Some like it cold: Temperature‐dependent habitat selection by narwhals

Abstract The narwhal (Monodon monoceros) is a high‐Arctic species inhabiting areas that are experiencing increases in sea temperatures, which together with reduction in sea ice are expected to modify the niches of several Arctic marine apex predators. The Scoresby Sound fjord complex in East Greenland is the summer residence for an isolated population of narwhals. The movements of 12 whales instrumented with Fastloc‐GPS transmitters were studied during summer in Scoresby Sound and at their offshore winter ground in 2017–2019. An additional four narwhals provided detailed hydrographic profiles on both summer and winter grounds. Data on diving of the whales were obtained from 20 satellite‐linked time‐depth recorders and 16 Acousonde™ recorders that also provided information on the temperature and depth of buzzes. In summer, the foraging whales targeted depths between 300 and 850 m where the preferred areas visited by the whales had temperatures ranging between 0.6 and 1.5°C (mean = 1.1°C, SD = 0.22). The highest probability of buzzing activity during summer was at a temperature of 0.7°C and at depths > 300 m. The whales targeted similar depths at their offshore winter ground where the temperature was slightly higher (range: 0.7–1.7°C, mean = 1.3°C, SD = 0.29). Both the probability of buzzing events and the spatial distribution of the whales in both seasons demonstrated a preferential selection of cold water. This was particularly pronounced in winter where cold coastal water was selected and warm Atlantic water farther offshore was avoided. It is unknown if the small temperature niche of whales while feeding is because prey is concentrated at these temperature gradients and is easier to capture at low temperatures, or because there are limitations in the thermoregulation of the whales. In any case, the small niche requirements together with their strong site fidelity emphasize the sensitivity of narwhals to changes in the thermal characteristics of their habitats.


| INTRODUC TI ON
Over the past 30 years, sea ice in the Arctic has declined by 3%-4% per decade, making the Arctic the area experiencing the most rapid degradation due to climate change (Parkinson & Cavalieri, 2002Serreze & Stroeve, 2015;Stroeve, Schroder, Tsamados, & Feltham, 2018). At the same time, the ocean has warmed on a global scale by an average of 0.11°C per decade from 1971 to 2010 (Pachauri, Qin, & Stocker, 2013). Increasing sea temperatures and reduction in sea ice are expected to modify the niches of Arctic marine apex predators although robust niche definitions are missing for most species (Chambault et al., 2018;Laidre et al., 2008).

While increasing sea temperatures may result in range expansions
for species that travel between temperate and polar waters, the range of Arctic species is conversely expected to decline leading to major changes in their habitat use (Simmonds & Isaac, 2007;Tynan & DeMaster, 1997).
The narwhal is a high-Arctic medium-sized cetacean that makes annual migrations from coastal areas in front of glaciers to offshore deep-water areas in Baffin Bay, the North Water, and the Greenland Sea Kenyon, Yurkowski, Orr, Barber, & Ferguson, 2018;Watt, Orr, & Ferguson, 2017). The various populations or stocks of narwhals are usually delineated based on their summer grounds (Heide-Jørgensen, Richard, Dietz, & Laidre, 2013;Watt et al., 2019). In East Greenland, narwhals in the Scoresby Sound fjord system are considered isolated from neighboring populations that summer north or south of Scoresby Sound . Narwhals routinely travel through water masses that are critical for the circulation in the North Atlantic. Glacial melt at the narwhals' summer grounds in Greenland increases the inflow of freshwater which then pushes the border of warm water masses south in the North Atlantic (Hunt et al., 2015). Large-scale freshwater anomalies have been observed in past decades and are predicted to occur with increasing frequency with global warming (Rahmstorf et al., 2015). This will maintain the cold-water influx to narwhal habitats but at the same time there is a general warming of the sea surface throughout the North Atlantic (Alexander et al., 2018) and at several of the narwhal summer grounds (Chambault et al., in review). Narwhals appear to avoid areas with inflow of warm Atlantic water (>4°C) in the Arctic, and this likely determines both their migrations and habitat selection (Chambault et al., in review).
Modeling of narwhal habitat selection has traditionally been conducted with correlative models including data derived from satellite-tracked whales, in which positions and diving depths are related to physical parameters (sea ice, bathymetry, etc.) or potential prey concentrations (Kenyon et al., 2018;Laidre, Heide-Jørgensen, Jørgensen, & Treble, 2004;Watt et al., 2017).
Based on this, wide areas of the Arctic are available as potential narwhal habitats, but are currently devoid of the whales. This discrepancy may be due in part to the limited understanding of the biological importance of potential foraging habitats, thus leading to erroneous assumptions about niche selection by narwhals. Also, the extrapolations of correlative habitat models to future climate scenarios are exceptionally risky without empirical data on niche selection. A more mechanistic niche description comprising environmental and biotic factors that affect the fitness of the whales is needed to understand the interactions between narwhals and their rapidly changing environment (e.g., Kearney, 2006). This is clearly a difficult task for cetaceans in general, and for narwhals in particular, as they prefer to live in logistically extremely challenging environments. Energetics-based niche modeling requires more data on metabolic and nutritional requirements of narwhals than are currently available. One improvement is to include the effects of changes in sea temperatures in niche modeling for narwhals. This is especially important for this Arctic marine mammal as changing sea temperatures may affect both the narwhal's homeothermy and its chances for encountering and capturing prey. It has also been shown for other cetaceans that sea temperatures play an important role in niche definition including the foraging success, survival, and distribution of the whales (Chambault et al., 2018;Derville et al., 2019;Lambert et al., 2014;Owen, Jenner, Micheline-Nicole, McCauley, & Andrews, 2018;Owen et al., 2018;Whitehead, McGill, & Worm, 2008;Wild et al., 2019).
Current models predict that sea temperatures will undergo marked changes in the Arctic both with reduced albedo from loss of sea ice and changing patterns of ocean circulation. As an endemic polar species with a limited distribution in the Atlantic part of the Arctic, the narwhal may be particularly susceptible to these changes. Furthermore, narwhals are considered highly sensitive to climate change due to lack of plasticity in movement patterns, low population size, low diet variability, and low genetic diversity Louis et al., 2020;Westbury, Petersen, Garde, Heide-Jørgensen, & Lorenzen, 2019).
In this study, we test the narwhal's vulnerability to changing climate conditions, by examining the narwhal's dependence on a narrow temperature niche in East Greenland. Four independent data sets obtained by use of satellite telemetry and biologging methods were used to illustrate the habitat usage/selection by narwhals.
Data from CTD transmitters were used to characterize the habitat's oceanography, Fastloc-GPS transmitters showed the spatial dispersal of the whales, acoustic tags provided data on depth and temperature of buzzing activities, time-depth recorders indicated the preferred diving depths while satellite-linked time-depth recorders provided concatenated data on diving activity.

K E Y W O R D S
buzzing, deep diving, East Greenland, high Arctic, oceanography, satellite tracking, thermal homeostasis 2 | MATERIAL S AND ME THODS

| Study area
The Scoresby Sound fjord system (hereafter Scoresby Sound) in East Greenland is the summer residence for an isolated population of narwhals. The fjord system is about 350 km long with many side branches of smaller fjords around one large island: Milne Land.
The detailed bathymetry of the fjord system is not well known but most of the inner part of the fjords have depths that range down to 1,000 m or deeper (Digby, 1953;Ryder, 1895). Extensive shallow areas with slightly sloping water depths are found in the northwestern part along Jameson Land. There are 12 active glaciers that feed into the fjord system supplemented by an inflow from the cold East Greenland current in the northern part of the entrance to the fjord system (Digby, 1953). The main current out of Scoresby Sound is in the southern part of the entrance. Sea ice forms in October in the inner parts of the fjord system, and by December, the entire fjord is ice-covered. The sea ice persists through June; however, an open water polynya is present throughout the winter at the opening of the Scoresby Sound fjord system.

| Study design
Four different approaches were used to illustrate the habitat usage of narwhals. Position data from 2 years of satellite tracking (August 2017-March 2019) were combined with oceanographic data to spatially identify the critical temperature ranges used by narwhals. The oceanographic data were collected independently by narwhals instrumented with CTD transmitters in the same 2 years. Archival tags recording acoustic activity by the whales in addition to depth and temperature provided information on the depth and temperature of potential foraging events in 2013-2018. Archival time-depth recorders showed the preferred diving depths, and satellite-linked timedepth recorders deployed in 2010-2013 provided concatenated data on diving activity (Table 1). The summer period was defined as August through September and the winter as December through March. The migration corridors used in October-November are probably not as important as the areas where the whales are more stationary and were therefore not included in the analysis.  Table 1) with three 8-mm delrin nylon pins secured with nylon washers and bolts on each end, following instrumentation techniques used in similar studies in Canada and West Greenland (Dietz et al., 2008;Heide-Jørgensen et al., 2003). The transmitters were programmed to collect an unrestricted number of FastLoc snapshots through August and September. In 2017, the transmitters were restricted to provide data only every seventh day in September. The Fastloc snapshots were transmitted to and relayed through the Argos Location and Data Collection System (www.argos -system. cls.fr). Postprocessing of GPS positions was conducted through the Wildlife Computers web portal. Fastloc-GPS is a GPS positioning system with the ability of faster acquisition of animal positions than traditional GPS (Bryant, 2007;Tomkiewicz, Fuller, Kie, & Bates, 2010).

| Oceanographic data
In both 2017 and 2018, two narwhals were instrumented with satellite transmitters that in addition to depth also recorded and transmitted data on in situ water temperature and salinity (Wildlife Computers Scout-CTD-370D, 12 × 6 × 3.5 cm, 316 g).
The CTD (conductivity-temperature-depth) tags were electrodebased, were powered by four AA lithium batteries, and had a temperature range of −3 to 40°C and a salinity range of 0-50 PSU.
The resolution was 0.001 for salinity and temperature, and 0.5 m for depth. The minimum requirement for a dive was 50 m and the minimum interval between dives was set to 15 min. For each qualifying dive, the deepest point was detected, after which sampling of the various parameters was continued at 1 Hz until reaching the surface. The first dive set the baseline or the minimum depth for a dive to be recorded and the next dive had to be ten percent deeper than the baseline dive to overwrite the data. This continued until the end of the summary period, when the CTD data and the closest position data were processed into Argos messages for transmission. When the tag was at the surface, the measured environmental data were transmitted through Argos Data Collection and Location System. The tags were programmed to capture the deepest profile in every 12-hr period, that is, a total of two profiles per day, and were set to transmit the profile repeatedly 12 times to increase the chance that a given profile would be received by an Argos satellite during the following 12-hr period. The tags were mounted in a similar way as the Fastloc-GPS transmitters mentioned above. Data from the CTD casts were collected at standard

| Acoustic recordings
Sixteen narwhals were instrumented with Acousonde™ acoustic tags (Table 1, www.acous onde.com), whose float had been modified to accommodate an Argos transmitter (Wildlife Computers SPOT5) in addition to a VHF transmitter (ATS Telemetry). The Acousonde recorders were attached to the skin with suction cups, on the rear half of the animal, to the side of the dorsal ridge, and released from the whales after four to ten days (see Blackwell et al., 2018). They were subsequently located and picked up at sea with Argos and VHF transmitters. A custom-written buzz detector (MATLAB, The MathWorks, Inc.) was used to identify buzzes in the records and all potential buzzes were verified manually in 13 of the Acousonde recordings with sufficient quality of the acoustic data. As was done for the GPS positions, the first 24 hr of acoustic data were excluded. In addition to acoustic sampling the Acousonde also provided data on depth (precision 0.5 m) and temperature (precision of 0.01°C) every one second. Since the temperature sensor on the Acousonde is embedded in epoxy, there is a delay in the temperature readings relative to the depth. It is therefore necessary to correct the measured temperatures to obtain the actual temperature values and a detailed description of how the corrected temperatures were obtained is provided in Appendix 1. The predicted occurrence of buzzes in relation to depth and temperature was modeled with generalized linear mixed-effects models with a Poisson response distribution with log-link (function glmer, package "lme4" (Bates, Maechler, Bolker, & Walker, 2015) in R 3.5.). The best model based on the lowest AIC score included temperature (nonlinearly, as second-order polynomial variable), depth (nonlinearly with natural splines with three degrees of freedom), and sex as fixed effects. Individual whales were included as a random effect. In order to distinguish between different layers of water column stratification and to isolate the layer known to have the highest buzzing activity (Blackwell et al., 2018), three depth strata were used for the modeling (0-100 m, 100-300 m, >300 m). This allowed the model to predict the effect of temperature on the buzzing rate separately for each depth stratum. The effect of sex on the temperature preference for buzzing was tested by comparing AIC and predicted values for models with and without sex. Note: Acousonde (Acou) tags provided data on acoustics, depth, and temperature; CTD tags provided data on position and oceanographic profiles (temperature and salinity at depth); FastLoc tags provided high-resolution (GPS-based) positions; SLTDR tag provided coarse data on diving activity in bins sampled in 6-hr periods, except for IDNO 3965 that provided a continuous record of dives at a resolution of 1 m for every second; and the SMRU CTD tag provided data on target depths of dives. Sample size is given in seconds (s)  but only five of the whales provided data during the winter period.

| Summer
The Buzzing, and therefore presumably foraging activity by the whales (e.g., Miller, Johnson, & Tyack, 2004), was highest at depths below 300 m with the highest predicted buzzing rate at 0.7°C in this depth category ( Figure 6, Table 2). This is slightly lower than the aver- of 12 whales that preferred temperatures lower than the average temperature. Less than 2% of the buzzes took place at temperatures above 2°C and less than 8.5% of the buzzes were above the thermocline at ~100 m depth. Including sex or not in the model did not change the estimates of the temperature preference for buzzing activity; however, the male buzzing rate was significantly lower than the females.

| Winter
The resolution of the target depths for the winter diving activity was less precise than for the summer diving activity, due to the use of long-term tags (SLTDR tags) that provided binned data on diving activity rather than archived data with all depth readings (Acousonde

F I G U R E 2
Cumulative time spent at depth in Scoresby Sound, presented as percentages for one narwhal instrumented with a SLTDR (left) with an exceptionally long record (#3963, 83 ds) and the mean of 16 narwhals instrumented with Acousondes (middle, 4-8 ds). Time spent at 0-10 m (not shown) was 44% for #3963 and 47% (SD = 9.62) for the Acousonde data. One SD is shown for the mean of the Acousonde data. Note the increased attendance to depths between 300 and 500 m. The right figure shows the percentages of the target depths (maximum depth reached during 915 dives) for one whale instrumented with a SMRU transmitter. For this, whale about 43% of the dives targeted depths < 10 m The overall selection of the water column properties at the target depths of narwhal diving in both summer and winter shows a narrow range of both temperature (0.5-1.5°C) and salinity (34.72-34.86PSU) values compared to the available habitats ( Figure 4). In winter, there appeared to be a secondary thermocline below 400 m but no obvious change in salinity.

| D ISCUSS I ON
Although Scoresby Sound is the world's largest fjord system (~14,000 km 2 ), very little information is available on the oceanographic conditions in the area and few in situ CTD profiles have been collected and all are outdated (Digby, 1953;Koch, 1945;Ryder, 1895;Ussing, 1934). Even though the sampling was coarse both spatially (grid cells) and vertically (depth intervals), the four nar- There were few oceanographic profiles outside the continental shelf, and it is well known that water from the warm Atlantic current, that is running north between Iceland and Greenland, dominates east of the shelf area, where narwhals are not known to occur (Brakstad, Våge, Håvik, & Moore, 2019;Dietz, Heide-Jørgensen, Glahder, & Born, 1994;Jochumsen et al., 2017). Although the warmest temperatures both on the summering ground inside Scoresby Sound and on the offshore wintering ground were detected at 400-500 m depth, followed by 300-400 m and 500-600 m depths, the temperatures were still consistently below 2°C. This was clearly also the upper limit of the temperature range for the deep-water (>300 m) buzzing activity. The main diving preference of the whales targeted areas both vertically and horizontally that met these temperature requirements.
During summer, the whales preferred areas where the temperature at the target depths was slightly colder than the water masses that were generally available in Scoresby Sound. The same was the case for the buzzing activity but no statistically significant difference in the selected and available temperatures could be detected. It may, however, be argued that the homogenous conditions of the hydrography in Scoresby Sound makes it difficult to select temperature ranges that varied much from the available temperatures. In winter, the situation was different as even though the whales selected temperatures slightly above the available temperature, the cold coastal water of polar origin was selected disproportionally relative to the large areas of warm Atlantic water further offshore.
The spatial distribution of the whales under extrinsic conditions provides a two-dimensional description of their habitat selection (Kenyon et al., 2018;Laidre, Heide-Jørgensen, Logsdon, et al., 2004).
The buzz activity in relation to depth and temperature provides a mechanistic approach to a vertical niche definition. The buzz activity in summer and diving activity in both summer and winter was concentrated at depths where the temperatures were below 2°C. The winter diving activity was more focused on dives to slightly deeper depths than in summer with a focus on depths between 300 and 600 m. The depth range selected by the whales targeted parts of the water column with temperatures below 2°C and confirms the niche selection of the summering area.
Temperature-dependent targeting of dive and buzz activity seems critical for narwhals when locating prey concentrations at depths, but the drivers behind this selection remain uncertain.
Narwhals feed on squid and fish and some of the target prey may be more abundant at certain depths and thermal gradients, or at the observed pycnocline around ~400 m, but the prey could also be easier to catch at lower temperatures for a relatively slow-moving F I G U R E 6 Predicted buzz rates (buzz/ min) (22,519 buzzes) in relation to depth and temperature from generalized mixedeffects model. The average temperature profile from the CTD casts is shown as the gray line. The black dotted lines mark the population estimates for the temperature with the highest predicted buzzing rates in each of the three depth categories. The estimates were 0-100 m: 1.9°C, 100-300 m −0.1°C, and >300 m: 0.7°C. The black triangles mark the individual specific estimates for the highest predicted buzzing rates (see Table 2 for details). Seven out of 12 whales selected temperatures that were lower than the average temperature. Extreme values below −2°C were omitted (<2% of the values). The number of buzzes for the three intervals is 0-100 m: 1.5 buzzes/1000 s, 100-300 m: 24.5 buzzes/1000 s, and >300 m: 47 buzzes/1,000 s. Extreme values below −2°C were omitted (<2% of the values) whale (Williams, Noren, & Glenn, 2010). The low temperatures may also serve to ensure thermal homeostasis. Narwhals appear to have a limited ability of transferring excess heat to the environment, in part due to the lack of a prominent dorsal fin to serve as a thermal window (Doidge, 1990). This has also been suggested for another whale endemic to the Arctic, the bowhead whale, Balaena mysticetus (Chambault et al., 2018). The significance of these findings for thermal balance in narwhals can only be speculated upon as studies of thermoregulation in cetaceans are still in their infancy compared to pinnipeds (Nieaber, Thomton, Horning, Polasek, & Mellish, 2010; TA B L E 2 List of whales instrumented with Acousonde recorders (see Table 1 for details on individuals) with the preferred depth, temperature, and predicted maximum buzzing per rate for each of three depth categories Noren, Williams, Berry, & Butler, 1999). In view of the strict thermal preference of narwhals (Figures 4 and 5), such studies are especially needed for the Arctic cetaceans that will be exposed to the most rapid increases in ocean temperatures (Alexander et al., 2018). ing with diving and foraging behavior of a high-Arctic cetacean, the narwhal, to estimate its habitat selection. Some cetaceans make seasonal migrations that cover very different habitats, for example, breeding at low latitudes and summering at high latitudes. Narwhals also migrate between summer and wintering grounds but they essentially remain within the same habitat and avoid excursions into water masses with temperatures higher than 2°C at the depths where feeding presumably occurs. The narrow temperature habitat together with the strong site fidelity confirms the sensitivity of narwhals  and demonstrates their limited ability to adapt to environmental changes. A mechanistic prediction of sea temperature changes in the high Arctic suggests that several of the present narwhal habitats are changing and that the optimal habitats for narwhal will be located further north by the end of this century (Alexander et al., 2018, Chambault et al., in review;Louis et al., 2020).

CO N FLI C T O F I NTE R E S T
There are no competing interests.

FRO M ACO US O N D E R ECO R D E R S
A laboratory experiment was carried out with the seven Acousondes used on narwhals with concurrent actual temperature measurements by a CTD. The CTD and the Acousondes were first submersed in water at around 1°C, after 45 min they were transferred to water of around 15°C, and after another 45 min they were transferred to water of around 0°C. The data are illustrated in the following figure. The gray dots are CTD measurements; the 7 black lines are the measured temperatures from the Acousondes.
It is clear how the Acousonde values decay exponentially toward the true temperature and that it takes about 15 min for the Acousondes to stabilize at the new temperature. We therefore applied the following model. Let X t denote the measured temperature at time t of a given Acousonde, and let μ t denote the true temperature at time t as measured by the CTD. Then, we assume that X t follows an Ornstein-Uhlenbeck process with a time varying mean of μ t + α , where α is an offset modeling a possible bias term, with decay rate β, and infinitesimal variance σ 2 . Here, β, α, and σ 2 are parameters to be estimated for each Acousonde. Thus, the model is a stochastic differential equation (SDE), where X t is the solution to the SDE where dW t are increments of a Brownian motion. We observe X t and μ t every 30 s, at time points t i = i∆ with ∆ = 30 s. We will write X t = X i and likewise for μ t . Then, the above model implies that X i + 1 = X i e −β∆ + (μ i + α) (1 − e −β∆ ) + ξ i + 1 where the ξ i 's are independent and normally distributed with mean 0 and variance σ 2 (1 -e −2β∆ ). The parameters can then be estimated by maximum likelihood, yielding the following estimates for the 7 Acousondes. dX t = − (X t − ( t + ))dt + dW t The typical time constants for the delay (the time it takes to decay to around 1/3 of the distance) of the 7 Acousondes in minutes are thus 1/(β × 60), with actual values: 7.23, 6.83, 6.02, 6.49, 6.8, 7.73 and 7.73.
Estimates of true temperature given the Acousonde measurements and the estimated parameters To estimate the true temperature, we use the Kalman filter. The Kalman filter is the standard algorithm to recover an unobserved coordinate from a linear Gaussian process. Thus, in the sequel, we assume μt unknown, and the goal is to recover μt based only on observations X i , i = 1,… n. To this end, we assume that the true temperature is a random process with Gaussian increments: where the η i s are independent and normally distributed with mean 0 and variance ̃2. The variance parameter ̃2 is estimated by: where n 1 and n 2 are the times where the Acousondes switched water temperature. We obtain ̃ = 0.00010296. The implementation of the Kalman filter then yields the estimates of the temperature as given in the following figure, where the green lines are the estimated temperatures. Notice that the gray points are not used in the estimation.
The Kalman filter seems to recover well the unknown temperature. In the following, the Kalman filter is applied to the records of four narwhals, with the parameters provided from the above analysis. The only parameter that has to be adapted to the setting of the tagged data is the variance of the true temperature, ̃2. Data from CTD tags attached to other animals in the same areas of East Greenland provide information about how temperature changes during normal diving patterns of narwhals.
The variability in the temperature measurements is very high at the surface (sometimes the temperature changes more than 2°C within one second, which is probably wrong), so only temperatures measured at depths of less than −30 m were used. Notice, however, that these two whales only dived to −200 m and −250 m, respectively, so we have no information about deeper dives. The estimate is ̃ = 0.002.
In the first plot, histograms of measured as well as estimated temperatures are presented. Note how the measured temperatures have a unimodal distribution, whereas the estimated temperatures follow a bimodal distribution. This is in agreement with the whales spending time in warmer waters either at the surface or at the bottom of the dives, and passing through colder waters when diving deep.
The next plots show example traces of 1 hr of measured and estimated temperatures together with depth. The black lines are the depths (scale on left axis), the green lines are the measured, and the blue lines are the estimated temperatures (scale on right axes). The dashed gray lines are at temperature 0°C, to show how temperature estimates are positive at the bottom of the dives, and only negative for short times when passing through the water column.
The last plots show the distribution of temperatures at different depths. Gray points are the measured temperatures, black points are the estimated temperatures, and the green curve is the average temperature through the water column in these waters. Fernando, Tusell (2011). Kalman Filtering in R. Journal of Statistical Software, 39(2).