A deep scattering layer under the North Pole pack ice

Abstract The 3.3 million km2 marine ecosystem around the North Pole, defined as the Central Arctic Ocean (CAO), is a blind spot on the map of the world’s fish stocks. The CAO essentially comprises the permanently ice-covered deep basins and ridges outside the continental shelves, and is only accessible by ice-breakers. Traditional trawling for assessing fish stocks is impossible under the thick pack ice, and coherent hydroacoustic surveys are unachievable due to ice-breaking noise. Consequently, nothing is known about the existence of any pelagic fish stocks in the CAO, although juveniles of Boreogadus saida richly occur at the surface associated with the sea ice and ice-associated Arctogadus glacialis has been reported as well. We here present a first indication of a possible mesopelagic fish stock in the CAO. We had the opportunity to analyse a geophysical hydroacoustic data set with 13 time windows of usable acoustic data over a transect from 84.4 oN in the Nansen Basin, across the North Pole (90.0 oN), to 82.4 oN in the Canada Basin. We discovered a deep scattering layer (DSL), suggesting the presence of zooplankton and fish, at 300-600 m of depth in the Atlantic water layer of the CAO. Maximum possible fish abundance and biomass was very low; values of ca. 2,000 individuals km-2 and ca. 50 kg km-2 were calculated for the DSL in the North-Pole area according to a model assuming that all acoustic backscatter represents 15-cm long B. saida and/or A. glacialis. The true abundance and biomass of fish is even lower than this, but cannot be quantified from this dataset due to backscatter originating from pneumatophores of physonect siphonophores that are known to occur in the area. Further studies on the DSL of the CAO should include sampling and identification of the backscattering organisms. From our study we can conclude that if the DSL of the CAO contains fish, their biomass is currently too low for any sustainable fishery.


Introduction
The "Central Arctic Ocean" (CAO) is defined as the permanently icecovered deep basins and ridges outside the continental shelves of the Arctic Ocean ( Fig. 1; PAME, 2013). For the CAO Large Marine Ecosystem (LME), there is an almost complete absence of knowledge of the occurrence of pelagic fish as a key ecological link between zooplankton and seals, whales and polar bears (Van Pelt et al., 2017). The reason for the absence of data from the CAO LME is related to the difficulty of accessing this remote cold area for on-site research due to its 2-3 m thick year-round sea-ice cover. Traditional trawling for assessing fish stocks is impossible under the pack ice. The area is only accessible by icebreakers, and acoustic data targeting the water column cannot be used if acquired while ice-breaking due to the high noise level. The fish stocks of the (partly) seasonally ice-covered Arctic shelf seas are relatively unknown as well, apart from those of the well-investigated, monitored and managed Barents Sea (Hansen et al., 2019;ICES, 2019a,b).
The CAO LME is changing rapidly as a result of global climate change. Large ecosystem transformations are likely happening with the loss of the sea ice, even before the baselines of its ecological communities are properly known and marine assessment and conservation measures are in place (Christiansen et al., 2014;Bluhm et al., 2015;Harris et al., 2018). The CAO is becoming a more dynamic ecosystem with global warming, as the marginal ice zone moves further north in summer and south again in winter (Overland and Wang, 2013). The extent of both the Arctic summer and winter minimum sea-ice cover has consistently been breaking negative records during the past decades: the minimum summer Arctic sea ice extent has decreased with an estimated loss rate of ca. 1 million km 2 (ca. 13.2%) per decade between 1979(IPCC, 2013Barnhart et al., 2015;Ding et al., 2017). Simultaneously, the sea ice is also thinning (Laxon et al., 2013, Kwok, 2018, with an estimated loss rate of ca. 3,100 km 3 (ca. 13.5%) per decade over the period 1979-2017. Climate models for the Arctic region predict a further decline of the summer sea-ice cover to below 1 million km 2 within the coming 30 years, depending on which political decisions are made and implemented at a global scale (Duarte et al., 2012;Screen and Williamson, 2017).
The rapid environmental changes in the CAO LME create new circumstances that may impact ecological and social values, including future commercial fisheries (Christiansen et al., 2014). A large part of the CAO is "High Seas" area ( Fig. S1), i.e., international waters beyond the Exclusive Economic Zones (EEZs) of the Arctic coastal states, where all states have rights and obligations according to the United Nations Convention on the Law of the Sea (UNCLOS, 1982). Lessons learnt from other marine LMEs worldwide are that (1) exploitation of newly accessible natural resources tends to precede scientific research and effective management measures, and (2) internationally shared fish stocks in High Seas are especially prone to overexploitation (McWhinnie, 2009). Recognising that, apart from a few anecdotal reports (e.g., single unidentified fish observed from submarine windows), nothing is known about the existence of any pelagic fish stocks (FiSCAO, , 2018, a precautionary approach has recently been taken for the CAO before any exploitation of its potential fishery resources. In November 2018, politicians from nine nations (including the five coastal states) and the EU decided to put "science first" and not to engage in commercial fishing for 16 years after the Agreement to Prevent Unregulated High Seas Fisheries in the CAO (Hoag, 2017;EU, 2019) has entered into force. This Agreement also includes the installation of a Joint Program of Scientific Research and Monitoring for the CAO. To date, the agreement has been ratified by nine of the ten Signatories. Commercial fishing does August and 15 September 2016. The stations were located along a transect between 83.2 • N in the Nansen Basin, across the Amundsen Basin, the North Pole, the Lomonosov Ridge, and ending in the Canada Basin at 82.4 • N. Each of the stations represents a time window during the expedition when the icebreaker was drifting with the sea ice at <1 knot, i.e., not causing excessive noise from breaking ice. Apart from Station 4 at the Lomonosov Ridge (1,244 m), the water depth at all stations was >2,000 m with maximum depth at Station 11 in the valley along the Gakkel Ridge (4,952 m). SD = the Sofia Deep, the hitherto northernmost reported mesopelagic scattering layer at 82.10 • N . The background map was extracted from the International Bathymetric Chart of the Arctic Ocean (IBCAO), Version 3.0 (Jakobsson et al., 2012). Physiographic classifications of the CAO and the High Seas (the area outside national Exclusive Economic Zones) of the Arctic Ocean are shown in Fig. S1. not occur in the CAO today because the yearly ice-free time window is still only local and short (some weeks). Furthermore, no resources of potential commercial value have so far been detected (FiSCAO, , 2018Snoeijs-Leijonmalm et al., 2020).
In this paper we present a first indication of what could be mesopelagic fish under the pack-ice cover of the CAO. We discovered a consistent deep scattering layer (DSL), i.e., a horizontal zone with living organisms (zooplankton and fish) detected as acoustic backscatter by an echosounder. The DSL was detected in the Atlantic water layer at 300-600 m of depth with highest density around the North Pole. DSLs are a near universal feature throughout the world's oceans at depths of 200-1,000 m (Irigoien et al., 2014;Klevjer et al., 2016). A mesopelagic DSL occurs across the entire Fram Strait at the Atlantic entrance to the Arctic Ocean (Gjøsaeter et al., 2020; Fig. 1), but whether a DSL would occur in the CAO outside the continental shelves remained an open question. We were able to perform the present study through access to data collected under the pack ice of the CAO during the "Arctic Ocean 2016" expedition with the Swedish ice-breaker (IB) Oden (Gårdfeldt and Lindgren, 2016). Acoustic and oceanographic data were collected along a transect from 84.4 • N in the Nansen Basin across the Amundsen Basin, the North Pole (90.0 • N), and the Lomonosov Ridge to the Canada Basin at 82.4 • N.

Area description
The boundaries of the CAO LME basically follow the shelf edge, as defined by the Arctic Council (PAME, 2013; Fig. 1), but in the Canadian Basin, the boundary between the CAO and Beaufort Sea is along 76 • N, leaving the deep southern portion of the basin as part of the Beaufort Sea LME. The CAO is the largest of the Arctic LMEs with an area of ca. 3.3 million km 2 , and essentially comprises the deep basins of the Arctic Ocean as well as ridges and sea mounts within the area (Fig. S1a). The most prominent feature of the CAO is the drifting sea ice that covers more or less the whole area during both winter and summer. However, the multi-year pack ice is disappearing fast and is being replaced by seasonal sea ice (Kwok, 2018).
The part of the Arctic Ocean that is most vulnerable to possible future exploitation of fishery resources is the ca. 2.8 million km 2 High Seas area outside the EEZs (Fig. S1b). The High Seas area overlaps to a large extent with the CAO, but includes also smaller portions of continental slopes and shelf regions, most notably the Chukchi Plateau (Fig. 1). The CAO and the High Seas contain about equal proportions of ridges (40-43%) and abyssal plains (33-34%) whereas the CAO contains more continental rises (22% compared to 15% for the High Seas) and the High Seas contains more continental shelf (3% compared to 0% for the CAO) (Fig. S1).

Acoustic sampling stations
Acoustic broadband backscatter data were collected from the water column between 15 August and 15 September 2016 during the "Arctic Ocean 2016" expedition with IB Oden (Gårdfeldt and Lindgren, 2016). The acoustic data grouped into 13 geographically separate areas with usable acoustic measurements corresponding to the geographical positions where station work (CTD and other) was conducted during the expedition (Fig. 1). Each of the stations represents a time window when the engines of the icebreaker were turned off and the ship was drifting with the Arctic transpolar ice drift (Spall, 2019) at <1 knot, i.e., not causing excessive noise from breaking ice. Apart from Station 4 at the Lomonosov Ridge (1,244 m), the water depth at all stations was >2,000 m with maximum depth at Station 11 in the valley along the Gakkel Ridge (4,952 m).
A comparison of the 13 areas showed that the minimum distance between them was 30 nmi (56 km) while the maximum distance within areas was 8 nmi (15 km). The midpoint of each area (average latitude and average longitude) is considered one sampling station (Table 1). The stations were labelled "Stations 1-13" along a transect starting at 83.2 • N in the Nansen Basin, across the Eurasian Basin (Nansen Basin, Gakkel Ridge, Amundsen Basin), the North Pole (90 • N), the Lomonosov Ridge, and ending in the Amerasian Basin (Alpha Ridge and Canada Basin) at 82.4 • N. All 13 stations fall within the CAO LME and 12 stations fall within the High Seas (the exception is Station 13 in the Norwegian EEZ) (Fig. S1).

Acoustic data collection and scrutinization
The acoustic data were originally collected for geophysical and oceanographic purposes to study double diffusive convection and vertical heat transport (Stranne et al., 2017) as well as mixed layer depth variability (Stranne et al., 2018), but we used them in the present paper as "data-of-opportunity" to detect living organisms. The data were not optimal due to the combination of high noise levels, acoustic backscatter measured over a single narrow frequency band, and low densities of organisms. However, after frequency selection, pulse compression and noise removal, the echograms were suitable for echo counting and confirming the existence of a DSL.
Acoustic backscatter was recorded from an echosounder (Simrad EK80) configured with a resonant (18 kHz) transducer  behind a 40-mm-thick polycarbonate plate ("window") mounted in the ice knife, aft of its cutting front. The echosounder pulse lengths were adjusted over a range of 1-8 ms with a linear up-sweep chirp from 15 to 30 kHz with fast up-ramping of the pulse, using 1.6% of the pulse duration for up-ramping. The transmit-power was set to 2000 W. It was synchronized with the on-board multibeam echosounder (12 kHz) and sub-bottom profiler (2.5-7 kHz) throughout the data collection period. The data were logged directly to large storage arrays in the Simrad raw format.
When inspecting the raw data, we found that the noise level within the echosounder band increased by 20-30 dB when the ship was steaming through and breaking ice, and for this reason the data could not be used. However, when the icebreaker was drifting with the sea ice at <1 knot, i.e., not causing excessive noise from breaking ice, the noise level was reduced to about − 140 dB rel. 1 W in active mode (Simrad, 2018), and we were able to observe backscatter from individual living organisms in the water column. Based on these observations, a plot of the ship speed was used to select all the data with ship speed < 1 knot. Only these data were further processed and scrutinized.
The data selected on the basis of ship speed was pre-processed using the software LSSS (Large-Scale Survey System; Korneliussen et al., 2006Korneliussen et al., , 2016 to remove noise and detect tracks, as well as to provide information that might be helpful when allocating backscatter to acoustic categories. Most of the remaining noise while the ship was drifting with the sea ice seemed to be electrical interference from the vessel's power system or other instruments. A typical signal-to-noise ratio of 10-12 dB could be achieved in the mesopelagic layer from 300 m down to 600 m depth. This mostly corresponded with the maximum depth of the observed backscattering of single targets. The data pre-processing (prior to scrutiny) proceeded as follows: 1. The nominal frequency was 18 kHz, but the original broadband data spanned over 15-30 kHz. The band 18 ± 2 kHz (16-20 kHz) was pulse-compressed and used in further processing because inspection of the data showed that background noise was at a minimum within this band (Fig. S2a). 2. The data were smoothed vertically with a Gaussian kernel of 0.5 m in diameter, i.e., the samples 0.25 m below and 0.25 m above were given half the weight of the sample centre during smoothing. 3. Spikes with duration longer than 1 ms were removed. In the slightly overlapping depth ranges 10-100 m, 90-250 m and 240-2400 m, spikes that were respectively 12 dB, 9 dB and 6 dB stronger than the The reason for requiring spikes to be stronger at short ranges is that the beam footprint is narrower at short ranges, i.e., small schools could more easily be detected by one ping at short ranges than at long ranges, and thereby be detected as a spike. The reason for the overlapping of the ranges is to be able to remove spikes that cross two depth ranges with different settings. 4. The time-varying ambient noise was quantified and corrected according to Korneliussen (2000). The measured samples were corrected by subtracting the estimated average noise for the measured sample, and by removing all samples that were weaker than the estimated high noise cut-off. This means that all measurements that were so weak that they were inside the noise pdf (probability density function) were set to zero, and all stronger measurements were reduced by the estimated average noise.
The pre-processed data were further scrutinized in the water column down to 1000 m using LSSS (Korneliussen et al., 2006(Korneliussen et al., , 2016. The density estimates were checked using echo counting, as described by Mitson and Wood (1961), later refined by Kieser and Mulligan (1983), and for target strength (TS) measurements by Ona and Hansen (1986). Most single targets and single-target tracks were detected in the mesopelagic layer at 300-450 m depth and only few below 600 m. Apart from intervals at a few stations, the epipelagic layer (the upper 300 m) was much noisier than the mesopelagic layer. Backscatter measurements within the first 0.4 s after transmission were generally noisier than those made in the period after 0.4 s. There was intermittent noise interference in the top 300 m of the water-column that was likely due to instruments on-board the ship. During periods where the noise was absent, the density of targets above 300 m seemed to be much lower than between 300 and 600 m. Unfortunately, the consistency of the target density above 300 m could not be assessed due to the intermittent noise. The data were stored in the LSSS database as vertical profiles with 10 m depth resolution where each profile was averaged over 25 pings.
The acoustic equipment was calibrated on-board IB Oden on 1 September 2015, following a standard methodology described by Foote et al. (1981). This was a centre calibration with a CU64 mm sphere, where the data were collected within 1 degree off axis. Examination of the calibration results showed that data collected at pulse lengths of 8, 4, and 2 ms were acceptable, whereas data for 1 ms pulse length could not be used due to an "overshoot" in amplitudes in the spectrum. The results showed a large calibration offset of >10 dB, with correction factors of 11.5, 12.0, and 15.8 for backscattering obtained with pulse lengths of 8, 4, and 2 ms. The corresponding corrections for individual TS were − 5.3, − 5.4, and − 6.0 dB at the same pulse lengths (Table 1). Since the calibration was performed one year before the measurements were made, we tested the stability of the Simrad EK80, 18 kHz, system onboard IB Oden by comparing the 2015 calibration with a later calibration in 2019, in both cases using the same pulse duration and power. The calibration was performed on the data processed as described above, with noise removal and filtering the outer parts of the FM data. The change in Gain (G 0 ) for echo integration and TS measurements, as defined in the equations of Ona et al. (2009), was only 0.6 dB weaker in 2019 compared to 2015. Also, the impedance of the four quadrants of the BITE menu of the EK80 software system was monitored and found equal in 2015 and 2019. The observed variation in gain from 2015 and 2019 was a bit larger than reported by Knudsen (2009) for 18 kHz systems, but within acceptable range. Based on our evaluation, we conclude that the data collected in 2016 is valid with the accuracy of ± 1 dB for both the Nautical Area Scattering Coefficient (NASC) and TS measurements, corresponding to ± 25% in density and backscattering cross section, derived from mean TS.

Acoustic data sets
Firstly, a "Full data set" (Data S1) was built by calculating NASC in m 2 nmi − 2 (Foote and Knudsen, 1994;MacLennan et al., 2002;Simmonds and MacLennan, 2005). To discriminate backscatter from fish or other organisms with similar target strength from background noise and weak echoes from plankton organisms, the minimum echo integrator threshold in terms of the volume backscattering strength (s v ) was gradually increased from the default setting of − 82 dB to about − 75 dB when most echoes from distinct single targets (Fig. S2c) remained but background noise and echoes from smaller organisms disappeared from the echograms.
Secondly, a "Target-tracking data set" (Data S2) was built by using a target-tracking algorithm, containing only the NASC that emerged from single-target tracks. The ES18-11 is a split-beam transducer, divided into four quadrants, so all targets in the beam giving an echo can be positioned in the beam. Following the position of a target over consecutive pings and adjusting for vessel movement allows for calculating the movement (speed and direction) of the targets. Single point targets were detected and connected into individual tracks according to the targettracking method proposed by Handegard et al. (2005). Tracking individual targets allowed us to discern among backscatter from organisms and noise occurring as spikes. To be accepted as valid, a track was required to contain at least eight detections. A track was accepted if <30%, and not more than five consecutive pings, missed detection ( Fig. S2b). Visual inspection showed that LSSS was able to track the specimens with these settings. The tracks were transformed into areas tightly surrounding the tracks in LSSS and the backscattering, in the form of NASC, from these areas was integrated and transferred to data files. Since target tracking is not very efficient in the shallower parts of the water column where the beam is narrow, the backscatter from depths less than ca. 300 m, i.e., received at times less than ca. 0.4 s, will mostly be lost. An investigation of a data subset showed that reducing the number of pings allowed missing from a track from 5 to 1 increased the number of tracks, but NASC did not change much (5%).
The "Full data set" included higher total NASC than the "Targettracking data set" at Stations 1-5 and 9, and a part of Station 6 ( Fig. S3b), which could be attributed to noise that passed the preprocessing filtering and were interpreted as real echoes. The total NASC values in the "Target-tracking data set" may be considered more reliable, since the criteria for accepting an echo as part of a track from a real scattering object were very strict (Fig. S2). We therefore consider the "Target-tracking data set" to contain minimum estimates of the backscattering objects. There is some uncertainty as to whether the "Full data set" can be considered maximum estimates since the noisereduction may also have removed some of the true backscatter along with the noise. We therefore consider the values in the "Full data set" as "high", but not maximum.

Target strength
In the present study neither species nor lengths are known. Consequently, the in situ target strength (TS) distribution estimated from the echosounder output (corrected for position in the beam) was used to calculate areal densities of scatterers. For most fish possessing a swim bladder, the TS varies with frequency. Fishes with swim bladders and similar size and morphology will, however, act as similar targets at 18 and 38 kHz. For another similar-sized gadoid fish Trisopterus esmarkii, the frequency response was found to be relatively flat (r(f) = 1) over a frequency range between 16 and 20 kHz to 38 kHz (Pedersen and Korneliussen, 2009). Consequently, we made no recalculation of the NASC from 16-20 to 38 kHz. To avoid TS distributions being biased towards higher values due to reduced signal-to-noise ratio in the outer parts of the beam, only TS measurements close to the acoustic axis (within 2 • ) were extracted.

Oceanographic data collection
During the "Arctic Ocean 2016" expedition with IB Oden, CTD data were collected with a SeaBird 911 equipped with dual SeaBird temperature (SBE 3) and conductivity (SBE 04 C) sensors. CTD measurements were available for 12 of the 13 acoustic sampling stations (Table 1). At one station, Station 4 at the Lomonosov Ridge (Fig. 1), no CTD was operated (Gårdfeldt and Lindgren, 2016). The CTD measurements for Station 1 were taken nearby the acoustic sampling station, but some weeks later. For the water in the mesopelagic layer (300-600 m), these latter measurements should be comparable for August and September as the deep water is isolated from the influence of the atmosphere and has rather stable conditions as shown by, e.g., year-long instruments deployed north of Svalbard (Renner et al., 2018). For the surface waters on the other hand, measurements a few weeks off in time in the CAO are not comparable as this layer is subject to both seasonal and short-time variability.

Detection of a DSL
Along part of the transect, from Station 1 to Station 7, a distinct DSL at 300-600 m of water depth was clearly documented, while it was more diffusebut presentfrom Station 8 to Station 13 ( Figs. 1 and 2a). Using the target-tracking possibilities of the split-beam transducer, we observed that individual targets were moving in the acoustic beam with speeds of up to 13 cm s − 1 , showing that they were living organisms (Fig. S2b). Schools of organisms did not occur in the data set and single target-tracking was possible for all data.
The mean NASC per station for the 300-600 m layer varied between 0.02 and 1.88 m 2 nmi − 2 in the "Target-tracking data set" and between 0.02 and 7.04 m 2 nmi − 2 in the "Full data set" (Table 2, Fig. S3b). The mean NASC was highest at the two stations closest to the North Pole (Stations 2 and 5), intermediate in the Amerasian Basin (Stations 6 and 7), and lowest near the Gakkel Ridge and in the Nansen Basin (Stations 11-13). Furthermore, the NASC was clearly concentrated between 300 and 600 m of depth at the North Pole, the Lomonosov Ridge, and in the Amerasian Basin, while near the Gakkel Ridge and in the Nansen Basin the NASC was more evenly distributed over the whole water column between 50 and 950 m (Fig. S4). In the Amundsen Basin (Stations 1, 9, 10), the DSL was most pronounced at 300-400 m of depth.
While our NASC data can be trusted for the 300-600 m stratum, they could be underestimated for the upper 300 m where the acoustic backscatter was often obscured by noise. However, time windows when this noise was absent or low suggest that the NASC were generally much lower in the upper 0-300 m layer than in the 300-600 m layer, providing distinctness to the DSL (Fig. 2).

NASC and water temperature
Along the whole transect studied, the central Arctic DSL coincided with the "Atlantic water layer". This deeper layer of warmer, saltier water delivered to the CAO from the Atlantic Ocean is located below the colder upper mixed layer and the halocline (Figs. 3 and S5). The DSL Salinity was lower above the Atlantic water layer (Figs. 3b and S5b), but showed only marginal differences between the stations at 300-600 m of depth (range 34.8-34.9). Also oxygen concentration (6.4-6.8 mL L − 1 ) and water density (1029-1031 kg m − 3 ) varied only marginally among stations in this depth interval (Fig. S5c,d).
The highest NASC was recorded at or below the depth of the temperature maximum in the North Pole area, the Amerasian Basin, and the Amundsen Basin (Figs. 3a and S5). The Nansen Basin differed from the other areas of the CAO studied by higher temperatures and higher NASC both higher up and deeper down in the water column (Figs. 3a and S4). >99% of the backscattering organisms in the DSL had ambient temperature within the range 0.4-1.2 • C, thus covering most of the temperatures within the 300-600 m depth range (0.3-1.6 • C). When comparing the distributions of ambient temperature and the temperatures available to the organisms in the same water layer, it appeared that the organisms were more concentrated at 1.0-1.2 • C than at 0.4-0.7 • C (Fig. 4). The distribution of ambient temperatures is skewed towards the warmer end with >80% of the organisms residing in waters with temperature 0.7-1.2 • C while this range covers only 55% of the temperature range within the DSL.

Target strength (TS) from single targets
Numerous tracks of single targets were detected in the DSL, providing good TS estimates that suggest the occurrence of fish-sized organisms (Data S3), and the acoustic records consisted of pronounced traces resembling those of fish (Figs. 2b and S2c). The TS data in the depth span 0-600 m ranged between − 69.9 and − 38.6 dB (N = 13,667 data points) with a mean value of − 46.8 dB (based on σ bs ), and were completely dominated by data from 300 to 600 m depth. Very few TS data (only 372 data points) were recorded in the upper 300 m, the uppermost one at 166 m. The frequency distribution of the TS values is unimodal and appears like a slightly skewed normal distribution (Fig. 5a); 50% of the distribution represented values between − 48.9 and − 45.8 dB, and 90% of the distribution represented values between − 51.6 and − 44.6 dB. There was a trend of increasing TS values with increasing depth by about 3 dB for the core of the data points (Fig. 5b). This could indicate the occurrence of slightly larger objects at greater depth although this could also be an artefact due to the detection level increasing with increasing distance to the target.

The CAO has a DSL with low abundances of organisms
Our study suggests that a DSL is a common and widespread feature in the mesopelagic zone of the CAO, at least in summer, because we detected this layer along our whole transect in August-September. The abundance of the organisms in the central Arctic DSL is low. The mean NASC (300-600 m) of up to 2 m 2 nmi − 2 with target tracking and up to 7 m 2 nmi − 2 without target tracking, is several orders of magnitude lower than those reported from the northern Atlantic Ocean. For example, the mean NASC (0-800 m) in the open ocean part of the Irminger Sea was found to vary between 319 and 1845 m 2 nmi − 2 (Anderson et al., 2005), and in the Norwegian Sea the mean NASC (0-1000 m) varied between 66 and 365 m 2 nmi − 2 . The Atlantic DSL was found to decrease north-westward from a mean Table 2 The mean area scattering coefficient (NASC) and target strength (TS) measured in the deep scattering layer (300-600 m), and body length, density and biomass calculated from NASC and TS assuming that the acoustic backscatter originates from the small endemic Arctic gadoids Boreogadus saida and/or Arctogadus glacialis. Values are given for both for the "Target-tracking data set" and the "Full data set". Body length was calculated using the empirical relationship between TS and body length for Boreogadus saida: TS = 21.8 log 10 L -72.7 proposed by Mamylov (1988). Stock density and biomass were calculated using the NASC, TS, and the lengthweight relationship for Boreogadus saida (Fig. 6). The density calculations were based on the measured TS values for Stations 1-6, but when <100 TS data points were measured in a specific depth layer at a specific station (Stations 7-13) the mean TS of all data (-46.8, n = 13,667; Fig. 5a NASC (0-900 m) of ca. 360 m 2 nmi − 2 in the eastern Norwegian Sea down to ca. 15 m 2 nmi − 2 in the colder waters north of Iceland (Norheim et al., 2016). Similarly, the mean NASC in the DSL (200-800 m) decreased northward from 215 m 2 nmi − 2 in the Norwegian Sea to 61 m 2 nmi − 2 west of Svalbard (Siegelman-Charbit and Planque, 2016). However, the low NASC recorded in the DSL north of Svalbard in the Sofia Deep (ca. 5 m 2 nmi − 2 ) and at the Yermak Plateau (ca. 7 m 2 nmi − 2 ) are comparable to the highest values recorded in our study in the CAO . In the former study, the DSL was found at ca. 150-720 m with temperature 1.2-1.5 • C, and thistogether with the present studysuggests that the DSL this far north coincides with the Atlantic water that enters the Arctic Ocean via Fram Strait and the Barents Sea (Rudels, 2019). The higher NASC in the warmest part of the vertical water column may be due to temperature preference in combination with light penetration (Schauer et al., 2002;Aksnes et al. 2017).

Possible identity of the organisms in the DSL
Our data provide a first indication of possible mesopelagic fish under the pack-ice cover of the Arctic Ocean. However, our data have two drawbacks with respect to the identity of the targets in the central Arctic DSL-: (1) no samples of organisms were taken, which implies that the acoustic backscatter could not be verified with information on the taxonomic composition of the pelagic fauna, and (2) since the acoustic backscatter was measured over a single narrow frequency band, we could not separate the backscatter from organisms with gas inclusions from organisms without. DSLs at mesopelagic depths usually consist of a mixture of small mesopelagic fishes, crustaceans, cephalopods and gelatinous zooplankton (Irigoien et al., 2014;Klevjer et al., 2016). In agreement with this, the DSL immediate south of the CAO is known to contain the small fishes polar cod (Boreogadus saida), juvenile beaked redfish (Sebastes mentella), spotted barracudina (Arctozenus risso), the lanternfishes (myctophids) Benthosema glaciale and Lampanyctus macdonaldi, the squid Gonatus fabricii, and large zooplankton such as krill, amphipods and gelatinous species Geoffroy et al., 2019). Small fishes have to our knowledge never been caught in zooplankton nets in the CAO, possibly because they tend to avoid nets (Sameoto et al., 2000).
If we assume that the organisms moving in the acoustic beam with speeds of up to 13 cm s − 1 and with TS values between − 69.9 and − 38.6 dB are small fishes with gas bladders, plausible candidates are the two endemic Arctic gadoids, polar cod Boreogadus saida (Mueter et al., 2016) and ice cod Arctogadus glacialis (Jordan et al., 2003), both single species in their genus and hereafter referred to as "Boreogadus" and "Arctogadus". These two species are known to have circumpolar distributions and  (Fig. S3). Plots for the NASC at all 13 stations are shown in Fig. S4. (b) Vertical distribution of salinity. Not used in the construction of these plots are: Station 2 (North Pole, values almost identical to Station 5), Station 4 (no CTD data available), and Stations 8 and 9 (marginal to the transect) (Fig. 1). The resolution in some parts of the transect, in particular in the Amerasian Basin, is low due to the low number of stations. The plot was made in Matlab R2016.
known to occur associated with the pack ice in the CAO (Andriashev et al., 1980;Gradinger and Bluhm, 2004;Melnikov and Chernova, 2013;David et al., 2016;Bluhm et al., 2017). Other circumpolar nekton that has been recorded in the CAO is the up to ca. 30 cm long squid Gonatus fabricii (Nesis, 1971(Nesis, , 2001Young, 1973). In the Norwegian Sea, larger individuals (>20 cm) of this species are seemingly restricted to depths between 400 and 1200 m (Arkhipkin and Bjørke, 1999 and references therein). However, due to lack of gas inclusions, and only a soft chitinous pen embedded in its body tissue, this squid would give much weaker echoes at 18 kHz than a similar-sized Boreogadus or Arctogadus.
The assumption that the DSL would include small gadoids could be considered reasonable because Boreogadus occurs in high abundancesand is a key speciesin the marine pelagic food webs of the Arctic shelf seas surrounding the CAO (Geoffroy et al., 2011;Hop and Gjøsaeter, 2013;Logerwell et al., 2015;Mueter et al., 2016;De Robertis et al., 2017) and Boreogadus juveniles are ubiquitous in the sea-ice habitat throughout the CAO (David et al., 2016 and references therein). Arctogadus has a circumpolar shelf distribution as well, but is apparently less abundant than Boreogadus (Aschan et al., 2009;Mecklenburg et al., 2018). Unique observations of both species were made from Soviet winter stations drifting with the sea ice in the Amerasian Basin near the Mendeleyev Ridge in the 1950s-1970s (Andriashev et al., 1980), where hundreds of individuals of these two gadoids were caught with hooks down to 25 m. In November-December most of the fishes were 10-20 cm long, but in February-March larger individuals (up to 43 cm) identified as Arctogadus were also caught. The specimens of Boreogadus that were caught rarely exceeded 15-17 cm.
The overall depth distribution of Arctogadus in the European Arctic ranges from 155 to 741 m, with most observations at depths of 300-400 m (Aschan et al., 2009). This would agree with the vertical distribution of the DSL in the CAO we discovered. A tentative suggestion that needs validation in future studies is that Arctogadus is a true pelagic species living at mesopelagic depth in the Atlantic layer of the CAO. The early studies from Soviet ice-drift stations showed that Arctogadus can occur associated with the sea ice in the CAO in winter (Andriashev et al., 1980), where they can feed on ice amphipods and juvenile Boreogadus. When the light returns in spring, Arctogadus would then descend to deeper water.
For Boreogadus a "sea-ice drift" hypothesis has been suggested for its transport from coastal spawning areas to the Amerasian and Eurasian Basins (Melnikov and Chernova, 2013;David et al., 2016). Young-ofthe-year from spawning areas in the Russian shelf seas would, according to this hypothesis, become associated with sea ice when it forms in autumn and then passively be transported to the CAO with the transpolar ice drift (Spall, 2019). Juvenile Boreogadus feed on ice-associated invertebrates, mainly amphipods, and can hide from predators in crevices in the ice (David et al., 2016). Based on our discovery of the central Arctic DSL, we could extend the hypothesis to include the possible descent of Boreogadus to the mesopelagic zone of the CAO when they grow older and forage on the crustaceans that dominate the zooplankton Fig. 4. Distributions of ambient temperature for the organisms and temperatures available to the organisms in the 300-600 m depth layer from 12 stations out of 13 (no temperature data were available for Station 4; Table S1). The ambient temperature for the organisms in the 300-600 m depth layer was calculated for each station by weighting the observed temperatures with area scattering coefficients (NASC) from the "Target-tracking data set". The calculations were conducted for 10-m depth intervals (corresponding to the resolution of the NASC) and binned into 0.1 • C temperature intervals. The temperatures available to the organisms was calculated using the same method but without weighting. The two distributions would be the same if the organisms would be distributed evenly in the 300-600 m depth layer.  1-13 (Fig. 1). Only TS data from within 2 • of the beam axis are included. (a) Histogram with mean and quartiles 1 and 3 (Q1 and Q3) for all 13,667 TS data points. (b) The relationship between all TS values and water depth, including separate mean TS values for 0-300 m and for 300-600 m, respectively. communities in the mesopelagic zone of the CAO (Kosobokova et al., 2011), as a coupling between shallow and mesopelagic ocean habitats. Data from the Beaufort Sea show Boreogadus to be vertically segregated by size in all months, with juveniles in the epipelagic (<100 m deep) layer and older Boreogadus in the Atlantic layer along the slopes of the Beaufort Sea at 200-500 m depth (Benoit et al., 2014;Geoffroy et al., 2016). It has been suggested that early hatching (already in January) enables the juvenile fish to reach a minimum size and capacity to avoid predation before they join their cannibalistic congeners in the mesopelagic Atlantic layer as larger individuals (Bouchard and Fortier, 2011). This strategy could be the same in the CAO with juveniles associated with sea ice (Gradinger and Bluhm, 2004;Melnikov and Chernova, 2013;David et al., 2016) and adults deeper down.
However, we must consider that (part of) the acoustic backscatter in our data could originate from pneumatophores (gas-filled floats) of physonect siphonophores (colonial hydrozoans). The pneumatophore is situated at the top of the colony to provide buoyancy and may be detected acoustically depending on frequency, depth, and size of the gas inclusion. Furthermore, physonect siphonophores are equipped with nectophores (swimming bells) that allow excellent motility (Costello et al., 2015). Many physonect siphonophores are known to inhabit the mesopelagic domain globally, and it has been argued that their contribution to backscatter must be determined to obtain better estimates of mesopelagic fish in the world's oceans (Proud et al., 2018). The physonect siphonopore Marrus orthocanna, an orange-coloured Arctic species with colony length of up to 2-3 m (R. R. Hopcroft, pers. comm.), is present in the CAO and the deep Beaufort Sea (Raskoff et al., 2010;Kosobokova et al., 2011). Another Arctic physonect siphonophore, Rudjakovia plicata, has been recorded in the CAO (Margulis, 1982;Kosobokova et al., 2011). R. plicata is not very well known (Mapstone, 2014), but is probably more abundant than initially thought (R.R. Hopcroft, pers. comm.; Raskoff et al. 2010;Kosobokova et al., 2011). Other zooplankton organisms, like chaetognaths, amphipods, pteropods, ctenophores and calycophoran siphonophores (without pneumatophore), are very weak scatterers with the exception of the pteropods, but these are mainly epipelagic (Kobayashi, 1974) and are seldom found at DSL depths. The echo returns at 18 kHz of the copepod Calanus hyperboreus and other similar-sized crustaceans that dominate the mesozooplankton biomass in the CAO (Kosobokova et al., 2011), are several orders of magnitude below what was observed in this study, and will most certainly not be detected at this frequency.

Maximum possible mesopelagic fish biomass
Catch data, including species identification and body size, are necessary for estimating fish density and biomass from NASC data. Since traditional trawling for fish sampling under the pack ice is impossible, it will probably take a long time before such data can be collected in the North-Pole area. By assuming that the individual tracks in the central Arctic DSL represent the endemic Arctic gadoids Boreogadus and/or Arctogadus, a rough estimate of maximum possible mesopelagic fish stock can be calculated from our NASC data. However, the true density and biomass of fish is lower and cannot be quantified from this dataset due to backscatter originating from pneumatophores of physonect siphonophores (Raskoff et al., 2010;Kosobokova et al., 2011).
Our unimodal TS frequency distribution within a narrow TS range indicates a relatively homogenous composition of the 13,667 individual targets. By applying the TS-length relationship for Boreogadus proposed by Mamylov (1988), this results in a hypothetical average body length of 15.4 cm. In the North-Pole area (Stations 1-5), the average body length would be ca. 15-17 cm, and in the Canada Basin (Station 6) ca. 10.5 cm ( Table 2). These lengths agree with literature data on the known lengths of Boreogadus and Arctogadus in the Arctic shelf seas surrounding the CAO (Fig. 6). The length-age models are very similar for Boreogadus and Arctogadus up to the age of about 2 years, but after that Arctogadus grows faster and can become twice as old and twice as long as Boreogadus (Christiansen et al., 2012;Nahrgang et al., 2014). If the acoustic backscatter in our data set would originate from Boreogadus, 80% of the fishes would be 1.8-4.7 years old and weigh 5-47 g, and if it would originate from Arctogadus, 80% of the fishes would be 1.5-3.6 years old and weigh 5-54 g (Table 3). Thus, the central Arctic DSL would largely be inhabited by a mixture of late juveniles and adults. However, it should be kept in mind that Boreogadus and Arctogadus are not the only fish candidates, e.g., the occurrence of myctophids  cannot be excluded.
Continuing with the assumption that the individual tracks in the DSL of the CAO represent Boreogadus, Arctogadus, or both, the maximum possible ranges of fish density and biomass in the central Arctic DSL would be 22-2200 individuals km − 2 and 0.5-50 kg km − 2 (Table 2). This is on average 720 individuals km − 2 and 14.5 kg km − 2 based on the "Target-tracking data set" or 1900 individuals km − 2 and 42 kg km − 2 based on the "Full data set" (n = 13 sampling stations). From these numbers it can be concluded thatif we have a mesopelagic stock of fish with swim bladders in the CAOit is widespread but very small. For comparison, the total stock biomass of Boreogadus in the 1.4 million km 2 Barents Sea was on average 777,000 tons in the period 1986-2017 (555 kg km − 2 ) with a maximum of 1.94 million tons in 2006 (1,386 kg km − 2 ) (van der Meeren and Prozorkevich, 2019). We emphasize that there is considerable uncertainty in our estimate for the CAO due to limited spatial coverage of our data with only one transect, only one summer month (15 August-15 September), the potential masking of backscatter by noise, and the possible contribution of siphonophores to the acoustic backscatter. Our density and biomass estimates are in the same order of magnitude as previous estimates of juvenile ice-associated Boreogadus in the sea-ice habitat of 5,000 individuals km − 2 and 19.3 kg km − 2 in the Eurasian Basin of the CAO (David et al., 2016), although these calculations were also based on a very limited number of observations in summer.
Our estimated maximum potential biomass of fish in the central Arctic DSL is in broad agreement with the low level of primary production in the CAO. Primary production in the ice-covered waters of the CAO is low, roughly on the order of 10 g C m − 2 year − 1 (Gosselin et al., 1997;Olli et al., 2007), which is equivalent to about 100 g wet weight biomass m − 2 year − 1 . With 10% transfer efficiency between successive trophic levels, this would give an annual production of 1 g m − 2 year − 1 (=1000 kg km − 2 year − 1 ) at the tertiary trophic level of planktivorous fish such as Boreogadus. This is an order of magnitude higher than our estimate of biomass of fish in the DSL. However, there are several explanations for this discrepancy that probably act in concert: a sizable fraction of the primary production in the CAO goes through the microbial loop, the Atlantic inflow to the CAO transports food items from the Atlantic Ocean to the CAO, and Arctogadus is partly at a higher trophic level than the tertiary, being a piscivore consuming juvenile Boreogadus (Andriashev et al., 1980). The reason why biomass was highest in the North-Pole area cannot be attributed to primary production; perhaps this is related to the flow of Atlantic water following the Lomonosov Ridge (Rudels 2015).
Indirect support for the occurrence (but low abundance) of a fish stock under the pack-ice is the occurrence (but low abundance) of fish predators in the CAO. The ringed seal Pusa hispida is regularly observed from icebreakers, even in the North-Pole area, as well as their predator the hypercarnivorous polar bear Ursus maritimus (Reeves, 1998;van Meurs and Splettstoesser, 2003). The diving behaviour of ringed seals is consistent with the depth of the central Arctic DSL, and avoidance of visual predation by mammals may be a possible reason for the DSL to remain at depths of 300-600 m. It has been shown that ringed seals in the Canada Basin can make > 20 dives down to 250-300 m within a few hours (Von Duyke et al., 2019) and they dive on average deeper in the Canada Basin than in the Bering and Chukchi Seas . Narwhals Monodon monoceros and beluga whales Delphinapterus leucas are known to feed on Boreogadus and Arctogadus (Bluhm and Gradinger, 2008). High numbers (about 1,000 individuals) of narwhals have been recorded deep into the pack ice in an aerial survey in the western Nansen Basin (Vacquié-Garcia et al., 2017), and in the eastern Nansen Basin this species has been observed north of 85 • N (Belikov and Boltunov, 2002). Beluga whales have been recorded by satellite-tracking to move north to beyond 80 • N in the Canada Basin (Hauser et al., 2014;.

The possible involvement of siphonophore pneumatophores
It cannot be ruled out that (at least part of) the NASC in our study originated from pneumatophores of physonect siphonophores. The TS of gas bubbles varies with bubble size and depth. Theoretical models show that the TS at 18 kHz for 1.8-3.1 mm diameter gas bubbles, residing at depths between 250 and 700 m, are mainly in the range − 60 to − 42 dB (Movie S1, Fig. S6), with peak values (resonance) in the range − 50 to − 40 dB. This is similar to the TS values we recorded in the DSL of the CAO, and siphonophores are therefore a group of organisms that may have contributed to the backscatter in our NASC values. Our conclusion is that with the current theoretical models we cannot separate the backscatter from a 2-4 mm pneumatophore with that from a 2-4 mm gas-filled swim bladder of a fish at 18 kHz.
Unfortunately, there is considerable uncertainty about the size of the  Tables S2 and S3. The data consist of mean length per age group for adult fish as given in tables and figures in the publications, but for juveniles (<1 year) the age was calculated from published polynomial regressions for each cm fish length in the length interval for which the equation was valid. Since the data were presented in different ways in the publications, the circles represent either primary data, population means or calculated values from published regressions for each cm fish length in the length interval for which the equation was valid. Thus, the R 2 values must be interpreted with care because they are not based on independent measurements. Within publications, data from the same area were treated as one population, and data from different publications were considered as different populations. For Boreogadus, fork length (FL) was converted to total length (TL) with the function TL = (1.03724*FL) − 0.0113, valid for the interval 4-30 cm, based on Nahrgang et al. (2014). Standard length (SL) was converted to TL with the function TL = (1.08849*SL) + 0.2736, valid for the interval 5-12 cm, based on David et al. (2016) and further validated up to 17 cm based on SL and TL length measurements in Melnikov and Chernova (2013). For Arctogadus, standard length (SL) was converted to total length (TL) with the function TL = (1.1033*SL) − 2.225, valid for the interval 17-41 cm, based on Andriashev et al. (1980) and Borkin and Mel'yantsev (1984).
gas inclusions in physonect siphonophores, and information on how the shape and size of their gas inclusions would vary with depth is lacking as well. For Nanomia bijuga with colony size up to 75 cm, average pneumatophore size of 3.27 × 1.16 mm (ca. 1.64 mm ESD = equivalent spherical diameter) and average gas inclusion size of 1.49 × 0.97 mm (ca. 1.12 mm ESD; n = 50) were measured after three weeks fixation in 10% formalin (Barham, 1963). During a siphonophore bloom in a fjord system in northern Norway it was found that high abundance of Nanomia cara was reflected in the acoustic backscatter at 18, 38 and 120 kHz in the water column slightly deeper than 300 m (Knutsen et al., 2018). Associated with the latter observations, a small number of pneumatophore diameters were measured, and the largest pneumatophore had an ESD of 1.12 mm, measured in formalin-conserved material about three weeks after collection (T. Knutsen, pers. comm.).
Of the two physonect siphonophore species that have been recorded in the CAO (Raskoff et al., 2010;Kosobokova et al., 2011), the colonies of Marrus orthocanna are larger (up to 2-3 m) than those of the two Nanomia species mentioned above, while the colony of Rudjakovia plicata (Margulis, 1982) is only up to 10-15 cm long (R.R. Hopcroft pers. comm.). The largest Marrus orthocanna colonies may have an estimated pneumatophore size of up to 5-10 mm in diameter (R.R. Hopcroft, pers. comm.). However, this and other species' pneumatophore sizes needs verification from in situ measurements. For the related species Marrus claudanielis two preserved pneumatophores measured 5.0-5.2 × 1.6-1.7 mm (ca. 2.46 mm ESD; Dunn et al., 2005). Altogether, it is evident that physonect siphonophores inhabit waters that overlap with the DSL we discovered in the CAO, and that presently it cannot be excluded that these siphonophores could contribute with acoustic backscatter of similar TS as small mesopelagic fish.

Conclusions
Despite great reduction of the sea-ice cover in the CAO, the area is still inaccessible for fishing vessels, preventing both commercial harvesting of fish and standard data collection for assessment of its fish stocks. Although we were not able to establish with certainty that the central Arctic DSL contains mesopelagic fishes, we discovered a clear DSL that is a typical habitat for zooplankton and small fishes worldwide. In the low-productive CAO ecosystem, the hypothetical fish in the DSL are likely to have a key role in the food web, providing a highenergy lipid-rich resource to higher trophic levels, such as narwhals, seals and polar bears, allowing mammals to survive even in the North-Pole area. An essential research priority is to sample organisms from the central Arctic DSL to establish their identity. This will not be an easy task since we found that the organisms are extremely sparse, relatively small, and possibly too fast to be caught in nets. From our study we can conclude that if the central Arctic DSL contains fish, their biomass is currently too low for any sustainable fishery. Northern fisheries are much more profitable south of the CAO in the Arctic shelf seas and sub-Arctic seas that are far more productive.

Author contributions
P.S.L. and H.G. conceived the idea. P.S.L. compiled the manuscript; all co-authors contributed to and reviewed the manuscript. Acoustic and CTD data collection was carried out by C.S., L.M., M.J., and K.G. The acoustic data sets were prepared by H.G., R.K., E.O., and T.K. The physiographic classification was made by M.J. Data analyses were carried out by R.I. and P.S.L.

Declaration of Competing Interest
The authors declare that they have no known competing financial interests or personal relationships that could have appeared to influence the work reported in this paper.

Table 3
Calculations of fish size and age if we assume that the organisms we recorded at TS between − 69.9 and − 38.6 dB, based on 13,667 TS measurements (Fig. 5a), would be polar cod Boreogadus saida and/or ice cod Arctogadus glacialis. Body length was calculated using the empirical relationship between TS and body length for Boreogadus saida: TS = 21.8 log 10 L -72.7 proposed by Mamylov (1988). For age and weight, the length-weight relationships for Boreogadus saida and Arctogadus glacialis in Fig. 6 were used.