluter et al., 2000;Huthnance, 2010).(Rovelli et al., 2015).Sponge ground data for Tranea reef are also included ( ).The solid line shows an empirical model fit (double-exponential decrease with water depth) to the compiled soft sediment data (Andersson et al., 2004).
climate feedbacks on organic matter export from the surface ocean flux (Bopp et al., 2001), emphasize the sensitivity of deepsea communities to climate change (Roberts and Cairns, 2014).

FIGURE 1 |
FIGURE 1 | (A) Bathymetry map of the Traena Deep Coral reef field with red dots representing known cold-water coral reef locations.(B) Sampling sites: two reefs were investigated and the symbols mark the locations of AqEC deployments.(C) Topography of Reef R1.The location of the AqEC deployment is indicated by the gray arrow.The size of the upstream AqEC footprint area, as indicated by the oval shape above the reef, was estimated using the equations in Berg et al. (2007) using the cross correlation among

FIGURE 2 |
FIGURE 2 | AqEC correlation data for consecutive 15 min bursts during the deployment at the reef R1, sponge ground and bare sediment sites.(A) Velocity data (x, y, z) after a 3 Hz low-pass filtering.

FIGURE 3 |
FIGURE 3 | Community respiration rates in bare sediment (left), sponge ground (center), and cold-water coral reef (right).AqEC, Aquatic Eddy Correlation.Bottom-up, Bottom-up community respiration rate.AqEC fluxes correspond to separate deployments over two reefs, two

TABLE 1 |
Station locations and instrument deployments.

TABLE 2 |
Polyp density of Reef R1 based on ROV video survey images.
n.b., Number of polyps were estimated based on calibrated snapshots (factor: 0.725).

TABLE 3 |
Estimation of the sponge density and biomass from ROV video survey images and of respiration rates.

TABLE 5 |
Carbon budget over the Traena Marine Protected Area (MPA), and the Norwegian Continental Shelf.