Sea temperature effects on depth use and habitat selection in a marine fish community

J Anim Ecol. 2021;90:1787–1800. | 1787wileyonlinelibrary.com/journal/jane Received: 26 October 2021 | Accepted: 30 March 2021 DOI: 10.1111/1365-2656.13497 RESEARCH ARTICLE Sea temperature effects on depth use and habitat selection in a marine fish community Carla Freitas1,2 | David VillegasRíos3,4 | Even Moland1,5 | Esben Moland Olsen1,5 This is an open access article under the terms of the Creative Commons AttributionNonCommercialNoDerivs License, which permits use and distribution in any medium, provided the original work is properly cited, the use is noncommercial and no modifications or adaptations are made. © 2021 Institute of Marine Research, Norway. Journal of Animal Ecology published by John Wiley & Sons Ltd on behalf of British Ecological Society 1Institute of Marine Research, Flødevigen, His, Norway 2Marine and Environmental Sciences Center (MARE), Funchal, Madeira, Portugal 3Instituto Mediterráneo de Estudios Avanzados (CSICUiB), Department of Ecology and Marine Resources, Esporles, Balearic Islands, Spain 4Instituto de Investigaciones Marinas (CSIC), Department of Ecology and Marine Resources, Vigo, Pontevedra, Spain 5Centre for Coastal Research (CCR), Department of Natural Sciences, University of Agder, Kristiansand, Norway Correspondence Carla Freitas Email: [email protected] Funding information H2020 Marie SkłodowskaCurie Actions, Grant/Award Number: 793627; Norges Forskningsråd, Grant/Award Number: 294926; Regionale forskningsfond Oslofjordfondet, Grant/Award Number: 272090 Handling Editor: Catherine Parr Abstract 1. Understanding the responses of aquatic animals to temperature variability is essential to predict impacts of future climate change and to inform conservation and management. Most ectotherms such as fish are expected to adjust their behaviour to avoid extreme temperatures and minimize acute changes in body temperature. In coastal Skagerrak, Norway, sea surface temperature (SST) ranges seasonally from 0 to over 20°C, representing a challenge to the fish community which includes cold- , cooland warmwater affinity species. 2. By acoustically tracking 111 individuals of Atlantic cod Gadus morhua, pollack Pollachius pollachius and ballan wrasse Labrus bergylta in 2015– 2018, we examined how coexisting species within a fish community adjusted their behaviour (i.e. vertical distribution in the water column and habitat selection) to cope with the thermal variation. 3. Mixedeffect models showed that thermal preference was a main driver of behaviour and habitat use of the fish community in a southern Norwegian fjord. Cod used colder waters, compared with pollack and ballan wrasse. Increases in SST during summer were associated with the use of deeper, colder waters by cod, especially by larger individuals, and conversely with the occupancy of shallower areas by pollack and ballan wrasse. During winter, when SST dropped and the thermal stratification reversed, pollack and ballan wrasse moved to deeper, relatively warmer areas, while cod selected shallower, colder habitats. Although habitat selection was affected by temperature, speciesspecific habitat selection was observed even when temperature was similar throughout habitats. 4. This study shows how cohabiting fish species respond to thermal heterogeneity, suggesting that (a) temperature regulates the access to the different depths and habitats and (b) behavioural plasticity may be an important factor for coping with temperature variability and potentially for adaptation to climate change. KEYWORDS acoustic telemetry, biologging, habitat use, resource selection functions, Vemco VPS 1788 | Journal of Animal Ecology FREITAS ET Al. 1 | INTRODUCTION Temperature is a main determinant of the distribution and behaviour of organisms. By directly impacting the thermal environment, climate change has increased the need to understand how species respond to temperature variability (Pacifici et al., 2015; Sinclair et al., 2016). Changes in climate are expected to alter existing environments and create novel ones, which may lead to local loss of organisms and their associated function in the ecosystem (Corrales et al., 2018; Pinsky et al., 2019; Yeruham et al., 2020). It has been suggested that behavioural responses, such as shifts in activity in space and time, may enable species to cope with thermally stressful environments, allowing them to maintain their functional roles in the ecosystem despite changed temperature (Fey et al., 2019; Wolff et al., 2020). Marine ecosystems provide fewer microclimates compared with terrestrial ecosystems, making marine animals, in particular ectotherms, more vulnerable to warming (Pinsky et al., 2019). However, behavioural thermoregulation is possible in the ocean. Some shark species, for example, undertake diel vertical migrations between deep cold waters, where temperature is favourable, and shallow feeding areas, which are less favourable thermally (Sims et al., 2006; Vaudo et al., 2016). Conversely, tunas and sea turtles, forage at depth and use warmer surface waters to thermoregulate (Freitas et al., 2019; Holland et al., 1992). Other organisms, such as American lobster Homarus americanus, seek optimal thermal habitats by moving seasonally in relation to water temperature (Jury & Watson, 2013). In the absence of other constraints, it is expected that ectotherms will seek suitable thermal conditions within their available habitat. However, the use of thermal refuges during unfavourable temperature conditions may be challenging if such behavioural thermoregulation is depriving species from vital resources. For instance, moving deeper can allow marine species to avoid hot surface temperatures but may limit their access to prime foraging habitats (Freitas et al., 2016). Shifting to deeper areas may also expose species to hypoxia, as oxygen availability typically decreases with depth (Deutsch et al., 2015). Furthermore, it is expected that changes in ocean temperature will pose significant risk to species that exhibit strong behavioural preferences for specific habitat types while generalist species are assumed to be less vulnerable (Matis et al., 2018). While thermal conditions are expected to directly constrain species distribution, changes in thermal conditions may also allow for species coexistence through thermal resource partitioning. More specifically, thermal preferences may enable species to coexist because the thermal regime will dictate when a particular habitat is available to each species (e.g.Attrill & Power, 2004; Crowder et al., 1981). Highlatitude marine ecosystems offer a unique opportunity to understand the patterns of resource use by species with different thermal affinity. For instance, sea surface temperature along the Norwegian Skagerrak coast ranges annually between 0°C and more than 20°C. Skagerrak is also characterized by temperature stratification inversions during winter (i.e. cool water lies above warmer water). Such inversions, typical for subpolar regions, form when the relatively warmer surface water of summer is trapped by the cooler, fresher conditions that exist during winter (Sprintall et al., 2019). Although sea temperature in Skagerrak has increased ~1°C in the last decades, seasonal oscillations and inversions in sea temperature are natural phenomena (Albretsen et al., 2012). Seasonal oscillation may, however, become even more pronounced in the future, given a predicted rise of 2– 4°C in sea surface temperature in this region by the end of the century (Dye et al., 2013; Gröger et al., 2019). Large seasonal variations in sea temperature in Skagerrak likely represent a challenge to the local fish community, which includes cold- , cooland warmwater affinity species, such as Atlantic cod Gadus morhua, pollack Pollachius pollachius and ballan wrasse Labrus bergylta, respectively. Cod is a coldwater species distributed in the northern areas of the North Atlantic. Their optimal temperature for growth ranges between 9 and 15°C (Björnsson & Steinarsson, 2002; Björnsson et al., 2007; Lafrance et al., 2005). Pollack occurs further south in the Atlantic, from Portugal to northern Norway and shows maximum growth at temperatures between 12 and 18°C (PersonLe Ruyet et al., 2006; Suquet et al., 1996). Finally, ballan wrasse is the most warmwateradapted species, occurring from Morocco up to southern Norway (Sayer & Treasurer, 1996). Juveniles (~15 g) grow faster at temperatures above 16°C (CavroisRogacki et al., 2019), whereas larger individuals (~300 g) also show increased physiological performance when temperature increases from 0 to 25°C, with low metabolic rates and inactivity at low temperatures, that is, 5– 10°C (Yuen et al., 2019). In coastal Skagerrak, cod typically avoids shallow waters when temperatures rises above 16°C and instead choose deeper waters that are up to 5°C cooler (Freitas et al., 2015, 2016). In this study, we used acoustic telemetry to track cooccurring cod, pollack and ballan wrasse in coastal Skagerrak. We analysed the intraspecific and interspecific variation in behaviour of these ecologically important demersal predators over a period of 3 years to understand how coexisting species with contrasting thermal affinity cope with the striking seasonal oscillations in water temperature. We hypothesized that (a) species with contrasting thermal tolerance will display different strategies (i.e. habitat selection and vertical distribution in the water column) to cope with the thermal variation and (b) within a species, individual traits such as body size will influence their behavioural plasticity to temperature change. 2 | MATERIALS AND METHODS 2.1 | Study area This study was carried out in a marineprotected area (MPA) located in the Tvedestrand fjord on the Norwegian Skagerrak coast (Figure 1). The MPA, established in 2012, is closed to all fishing activities and this qualifies as a notake marine reserve where fish behaviour can be studied in the absence of any disturbance from fishing gear. The MPA comprises waters ranging from 0 to 90 m | 1789 Journal of Animal Ecology FREITAS ET Al. depth. Seafloor habitats were mapped in 2013 using video surveys (Freitas et al., 2016). The following eight habitat types are found in the study area (Figure 1): Eelgrass (soft substrate covered by eelgrass Zostera marina; present at depths 0– 6 m); Vegetated hard substrate (hard substrate covered by macroalgae; present at depths 0– 17 m depth approximately); Nonvegetated hard substrate (hard substrate with no macroalgae; present below 17 m depth approximately); Coral rocky substrate (uneven hard substrate covered by soft corals and patchy macroalgae); Anemone rocky wall (steep rock wall covered by anemones and other invertebrates); Boulders (bare boulders; present in the steep western margin of the fjord); Sand (soft, oxygenated substrate) and Soft anoxic substrate. Sea temperature in the study area was monitored using temperature loggers deployed at six different depths: 1, 5, 10, 15, 20 and 33 m. Temperatures at 1– 20 m were recorded using Hobo UA0026 loggers, which registered temperature each hour. Temperature at 33 m was recorded using a Vemco V13T1x sensor, logging temperature each 500– 700 s. All temperature records were later averaged daily. Temperature at 1 m depth is hereafter referred to as sea surface temperature. FIGURE 1 Maps of the study area in Tvedestrand (Tved.), Norway, showing the habitat map (a) and filtered locations obtained from cod (b), pollack (c) and ballan wrasse (d). Panel d shows also the network of acoustic receivers and borders of the notake marine reserve (NTMR). Pictures in panel e show the following habitats: eelgrass; H_An, rock wall with anemones; H_Bo, boulders, H_Co, Hard substrate with corals; H_Nv, nonvegetated hard substrate; H_Ve, Vegetated hard substrate and; Sand, oxygenated soft substrate. Soft anoxic substrate not shown (a) (b) (c) (d) (e) 1790 | Journal of Animal Ecology FREITAS ET Al. 2.2 | Fish telemetry A total of 111 fish (75 cod, 19 pollack and 17 ballan wrasse) were tagged in 2015– 2017, using either Vemco transmitters V9P or V13P, equipped with pressure sensors (Table 1). Transmitters were set to transmit a signal every 110– 250 s, with a random interval to reduce code collision. Tagging procedures have been described elsewhere (Olsen et al., 2012; VillegasRíos et al., 2017). A Vemco positioning system (VPS) was used to obtain information on fish depth and horizontal location (Figure 1). The VPS system was composed of (a) 33 acoustic receivers (Vemco VR2W, 69 kHz), used to record data transmitted by the tagged fish; (b) 33 synchronization tags, ‘Synctags’ (Vemco V164x), moored along with each receiver to correct for clock drift between receivers and (c) four reference tags (three Vemco V131x and one Vemco V13T1x) placed within the receiver grid to measure system performance during location estimation. Earlier performance tests indicated that the median location error of the VPS array was 1.8– 4.4 m (Freitas et al., 2016). Two datasets were obtained from this telemetry system: (a) detection data (fish ID, detection time, depth) downloaded from the receivers and (b) fish location data (fish ID, detection time, depth, latitude, longitude, horizontal position error) provided by Vemco after postprocessing of the detection data. Horizontal position error (HPE) is a relative, unitless estimate of how sensitive a calculated position is to errors in its inputs (Smith, 2013). Locations with the highest HPE values were removed from the dataset (7.7% of the locations), based on a tradeoff between accuracy and quantity, as described in Freitas et al. (2016). Locations outside the study area (i.e. the notake marine reserve; Figure 1) were also removed from the dataset. Data were inspected for individual fate, namely death and dispersal, following the method described in VillegasRíos et al. (2020). Data obtained after natural mortality events, characterized by stabilized vertical and horizontal movements, were removed from the datasets. 2.3 | Data analysis 2.3.1 | Thermal habitat and temperature use Temperature profiles were generated from the available daily temperature data at 1, 5, 10, 15, 20 and 33 m, using linear interpolation in between measurements. Temperature at >33 m was assumed to be the same as at 33 m depth. Based on the analysis of the temperature profiles (Figure 2), four seasons were defined: winter (January– March), spring (April and May), summer (June– September) and fall (October– December). Winter was defined as the months when surface temperature was typically colder than deeper water masses. Conversely, summer was defined as the months when surface layers were warmer than deeper water masses (Figure 2). Average depth used by each fish during day and night was calculated using the detection data. For a given day, average depth during the day was calculated as the average of all depths recorded between sunrise and sunset (i.e. when solar elevation was ≥0°). Average depth during the night was calculated as the average of all depths recorded from sunset in the previous day to sunrise of that day. Note that repeated depths records (i.e. the same depth record received simultaneously by multiple receivers) were removed from the dataset before calculating mean depths. Temperatures at mean daily depths were later calculated by linear interpolation of temperatures recorded by temperature loggers. 2.3.2 | Vertical distribution relative to sea temperature Generalized additive mixedeffect models (GAMMs) were used to test whether depth use was affected by sea surface temperature and fish body length. Separate models were fitted for each species during the day and night and took the following form: Depthi,j denotes the average depth used during daytime or nighttime by individual i at day j. T1mj is the average sea surface temperature at day j, modelled as a smooth function (penalized regression splines, with 4 knots). Leni is the fish body length, modelled as a linear term, with slope β. The term ƒ (T1mj, Len) tests for an interaction between temperature and body length. The random intercept αi allows for a random variation around the intercept α and is assumed to be normally distributed with mean 0 and variance 𝜎2 i . The term εij is independently normally distributed noise. An autoregressive process of order 1, corAR1 (see Zuur et al., 2009), was added to the random structure of the model to take temporal autocorrelation into account. Model selection was performed based on the Akaike information criterion, AIC (Table S1). Model validation took place by inspecting residual distributions and also residuals against fitted values and covariates, as recommended by Zuur and Ieno (2016). Depthi,j =𝛼+ f(T1mj) +𝛽 Len + f(T1mj, Len) +𝛼 i +𝜀 ij. TABLE 1 Summary data for the 111 fish acoustically tracked in Tvedestrand fjord from May 2015 to April 2018. Selected VPS locations refer to the number of VPS locations used in the analyses (10 best VPS locations per day; see Section 2.3.3) Species Number of individuals Body length, cm: min– max (M ± SD) Days tracked: min– max (M ± SD) Number of detections Number of VPS locations Selected VPS locations Cod 75 34– 74 (48 ± 9) 29– 884 (265 ± 176) 12,698,625 830,134 126,874 Pollack 19 35– 52 (40 ± 5) 52– 670 (311 ± 175) 4,373,925 328,920 40,710 Ballan wrasse 17 26– 42 (35 ± 5) 57– 702 (389 ± 222) 2,576,268 139,128 32,107 Total 111 29– 884 (292 ± 187) 19,648,818 1,298,182 199,691 | 1791 Journal of Animal Ecology FREITAS ET Al. Models were fitted in the R software (R Core Team, 2018) using the package mgcv, function gamm (Wood, 2006). 2.3.3 | Habitat selection Bottom habitat type was extracted for each fish VPS position using QGIS software. To investigate whether fish were close to the bottom or in the water column, bottom depth was also extracted for each position, using 5 m resolution bathymetry data available from the Institute of Marine Research. Resource selection functions (RSFs) were used to quantify habitat selection by the study species. An RSF estimates the probability of that animal using a certain resource proportional to the availability of that resource in the environment (Boyce & McDonald, 1999; Johnson et al., 2006; Manly et al., 2002). We estimated RSFs using a useavailability design and used logistic regression to compare the habitat selected by fish (VPS locations, coded as ‘1’) to what was theoretically available (random locations, coded as ‘0’)— see Johnson et al. (2006). An RSF above 1 denotes selection (i.e. disproportional use relative to availability), whereas an RSF below 1 denotes avoidance; an RSF equal to 1 denotes proportional use relative to availability. The number of locations obtained per day for each fish varied from 1 to 327 (M ± SD = 55.6 ± 56.4), distributed generally through 1– 3 habitats. To decrease pseudoreplication of data, improve temporal independency and decrease unbaled observations between days, we selected 10 positions per day for each fish, by retaining the best 10 positions (i.e. with the lowest HPE). Hence, the resulting dataset used for further analysis of habitat selection and use had a maximum of 10 locations per day for each fish (M ± SD = 8.6 ± 2.9). To represent availability, we drew random points from the area detectable by our array of receivers, using the same number of points as in the sample of used locations (n = 199,691, see Section 3) and extracted habitat type for these positions. Following the recommendations in McDonald (2013), RSFs FIGURE 2 Water temperature recorded in Tvedestrand between April 2015 and April 2018, and average depth used by cod (N = 75), pollack (N = 19) and ballan wrasses (N = 17) during the day (magenta dots) and night (white dots). Dashed vertical lines delineate seasons: summer, fall, winter and spring. Temperatures below 33 m were assumed to be equal to the 33 m measurements (see Section 2.3.1) 2016 2016 2016 2017 2017 2017 2018 2018 2018 Summer SummerSummerWinter Winter Winter Month and year Fall FallSpring Spring Fall 110035318 1792 | Journal of Animal Ecology FREITAS ET Al. were estimated from the logistic regression coefficients, using an exponential link: In the equation, w(x) is the RSF and β1 to βn represent the coefficients for the variable X1 to Xn. To account for differences in sampling intensity and random variation between individuals, random intercepts for each individual were added to the logistic model. Telemetry location data are often autocorrelated, which does not influence estimates of model coefficients but can deflate standard errors (Fieberg et al., 2010; Schielzeth et al., 2020). To minimize temporal autocorrelation, we added Julian day, nested within individual, to the randomeffects structure. We fitted 12 RSFs, representing different combinations of seasons and species. Models were fitted using the function glmer() of the r package lme4 (Bates et al., 2015). 2.3.4 | Habitat use Following Lele et al. (2013), a habitat use distribution can be used to quantify the probability that a used habitat unit is of type x, that is, p(x). Considering only the set of used locations, we quantified p for the most common habitats in the study area (eelgrass, vegetated hard substrate, nonvegetated hard substrate and sand), using GAMMs, with a binomial link: In this model, pij denotes the probability that a used habitat unit is of type x. Surface temperature (T1m), diel period, that is, day or night (Diel) and fish body length (Len) were included as explanatory variables. The interaction Diel × Len was used to investigate whether diel changes in the probability of using habitat x were dependent on fish body size. Individual ID was used as randomeffect variable (αi) to account for individual variability. An autocorrelation structure of type AR1 was used to account for temporal autocorrelation. Model selection was performed based on the AIC (Table S2). Models were fitted using the function bam() of the r package mgcv (Wood et al., 2015). 3 | RESULTS Individual cod, pollack and ballan wrasse were tracked for 29– 884 days (M ± SD: 292 ± 187), between May 2015 and April 2018 (Table 1). A total of 19,648,818 detections and 1,298,182 VPS locations were obtained. From these, 199,691 VPS locations were selected and further used in the habitat analyses. 3.1 | Thermal habitat and temperature use High thermal heterogeneity was observed in Tvedestrand fjord, both temporally (throughout the year) and spatially (vertically in the water column). Sea temperature ranged between −1.2 and 21.2°C at 1 m depth and between 6.1 and 13.5°C at 33 m depth. During summer, surface layers surpassed 15°C, whereas in winter they remained below 5°C (Figure 2). During the transition seasons (spring and fall), no vertical thermal gradients were observed, that is, water temperature was nearly homogeneous throughout the water column (Figure 2). Overall, cod and pollack used similar temperature ranges, between 0 and 19.3°C, while ballan wrasse used temperatures from 4.6°C to 19.4°C. During spring and autumn, when water temperatures were similar throughout the water column, no substantial differences in temperature use were found between species (Figure 3). During summer (June– September), cod and pollack showed affinity to colder waters, using median temperatures 2.5°C colder than ballan wrasses. In winter (January– March), cod used colder waters compared with both pollack and ballan wrasse (Figure 3). Note that the latter two species used similar median winter temperatures, but pollack sometimes explored colder areas (Figure 3). 3.2 | Depth use relative to sea temperature Cod, pollack and ballan wrasse were detected from surface to a maximum of 62, 75 and 76 m depth, respectively. The largest mean depth during the day or night was 54, 69 and 76 m for cod, pollack and ballan wrasse, respectively (Figure 2). A significant relationship was found between the vertical distribution of each species and sea surface temperature (Figures 2 and 4; Table 2). Responses to temperature were, however, contrasting between species. Pollack and ballan wrasse occupied shallower waters when sea surface temperature increased, moving to deeper (warmer) water layers when surface temperature decreased in winter (Figures 2 and 4). This pattern was particularly evident for ballan wrasse. Cod, in opposite, constrained themselves to deeper (colder) waters when surface temperature increased in summer, occupying shallower areas when surface temperature dropped (Figures 2 and 4). In winter, in particular when sea surface temperature fell under 5°C, cod moved again to deeper (warmer) areas. Body size had a significant effect on the vertical position of cod relative to temperature during the night, with larger individuals located in deeper, colder areas when surface temperature increased (Figure 4; Table 2). No relationship was found between depth use and body size for pollack and ballan wrasse (Table 2). 3.3 | Habitat selection RSFs, used to compare selected habitats relative to habitat availability, revealed interspecific and seasonal differences in habitat selection (Figure 5). Of the three species, cod used the widest range of habitats. During spring and fall, when there were no vertical temperature constrains (i.e. temperatures were homogeneous throughout the water column), cod selected eelgrass and hard bottom habitats (i.e. anemone rock walls, coral rock substrates, vegetated and w( x ) = exp( 𝛽 1 X 1 +𝛽 1 X 1 +…𝛽 n X n). Logit(pij) =𝛼+ f(T1mj) +𝛽 1Dielj +𝛽 2Leni +𝛽 3Dielj × Leni +𝛼 i +𝜀 ij. | 1793 Journal of Animal Ecology FREITAS ET Al. FIGURE 3 Temperature use by cod, pollack and ballan wrasse in Tvedestrand fjord, Norway. Data are shown by season and diel period. Number of daily observations is given above each box. Boxes show the median, as well as the first and third quartiles (the 25th and 75th percentiles). Notches give a rough 95% confidence interval for comparing medians— see McGill et al. (1978). Black dots are extreme values that extend outside one and a half times the interquartile range ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 7892055 711 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 8212139 724 18149794 1880 19329790 1897 ● ● ● ● ● ● ● ● ● ● ● ● 18834812 1668 ● ● ● ● ● ● ● ● ● ● ● ● ● ● 19034777 1674 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 18072930 1542 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 18232913 1535 Spring Summer Fall Winter Night Day CodPollack Ballan CodPollack Ballan CodPollackBallan CodPollackBallan 0 5 10 15 20 0 5 10 15 20 Temperature (°C) FIGURE 4 Average depth used by cod, pollack and ballan wrasse during the day and night as a function of sea surface temperature, as predicted from GAMMs. Solid lines are estimated mean effects and dashed lines are 95% confidence intervals. Body size was found to have a significant effect on depth use by cod during the night, as illustrated in the inset 1794 | Journal of Animal Ecology FREITAS ET Al. nonvegetated hard substrate), avoiding sand and anoxic habitats. In summer, cod decreased the preference for eelgrass and vegetated hard habitats located in shallower areas. In contrast, ballan wrasse showed a preference for eelgrass during spring and summer months when temperatures were higher in the shallows. Both pollack and ballan wrasse exhibited a high preference for rocky walls with anemones yearround, especially during summer. In addition, pollack selected boulders, as well as vegetated and nonvegetated hard substrates, avoiding eelgrass, hard bottom with corals and sand. All species avoided anoxic habitats, except pollack which selected areas above anoxic fjord bottom during spring (Figures 2 and 5). This does not imply that pollack was necessarily located in anoxic layers, as this species swam on average 26.4 (±3.8) m above the seafloor when detected over anoxic habitats (Figure 6). 3.4 | Habitat use For all three species, habitat use was clearly associated with sea surface temperature (Figure 7; Table S3). Furthermore, habitat use by cod was significantly affected by diel period and individual body size (Table S3). Cod showed a higher probability of using eelgrass and vegetated hard substrate, available in shallow areas, when sea surface temperature was between 5 and 15°C, using deeper TABLE 2 Parameter estimates for generalized additive mixed models (GAMMs) used to model the mean depth used during day and night (response variables) by cod, pollack and ballan wrasse in Tvedestrand, Norway. Explanatory variables were surface temperature (T1m), fitted as a smooth term, body size (Len), fitted as a linear term, as well as the interaction between these [ti(T1m, Len)] Response Species Smooth term EDF F p value Linear term βSE t Mean depth at day Cod s(T1m) 2.94 53.27 <0.001 Pollack s(T1m) 2.20 8.46 <0.001 Ballan s(T1m) 2.74 81.77 <0.001 Intercept 25.90 10.17 2.55 ti(T1m, Len) 4.37 4.98 <0.001 Len −0.18 0.29 −0.61 Mean depth at night Cod s(T1m) 2.82 16.85 <0.001 Intercept 6.11 3.03 2.02 ti(T1m, Len) 1.00 5.59 0.018 Len 0.13 0.06 2.11 Pollack s(T1m) 1.00 8.02 0.005 Ballan s(T1m) 2.51 38.04 <0.001 FIGURE 5 Resource selection functions (RSFs, with 95% confidence intervals) for cod, pollack and ballan wrasse in coastal Skagerrak in different seasons. RSFs provide the probability of using a habitat relative to its availability. RSFs above 1 denote selection (disproportional use relative to availability); below 1 denote avoidance and equal to 1 denote proportional use relative to availability. Note different ranges in the yaxis. H_An, rock wall with anemones; H_Bo, boulders, H_Co, Hard substrate with corals; H_Nv, nonvegetated hard substrate; H_Ve, Vegetated hard substrate; Sanox, Soft anoxic substrate | 1795 Journal of Animal Ecology FREITAS ET Al. FIGURE 6 Mean (±SE) vertical position of tagged cod, pollack and ballan wrasse relative to bottom depth as a function of habitat type in Tvedestrand fjord. Number of individuals in each habitat is provided for each species. Fish positions below sea bottom may arise from inaccuracies on bottom bathymetry and/or on fish depth and location data. Abbreviations see Figure 5 FIGURE 7 Probability that a used habitat unit is of type x, p(x), during the day and night, as a function of sea surface temperature, as predicted from GAMMs for cod, pollack and ballan wrasse in Tvedestrand. Habitat types are provided at the top of each panel row. Solid lines are estimated mean effects and dashed lines are 95% confidence intervals