Individual variability in activity patterns in Atlantic halibut ( Hippoglossus hippoglossus ) revealed using pop‐up satellite tags

Abstract

Understanding fish behaviour and activity patterns is essential for interpreting their ecology and the processes that shape population dynamics, yet such information remains limited for wild fishes because observing individuals in situ is challenging. Recent technological advances make it possible to collect high‐resolution movements and activity data over extended periods, opening the door to detailed descriptions of fine‐scale behaviours and their temporal variability. We evaluated the coarse and fine‐scale activity patterns of 25 Atlantic halibut (Hippoglossus hippoglossus) individuals throughout a complete seasonal cycle in the Gulf of St. Lawrence using acceleration data at a resolution of 5 s extracted from 25 pop‐up satellite archival tags (PSAT). We sought to classify individuals into behavioural contingents within the population and to identify specific behaviours that could be identified with the fine resolution of the available data using dynamic factor analysis, visual representations and variance partitioning. Despite some common general patterns of activity during reproduction, the data were characterized by a high level of individual variability in the amount and patterns of activity, with each halibut exhibiting a unique activity profile across the year. Periodic daily behaviours (diurnal and nocturnal) were identified for several individuals but with no clear pattern in their recurrence over time within individuals or coherence between individuals. Overall, individual variability in activity dominated over the common patterns detected, with little evidence of distinct activity contingents among individuals. The high‐frequency activity data available from PSATs in this study challenge the traditional view of halibut and flatfish in general, as passive bottom dwellers exhibiting simple, perhaps periodic behaviours, instead supporting growing evidence that fish exhibit substantial individual behavioural differences.

1. INTRODUCTION

Across animal taxa, variability in behaviour is observed at the individual level (Briffa & Weiss, 2011; Dingemanse et al., 2010; Japyassú & Malange, 2014; Snell‐Rood & Steck, 2019; Wilson et al., 2011). Individual variability is particularly important in vertebrates, which have evolved complex behaviours in response to a combination of intrinsic (e.g. reproductive status or nutritional needs), demographic (e.g. sex ratio, age and size structure) and extrinsic (e.g. predation, habitat characteristics) factors (Budaev et al., 1999; Dingemanse et al., 2007; Shettleworth, 2010). Understanding individual variability in behaviour is important as it can impact population dynamics (Bolnick et al., 2003, Gondin 1997, Baptista et al., 2020) by generating variability in dispersal patterns (Cote et al., 2010; Fraser et al., 2001), reproductive success (Mittelbach et al., 2014) and survival (Cote et al., 2010; Smith & Blumstein, 2008).

Fishes were underrepresented in early publications on behavioural ecology (Godin, 1997), as their behaviour was considered simple and instinctive (Budaev & Brown, 2011; Magurran, 1986). However, high variability in behaviour has commonly been observed among individuals (Bolnick et al., 2003; Conrad et al., 2011; Kristiansen & Fernö, 2007; Metcalfe et al., 2016; Millot et al., 2014; Rupia et al., 2016) and within individuals over multiple time scales (Bell & Stamps, 2004; Fürtbauer et al., 2015; Reebs, 2002; Robert et al., 2013; Roy et al., 2017; Roy & Bhat, 2018; Shaw, 2020). Nevertheless, knowledge of individual fish behaviour in the wild remains scarce and limited to particular taxa given the complexity of observing individuals in situ (Broell et al., 2013). For this reason, fish behaviour is still often conceptualized at the population or species level (Magurran, 1986; Punt et al., 2015; Tyler & Rose, 1994), and individuals are assumed to exhibit uniform behaviour, ignoring variability at the individual level (Budaev et al., 2015; Ringler, 1983). Ringler (1983) discussed the impact of variation in individual foraging tactics (e.g. changes in strike pattern, preference or feeding location) and its impact on maintenance, growth and reproduction that in turn affect population recruitment and mortality. For example, individual differences in ‘risk‐taking’ behaviour (e.g. wandering out of refuge more frequently) modulate predation risk as well as access to resources and mates (Biro et al., 2003; Hulthén et al., 2017; Roy et al., 2017). More interest is now directed towards fish behaviour with the consideration that it represents a key component of population dynamics (Budaev & Brown, 2011).

Electronic tagging, particularly through the use of pop‐up satellite archival tags (PSAT), has enabled the study of individual fish behaviour in situ, and these tags are now widely used to study the behaviour of large‐bodied fish (Armsworthy et al., 2014; Block et al., 2001; Gunn & Block, 2001; Hussey et al., 2015; Lutcavage et al., 1999). PSATs are installed on the fish with a preprogrammed date when the tags will be released and float back to the surface. PSATs record and archive various data time series (e.g. temperature, depth, light level), from which a summary can be transmitted directly through the Argos satellite network upon release (Block et al., 2001; Broell et al., 2016; Fisher et al., 2017; Loher & Seitz, 2006). Data are typically transmitted in a summarized form due to constraints on battery capacity as the amount of data is transmitted through satellite. However, if tags are physically recovered after popping up, the full archived datasets can be accessed. The high‐resolution data archived in PSATs facilitate inferences of behaviours such as spawning, migration and foraging, as well as temporal variability in behavioural expression on scales ranging from minutes to seasons (Broell et al., 2013; Gatti et al., 2020; Knotek et al., 2020; Le Bris et al., 2018; Nielsen et al., 2018; Ransier et al., 2024; Seitz et al., 2005, 2011; Skubel et al., 2020). Scott et al. (2016) used depth records from recovered PSATs to show that Pacific halibut (Hippoglossus stenolepis) exhibited seasonal variation in ‘periodic behaviour’, behavioural patterns linked to environmental cycles such as daylight, tide cycle or season (Li et al., 2012). Another advantage of recovering PSATs is to access the accelerometer data available in the newer generations of PSATs. These data are useful for quantification of survival rate (Nielsen et al., 2018), parametrization of movement models (Gatti et al., 2021), vessel avoidance (Rohan et al., 2024), and identification of periodic behaviour in combination with measurements of environmental conditions (Skubel et al., 2020).

Atlantic halibut (Hippoglossus hippoglossus Linnaeus 1758) in the Gulf of St. Lawrence (GSL), Canada, has been the subject of an extensive tagging programme focused on identifying migration patterns and spawning locations (Fisher et al., 2017; Gatti et al., 2020, 2021; James et al., 2020; Le Bris et al., 2018; Murphy et al., 2017). A total of 114 PSATs were deployed between 2013 and 2017 on large halibut, of which 65 were physically recovered (Gatti et al., 2020). A geolocation model relying on depth and temperature data revealed that a large proportion of individuals migrated between coastal feeding grounds in the summer and deep offshore spawning grounds in the winter, whereas some individuals remained in deep waters year round (Gatti et al., 2020; James et al., 2020; Le Bris et al., 2018). These studies also suggested individual variability in migration behaviour (Gatti et al., 2020; Ransier et al., 2024). This supports the contingent hypothesis, which posits that a population comprises multiple groups, each exhibiting a common migratory behaviour that differs between groups (Clark, 1968; Secor, 1999). PSATs deployed in 2017 were equipped with an accelerometer, providing a unique opportunity to use acceleration data to further infer fine‐scale individual behavioural patterns of halibut over a full seasonal cycle. In the present study, we use acceleration data from recovered PSATs to (1) characterize the extent and diversity of halibut activity and (2) reveal daily‐to‐seasonal temporal variability in behaviour across different geographic sectors of the GSL. We also aim to identify contingents, inferred from recurring patterns of activity shared by groups of individuals.

2. MATERIALS AND METHODS

2.1. Satellite tagging operations and data collection

A total of 36 PSATs (MiniPat Wildlife Computers Inc., Redmond, WA, USA) were deployed on Atlantic halibut across the GSL between 15 September and 23 October 2017 (Figure 1). Tagging operations took place during a collaborative halibut longline survey co‐ordinated by Fisheries and Oceans Canada. Tagging was undertaken on chartered fishing vessels in the waters of the Esquiman Channel, Gaspe Peninsula, Anticosti Gyre, Cape Breton and Prince Edward Island (Figure 1). To avoid physical damage to the fish, halibut were caught using longlines equipped with baited number 14 and number 16 circle hooks and were carefully brought onboard to be handled on deck. The fork length of each halibut was measured to the nearest centimetre, and individuals over 130 cm that appeared healthy were tagged with a PSAT. Tagged halibut ranged in size between 130 and 202 cm (averaging 151 cm; Table 1) and mostly consisted of mature individuals. Marshall et al. (2023) determined the sexes for 19 individuals using either genetics sex determination or visual analysis of data collected by the PSATs related to spawning rises to identify typical distinct behaviour for males and females. Tagging operations were carried out following the methods developed by Seitz et al. (2005) for Pacific halibut. PSATs were attached using a 10‐cm 180‐kg test nylon monofilament tether linked to a sterilized titanium dart. The anchoring dart was inserted under the pterygiophores on the dorsal fin from the eyed side of the fish. Two moored control PSATs were also deployed to record environmental parameters and provide a baseline for the effect of tidal currents on acceleration recordings: one in shallow waters (48 m) near Port au Choix, Newfoundland, and another in the deep Laurentian channel at the level of Cabot Strait (342 m) (Figure 1).

FIGURE 1.

Map of the Gulf of St. Lawrence (Canada), showing deployment (circle) and pop‐up (star) locations for the 25 recovered MiniPAT pop‐up satellite archival tags (PSAT) that were deployed on Atlantic halibut (Hippoglossus hippoglossus) in 2017 and subsequently recovered, and two moored tags. The coloured symbols represent the deployment region based on Gatti et al. (2020).

TABLE 1.

Summary of the physically recovered PSATs deployed on Atlantic halibut and on moorings in the Gulf of St. Lawrence in 2017.

PSAT ID	Fork length (cm)	Sex	Deployment				Number of days at sea	Linear distance (km) from tagging to pop‐up	Estimated travelled distance (km)	Percentage of the year spent active
			Date (2017)	Depth (m)	Latitude	Longitude
16P2437	164	F, g	09–25	206.79	49.73	−59.12	341	4.41	913	34.71
17P0086	146	F, d	09–19	29.28	46.70	−63.70	347	19.33	1321	14.35
17P0087	148	–	10–03	131.76	48.46	−63.86	333	13.06	78	3.83
17P0126	145	–	10–23	201.30	49.98	−65.13	313	15.38	25	5.31
17P0127	147	F, g	10–23	201.30	49.98	−65.13	276	–	127	7.61
17P0135	159	–	09–25	206.79	49.73	−59.12	172	21.52	426	19.27
17P0136	138	F, g	10–03	210.45	48.84	−63.74	333	7.84	90	11.41
17P0138	166	F, d	10–03	210.45	48.84	−63.74	333	42.54	96	7.62
17P0140	161	–	10–03	131.76	48.46	−63.86	333	3.77	474	4.86
17P0143	146	F, d	09–27	290.97	50.46	−58.02	339	5.54	1345	13.60
17P0147	171	F, g	10–10	215.94	49.45	−63.80	326	10.63	70	11.81
17P0148	171	F, d	09–26	109.80	49.37	−58.66	340	36.95	154	24.55
17P0149	150	F, g	10–23	201.30	49.98	−65.13	171	40.16	412	6.08
17P0150	155	F, d	09–25	206.79	49.73	−59.12	341	14.22	372	28.62
17P0154	180	F, g	10–23	210.45	49.98	−65.15	313	30.28	238	16.63
17P0155	160	F, g	09–25	206.79	49.73	−59.12	341	8.59	130	24.45
17P0159	146	F, g	09–27	287.31	50.45	−58.03	339	1.51	9	10.57
17P0162	143	F, g	10–23	201.30	49.98	−65.13	313	4.29	399	9.07
17P0165	130	–	09–16	162.87	47.21	−60.52	351	6.16	332	11.89
17P0166	135	M, g	10–03	210.45	48.84	−63.74	333	10.51	260	15.84
17P0169	130	F, d	09–27	289.14	50.46	−58.02	298	34.90	320	19.01
17P0170	139	M, d	10–03	210.45	48.84	−63.74	333	1.78	647	23.18
17P0171	165	F, g	10–03	131.76	48.46	−63.86	333	9.65	411	7.68
17P0173	168	–	10–15	32.94	46.65	−63.44	293	314.87	326	10.94
17P0176	202	F, d	09–23	43.92	46.95	−63.66	343	86.83	236	12.81
Moored PSATs
17P0197	–	–	09–14	342	47.46	−59.20	348	–	–	–
17P0198	–	–	09–25	48	50.73	−27.37	344	–	–	–

Note: Sex was based on the results of Marshall et al. (2023) using either genetics (g) or depth profile analysis (d) for each individual. The number of days at sea in bold font corresponds to individuals whose tags did not remain attached for more than 300 days and are therefore not considered full time series. Linear distance (km) from tagging to pop‐up and estimated travelled distance (km) are based on calculation from Gatti et al. (2020).

Abbreviation: PSAT, pop‐up satellite archival tag.

The preprogrammed pop‐up date for all PSATs was 31 August 2018. When PSATs detached, they floated to the surface and initiated the transmission of a summary of recorded data through the Argos satellite network, as well as current GPS positions. Popped‐up PSATs were recovered using chartered industry vessels equipped with CLS ARGOS RXG‐134 goniometers (CLS America Inc., Lanham, MD, USA). The goniometer directional antenna detects pings broadcast by floating PSATs within a range of 5 nm, providing an angle relative to the vessel's bow as well as the relative strength of the signal, used as a proxy for distance.

2.2. PSAT data and tilt estimated from acceleration data

PSATs recorded four separate metrics at a 5‐s resolution throughout the immersion period: depth (resolution = 0.5 m, range of accuracy is ±0.34 m at the surface and ±5.38 m at 440 m), temperature (resolution = 0.05°C, ±0.1°C accuracy), light intensity levels (±5 × 10⁻¹² W cm⁻²) and tri‐axial acceleration (range: −2 to 2 g ± 0.05 g). When a fish is motionless, the positively buoyant PSAT floats freely above the fish and records either the full force of gravity (1 g = 9.81 m s⁻²) or a small inclination along the three orthogonal axes resulting from ambient currents (e.g. tidal currents). When a fish becomes active and swims, the accelerometer detects changes in inclination in one or more of the axes (Figure 2). As a fish moves, the PSAT tilts backwards, with tilt angle increasing towards the horizontal position as speed increases, as previously demonstrated for Pacific halibut (Nielsen et al., 2018). In accordance with conclusions from Nielsen et al. (2018) on acceleration data from PSATs, acceleration values from each of three axes (A _X , A _Y , A _Z ) contribute to the calculation of the tilt angle, expressed in degrees, using the formula developed by Pedley (2013):

$$\mathbf{\text{Tilt angle}}=\mathbf{\text{acos}}\left(\frac{\mathit{Az}}{\sqrt{\mathit{A}{\mathit{x}}^{\mathbf{2}}+\mathit{A}{\mathit{y}}^{\mathbf{2}}+\mathit{A}{\mathit{z}}^{\mathbf{2}}}}\right)\times \frac{\mathbf{180}}{\mathit{\pi }-\mathbf{180}}$$ (1)

FIGURE 2.

Representation of a pop‐up satellite archival tag (PSAT) floating while the fish is motionless and tilted back when the fish is in motion.

2.3. Tilt as a proxy for activity

Tilt values obtained for individual fish are distributed from 0 to 180° and typically comprise a combination of (1) small values corresponding to static positions, slow movements and a portion of those values consistent with ambient currents, such as the tide effect; and (2) larger values that corresponded to stronger movement at different swimming speeds. A tilt value of 0° implies the tag is floating vertically above the fish. We fit a Gaussian mixture model assuming two modes using the R package ‘mixtools’ (version 1.2.0) (Benaglia et al., 2009) in R (version 4.1.2) (R Core Team, 2021) to calculate a threshold to separate values corresponding to inactivity or current effects from those corresponding to fish movement (Figure 3) (Nielsen et al., 2018). Based on values observed across the 25 individuals, and the two moored tags which exclusively reflected tilt values associated with ocean currents, a threshold value of 38.5° was assumed to separate tilt values inferred to reflect inactivity (0–38.5°), which represented 85.44% [standard deviation (SD) = 16.57%, n = 25] of all observations, from tilt values assumed to reflect fish activity (range: 38.5–180°, mean: 64.59°), which represented 14.55% (SD = 16.57%, n = 25) of all observations.

FIGURE 3.

Distribution of tilt values for pop‐up satellite archival tags (PSAT) attached to Atlantic halibut using a dart and tether mechanism, as well as for the two moored PSATs from the Gulf of St. Lawrence. Tilt values were calculated from archived data (5‐s resolution). The threshold (38.5°) used to divide static mode from active mode is represented by the black vertical line.

Due to the inability to quantitatively account for the contribution of currents to tilt values, we focused our analysis on a categorization of the data, with tilt values categorized (5‐s period) as either active = 1 or inactive = 0 based on the activity threshold to create continuous activity time series for each individual. We note that the inactive state includes slow movements that cannot be distinguished reliably from the effect of currents. The sum and proportion of active moments comprising those time series were calculated for different temporal resolutions (15 min, 60 min and 1 day) to facilitate the interpretation of activity patterns in subsequent analyses.

2.4. Effects of tidal currents on tilt data

Plots of the tilt data from the moored tags and from two halibut with low activity level clearly showed periodic variation consistent with tag movement consistent with the tidal cycle (Figure 4). The vast majority (98.5%) of the tilt values recorded by moored tags were smaller than our calculated threshold value of 38.5°, indicating that the active mode of values for halibut can be attributed to moderate to fast swimming speeds (Nielsen et al., 2018). However, we were unable to distinguish between tag tilt values consistent with currents and slow movement speeds and therefore could not evaluate slow movements of halibut in this study.

FIGURE 4.

Two‐dimensional time plots of tilt values from the pop‐up satellite archival tags (PSAT) of two individual Atlantic halibut and a moored tag between September 2017 and August 2018, where the influence of tidal currents on tilt is evident, as represented by diagonal patterns in the tilt values. Hourly mean tilt values are presented using a continuous colour gradient, where dark colours represent relatively low tilt values (assumed to be associated with immobility) and light colours (light blue through yellow) represent relatively high tilt values (assumed to be associated with activity). The white dots indicate the times of high tide and low tide. The beginning of each time period is represented by the dash, and time is in Atlantic Standard Time (AST).

2.5. General population patterns in activity

We used dynamic factor analysis (DFA) to identify and describe general patterns in activity level in GSL halibut. DFA, an analogue of factor analysis but for time series, is a multivariate data‐reduction method that is used to detect patterns in time series. It aims to identify a small number of dimensionless hidden (underlying) common patterns (hereafter common trends) shared across a group of time series. These common trends are unitless and ‘mimic’ the general dynamic that can be found in the group of time series. A factor loading on each common trend is estimated for each original time series (individual halibut activity series) and represents the degree of association of the time series with that trend. For our data, the activity level of each halibut was modelled as a linear combination of common trends and factor loadings plus some noise (error) that is naturally present.

Daily activity level ($y_t$) was estimated by summing the number of observations classified as active for each day (t corresponds to 1 day). Following Zuur, Fryer, et al. (2003), each resulting series was standardized (i.e. z‐scored, $y_t^{*}$) by subtracting the mean $\left(\overline{y}\right)$ and dividing by the standard deviation ($\sigma$). A matrix of n × T (where n is the number of individuals and T is the total number of days of the series) was then created.

In this analysis, only the data from individuals for which the full annual data series were available (n = 20) were considered. We modelled trends in standardized daily activity level as a combination of m common trends (${\alpha }_t$) and both process errors ($w_t$, random fluctuation in the system dynamics) and observation errors (${ϵ}_t$, observation errors). This was performed using the MARSS package (Holmes et al., 2021, 2022, 2024) in R, version 4.1.2 (R Core Team, 2021), with the term form specified as ‘dfa’ and a specified number of common trends (m). The model was constructed as follows:

$${\mathit{y}}_{\mathit{t}}^{*}={\mathit{Z\alpha }}_{\mathit{t}}+{\mathit{ϵ}}_{\mathit{t}}\mathbf{\text{where}}{\mathit{ϵ}}_{\mathit{t}}~\mathbf{MVN}\left(\mathbf{0},\mathit{R}\right)$$

$${\mathit{\alpha }}_{\mathit{t}}={\mathit{\alpha }}_{\mathit{t}-\mathbf{1}}+{\mathit{w}}_{\mathit{t}}\mathbf{\text{where}}{\mathit{w}}_{\mathit{t}}~\mathbf{MVN}\left(\mathbf{0},{\mathit{Q}}_{\mathit{t}}\right)$$ (2)

where Z is a $20\times m$ matrix of factor loadings on the common trends, which is a measure of the relationship between an individual's activity‐level time series and a common trend. ${ϵ}_t$ is an n × T column vector of the non‐process errors and $w_t$ is an m × T matrix of the observation errors; it is assumed that both are independent multivariate normally distributed with a mean of zero and respective covariance matrices. The variance–covariance matrices R (observation variance) and Q (process error) were, respectively, set as diagonal and unequal, and identity.

We compared DFA models comprising either 1, 2 or 3 common trends. Akaike's information criterion (AIC), calculated as twice the difference between the log‐likelihood value (measure of fit) and the number of parameters (penalty), was used to compare the relative evidence for each model (Holmes et al., 2022; Zuur, Fryer, et al., 2003). Apart from the AIC, we used the number of parameters (penalty) to make an arbitrary decision for the best model considering that increasing the number of common trends in the model improves the fit as the trend number (m) approaches the number of individuals (n); it also increases the number of parameters to be estimated, leading to an increasingly high penalty (Zuur, Tuck, & Bailey, 2003). The goal is to identify the minimum number of trends (m), without losing too much information. Model validation included plotting of residuals, assessing the adequacy of model fits and considering the ecological interpretation and significance of the results.

Factor loadings for each trend were rotated using the varimax rotation function in MARSS package to improve their interpretability. An arbitrary cut‐off level for the loadings was set at 0.025 (moderate) and at 0.050 (high) to determine the level of association of a time series with the common trend (${\alpha }_t)$ (Zuur, Fryer, et al., 2003). We hypothesized that halibut released together or nearby should be more likely to be similarly associated with the common trends. Therefore, we considered the five broad geographical regions where tagging was carried out, as defined by Gatti et al. (2020) (Figure 3).

Separate from the DFA and to verify the effect of season on activity level, a repeated‐measures analysis of variance (ANOVA) was conducted using seasonal mean of activity level for each individual as the dependent variable and season as the independent categorical factor; individuals with incomplete series where removed. After the ANOVA, a post hoc paired t‐test with Bonferroni correction was used to compare each pair of seasons.

2.6. Characterization of periodic activity

Following the approach of Scott et al. (2016) to visualize periodic variation in activity on different time scales, activity‐level plots and two‐dimensional (2D) time series plots were constructed. The activity‐level plots allowed for the visualization of patterns in activity levels at the seasonal and monthly scales within groups of tagged individuals. 2D time series plots for day of the year (x‐axis) and hour of the day (y‐axis), with colour used to represent mean activity level, were used to detect potential diel periodicity patterns over the year for each halibut. Both plots were constructed using hourly proportion of active moments.

Based on the results from the 2D time plots, we also investigated a potential link between ambient light and activity level by utilizing a non‐linear regression function in R. Locally estimated scatterplot smoothing (LOESS) curves were fitted to the hourly mean light intensity and the hourly proportion of activity data. This was tested for each individual and group (n = 25), for both sunlight hours and the entire 24‐h period.

Additionally, we sought to characterize if activity frequency varied between day and night for each halibut. Because halibut were sometimes more active in daytime and sometimes during night‐time, a comparison of the summarized day versus night values would not have been informative. Instead, we created a metric to measure relative change in activity between these periods. The beginning of day and night was defined using the local time (Atlantic Standard Time) of sunrise and sunset from the National Council of Canada. We measured the association activity to the defined day and night periods for each individual halibut h by partitioning the variance using the following R ²‐based calculation based on the frequency of active moments in distinct 15‐min periods:

$$R_h^2=100\times \left(1-\frac{\sum _{t=1}^{T_h}\sum _{d=1}^2\sum _{i_{t,d}=1}^{I_{t,d}}{\left(y_{i_{t,d},t,d,h}-{\overline{y}}_{t,d,h}\right)}^2}{\sum _{t=1}^{T_h}\sum _{i_t=1}^{I_t}{\left(y_{i_t,t,h}-{\overline{y}}_{t,h}\right)}^2}\right)$$ (3)

where

$y_{i_{t,d},t,d,h}$ is the sum of active moments in period $i_{t,d}$ occurring on consecutive day t, during diel portion d (daytime, d = 1; night‐time, d = 2) for halibut h;

$y_{i_t,t,h}$ is the sum of active moments in period $i_t$ occurring on consecutive day t for halibut h;

${\overline{y}}_{d,t,h}$ is the mean number of active moments per period on consecutive day t, diel portion d for halibut h;

${\overline{y}}_{t,h}$ is the mean number of active moments per period on consecutive day t for halibut h;

$T_h$ is the total number of days of monitoring for halibut h; and

$I_{t,d}$ and $I_t$ are, respectively, the total number of periods for day t and diel portion d, and for day t only.

An overall, across individuals, measure of explained variance, $R^2,$ is obtained by adding a summation over individual halibut to Equation (3):

$$R^2=100\times \left(1-\frac{\sum _h^H\sum _{t=1}^{T_h}\sum _{d=1}^2\sum _{i_{t,d}=1}^{I_{t,d}}{\left(y_{i_{t,d},t,d,h}-{\overline{y}}_{t,d,h}\right)}^2}{\sum _h^H\sum _{t=1}^{T_h}\sum _{i_t=1}^{I_t}{\left(y_{i_t,t,h}-{\overline{y}}_{t,h}\right)}^2}\right)$$ (4)

where H is the total number of halibut in this study.

R ² values are expressed in percentage and correspond to the variation in activity level that was attributable to the periodic intervals of day and night.

3. RESULTS

3.1. General population patterns in activity

The DFA model with three trends ($m$ = 3) was selected as the most suitable model to represent the general patterns observed in the activity level of tagged halibut (Table 2). We note however that the first trend (trend A) is nearly identical to that estimated for models 1 and 2, and the second trend (trend B) is very similar to the second trend from model 2 (Figure 5; Figure S1). Trend A was considered the strongest of the three, because it was present in models 1, 2 and 3, whereas trend B and trend C represented weaker trends, hence weaker drivers of the patterns (Figure 6) (Zuur, Fryer, et al., 2003). A positive value of the common trend indicates an upward shared pattern in the activity‐level time series, and a negative value indicates a downward shared pattern. The strength of association of individual halibut time series to a trend is indicated by the factor loading, with a negative loading indicating that the data follow a pattern opposite the common trend.

TABLE 2.

Comparison of dynamic factor analysis models incorporating different (1;3) common trends (m).

Model	Trends (m)	Log likelihood	AICc	Δ i	w i	Number of parameters
1	1	−9028.759	18,158.196	1101	1.108 e‐239	50
2	2	−8715.509	17,580.502	522.8	3.087 e‐114	74
3	3	−8430.609	17,057.660	0.0	0	97

Note: Δ _i is the Akaike information criterion of model i minus the minimum AICc of the considered models, and w _i is the Akaike weight for model i.

Abbreviation: AIC, Akaike information criterion.

FIGURE 5.

Individual standardized daily mean activity level (grey dots) and predictions from the dynamic factor analysis (DFA) models (coloured lines) for 20 Atlantic halibut tagged in the Gulf of St. Lawrence, Canada (panels). Predictions from the three models are distinguished using different colours: orange (model 1), blue (model 2) and red (model 3). The geographical region of each individual is identified by the colour used to display the pop‐up satellite archival tag ID in the top left of each panel.

FIGURE 6.

Trends obtained by the model containing three common trends (i.e. model 3). Common trends are unitless. The coloured boxes correspond to the delimitation of the different seasonal periods.

Based on the overall patterns observed in each trend, we denote three distinct periods of activity across the year, corresponding to (1) September–December (108 days, hereafter ‘fall’); (2) January–mid‐April (106 days, hereafter ‘winter’); and (3) mid‐April–August (138 days, hereafter ‘spring–summer’). Based on these characteristics and patterns in the trends, we label the trends as follows: trend A, ‘early winter peak, high activity’; trend B, ‘late winter peak, moderate activity’; and trend C, ‘spring–summer increasing activity’.

Common trends A (‘early winter peak, high activity’) and B (‘late winter peak, moderate activity’) were characterized by a low activity level during the fall, followed by an increase in activity level across winter, with trend A peaking earlier and at higher levels relative to the highest peak of trend B (Figure 6). After mid‐April, trends A and B exhibited a decrease in global activity level characterized by small time scale (weekly) variation, but with an overall higher level of activity compared to the fall. Trend C (‘spring–summer increasing activity’) was characterized by the highest activity level at the beginning of the tagging period, followed by two small peaks of activity in January and mid‐February, and a progressive increase after March that reached its highest level through spring and summer. Each common trend showed a different timing in the peak of activity level.

Individual halibut associated differently with the three common trends based on factor loadings of the DFA (Figure 7). Overall, differences between halibut in association with the trends were partially related to sex and tagging locations. Trend A (‘early winter peak, high activity’) was positively related to the individuals from the Gaspe Peninsula (five out of seven individuals), with two in particular (17P0166, 17P0170) exhibiting a stronger association with that trend, meaning high factor loading values and therefore a high activity level during winter. Those two halibut were the only individuals identified as males using the methods of Marshall et al. (2023). The only individual tagged off of Cape Breton was only positively related to trend A. Three of four individuals from Anticosti Gyre were positively associated with trend B (‘late winter peak, moderate activity’). Five of six individuals from the Esquiman Channel were strongly and positively associated with trend B, and four were negatively associated with trend C (‘spring–summer increasing activity’), indicating that their time series followed the opposite pattern from the trend. One individual from Prince Edward Island was associated most strongly with trend B, and the other individual associated most strongly with trend C.

FIGURE 7.

Individual factor loadings for the three common trends from the selected model. The vertical dashed lines show the arbitrary cut‐off level of (±) 0.025 and 0.05 based on Zuur, Fryer, et al. (2003) to assess the extent to which an individual is associated with each trend. The colour of the pop‐up satellite archival tag (PSAT) ID corresponds to the tagging location. The fork length of each individual is indicated on the left side of the panel. Individuals were grouped by sex (female, male, unknown) based on the results of Marshall et al. (2023) using either genetics (g) or behavioural analysis (b) for each individual.

Overall, halibut activity exhibited three main seasonal periods captured by three common trends that peaked at different times. Individual fish varied in how strongly they followed these trends, as reflected in their factor loadings, with some associations influenced by sex and tagging location.

Using the percentage of activity level presented in Table 3, it was possible to identify the differences from individuals tagged in different regions and across the three seasonal periods. Individuals tagged in the Esquiman Channel exhibited the highest percentage of activity over the year (22.08%, SD = 19.27, n = 8) (Table 3). Individuals tagged off of the Gaspe Peninsula were active 10.63% (SD = 15.46, n = 7) of the time over the year and exhibited the highest increase in activity during the winter period, which was primarily attributable to two individuals (Figure 8; PSAT ID 17P0166, 17P0170). Halibut from the Anticosti Gyre were the least active (9.85%, SD = 10.40, n = 6). The individual tagged off of Cape Breton was active 11.89% over the year and the three individuals tagged off of Prince Edward Island 13.15% of the time (SD = 14.58, n = 3).

TABLE 3.

Annual and seasonal mean percentage of active moments (%) based on 15‐min intervals with standard deviation (±) for halibut from each region where more than one halibut was tagged.

Regions	Number of halibut	Tagging site		Annual	Seasonal
		Depth (m)	Temperature (°)		Fall	Winter	Summer
Anticosti Gyre	6	112.16	4.01	9.85 ± 10.40	5.86 ± 4.96	10.92 ± 11.28	10.14 ± 10.51
Cape Breton	1	89.38	4.63	11.89	6.65	16.31	11.72
Gaspe Peninsula	7	96.57	3.69	10.63 ± 15.46	3.93 ± 4.67	14.56 ± 21.00	10.10 ± 11.38
Esquiman Channel	8	123.25	4.39	22.08 ± 19.27	16.02 ± 14.22	24.79 ± 19.97	22.75 ± 20.20
Prince Edward Island	3	19.33	3.13	13.15 ± 14.58	5.96 ± 4.65	14.36 ± 15.60	16.36 ± 2.54

Note: Tagging site depth was recorded directly from the tagging vessel, and temperature was calculated based on the first hour after tagging. All values represent averages for individuals within the same region, except for Cape Breton.

FIGURE 8.

Activity plot for 25 halibut tagged in the Gulf of St. Lawrence (GSL) and two moored tags over the deployment period between September 2017 and August 2018. Daily proportion of active values is presented using the continuous gradient of colour, where dark colours represent a relatively low proportion of active values and light colours represent a relatively high proportion of active values. The colour code for the regional designation of PSAT (pop‐up satellite archival tag) IDs is presented in the legend. Note that elevated tilt values were observed for the moored tag 17P0198 at the beginning of February; however, their cause is unknown.

Although environmental conditions such as depth and temperature differed among tagging regions (Table 3), these differences alone cannot explain the observed variability in halibut activity. Considerable variability was also observed within regions, suggesting the influence of additional factors, including seasonal changes on activity level. Across all regions, ANOVA revealed a significant effect of season on activity levels (F(2, 44) = 13.17, p < 0.001) and also showed that mean activity levels for all tagged halibut differed significantly among all seasons (p < 0.001). Specifically, activity was lowest in fall, higher in summer and highest in winter.

3.2. Individual variability in activity level

We observed high individual variability in activity level, which was particularly evident in the proportion of active and passive moments. The most active individual (16P2437) was active 34.71% of the time, whereas the least active individual (17P0087) was active only 3.83% of the time (Figure 8). Between these two extremes of activity level, a gradient was observed, with a mean of 13.29% (SD = 8.36, n = 25), indicating that halibut spend a considerable amount of time inactive. Although regional differences appear to emerge, it is important to note that within the regions, some individuals, based on the DFA results, exhibit activity profiles that are more similar to those of individuals in other regions than to those within their own region (e.g. Figure 12; 17P0165 and 17P0136). The results indicated strong individual differences in activity, with each halibut exhibiting a distinct yearly activity pattern. Both mean activity levels and temporal variability varied substantially among individuals.

FIGURE 12.

Locally estimated scatterplot smoothing (LOESS) model for the relation between hourly proportion of activity and hourly mean light intensity for the 25 halibut over the entire deployment period. The smooth trend lines based on the LOESS regressions are shown. The left panel includes only the data collected during daytime, and the right panel includes both daytime and night‐time data.

3.3. Diel patterns of activity levels

There were compelling diurnal and nocturnal patterns of activity spanning several days in the time series of individual halibut, with a strong demarcation based on synchronization with sunrise and sunset times (Figure 9). In most instances, these diel periodic patterns were associated with higher activity during daylight hours over a succession of days (Figure 10; 16P2437 from November to December); however, in others a period of general low nocturnal activity was associated with complete inactivity in the daytime (17P0150 in October). Instances of moderate nocturnal periodic activity were less common (e.g. 17P0150 in June). Interestingly, some individuals exhibited both periodic nocturnal and diurnal activity at different times throughout the year (Figure 11; 17P0086 and 17P0148); however, the seasonal timing of these periods varied among individuals. Each individual seemed to exhibit a unique combination of periodic patterns (Figure 11). Overall, halibut exhibited distinct diel activity patterns synchronized with sunrise and sunset but not light intensity, as revealed by the LOESS model (Figure 12); LOESS models for each individual yielded a similar result (see Figure S2). These results indicate that light intensity is not a direct driver of halibut activity, but the 2D time series plot shows a clear increase in activity at a defined moment throughout the year.

FIGURE 9.

Two‐dimensional time plot of tilt values for one individual [PSAT (pop‐up satellite archival tag) ID 16P2437] between October 2017 and August 2018. Hourly mean tilt values are presented using a continuous colour gradient, where dark colours represent relatively low tilt values and light colours represent relatively high tilt values. The black lines correspond to the sunrise and sunset hours. The two zoomed panels show examples of the untransformed tilt data used to construct the two‐dimensional time plot; the shaded areas correspond to the night. The upper zoomed panel shows a period of nocturnal activity during which 94.76% of all active moments were during the night, and the lower zoomed panel shows a period of intense diurnal activity during which 89.96% of all active moments were during daylight. The beginning of each time period is identified by a dash, and all times are given in Atlantic Standard Time (AST).

FIGURE 10.

Two‐dimensional time plots of tilt values for four selected halibut tagged in the Gulf of St. Lawrence (GSL), showing different profiles of periodic activity. (a) Individual exhibiting recurrent daytime activity, while mostly inactive at night; (b) individual exhibiting recurrent night‐time activity levels generally higher than those observed in the daytime; (c) individual exhibiting variable periodical activity during the day and the night; (d) individual mostly inactive throughout the year with little apparent periodic activity. Hourly mean tilt values are presented using a continuous colour gradient, where dark colours represent relatively low tilt values and light colours represent relatively high tilt values. The thick black lines correspond to the sunrise and sunset hours. The beginning of each time period is identified by a dash, and all times are given in Atlantic Standard Time (AST).

FIGURE 11.

Activity plots representing the sum of active moments (5‐s periods) per 15 min for daytime (grey line) and night‐time (blue line) for each halibut tagged. R ² values expressed in percentage are provided in each panel and correspond to the variation in activity level that is attributable to the distinction between day and night.

Based on visual examination of the 2D time plot and the activity plots (Figures 10 and 11), daylight activity during the year was identified at times for 12 individuals (48% of all halibut). This daytime periodic pattern was observed between spring and summer for 11 individuals and between fall and winter for 4 individuals. The nightly periodic pattern was observed in nine individuals (36% of all halibut). The nightly periodic activity pattern was equally observed between fall and spring/summer with little evidence of occurrence throughout winter. The nightly pattern was observed throughout the deployment time series for only one halibut (17P0135; Figure 10b), but the series was truncated to a duration of 6 months due to the early release of the tag, so it is unknown if the nightly activity pattern would have persisted through the summer.

Values of R ² index, which represents the variation in activity level in the series that can be attributed to the day and night cycle, exceeded 10% for 16 individuals (Figure 11). The R ² index had a mean of 18% for all individuals, with individuals ranging between 6.28% (17P0087) and 32.00% (16P2437). Figure 11 highlights two key observations. First, the contrasting day (grey) and night (blue) activity lines reveal pronounced diel periodicity in some individuals at certain times of the year (e.g. 17P0170 during summer). In addition, the activity lines illustrate that each fish exhibits a distinctly individual activity pattern. Some regional differences were also observed, with Esquiman Channel characterized by the highest mean R ² index value of 21.81% (SD = 6.67%, n = 8), and Anticosti Gyre, Gaspe Peninsula and Cape Breton with the lowest values of 9.19% (SD = 3.77%, n = 6), 11.01% (SD = 3.45%, n = 7) and 11.05%, respectively. Prince Edward Island had an intermediate R ² index value of 13.22% (SD = 4.02%, n = 3).

Overall, several halibut exhibited diel activity patterns during parts of the year, but the timing of these patterns varied widely. There was little consistency, either within individuals over time or across individuals, in whether periodic activity was more present during the day or at night.

4. DISCUSSION

The primary aim of the present study was to evaluate both coarse and fine‐scale individual activity patterns of Atlantic halibut throughout an entire seasonal cycle. The study also aimed to investigate evidence for contingents in behaviour within the GSL population and identify specific behaviours based on the resolution of the available data. Although some common general patterns in activity were observed, the results revealed a surprisingly high level of individual variability in activity, with each halibut exhibiting a unique temporal activity profile over the year. Activity widely varied among individuals in terms of mean level through the year and in terms of temporal variability in observed patterns. A diel pattern in activity was evident in several individuals during parts of the year; however, there were differences among individuals in the timing of specific diel activity patterns, and there was a lack of consistency over time within individual and generally between individuals as to whether halibut were more active during the daytime relative to night, or vice versa. Overall, individual variability in activity dominated over any common patterns that were detected. As a result, no evidence suggests it was possible to assign individual halibut to distinct activity contingents. However, common to all individuals, activity levels were generally low [active on average 13.29% of the time (SD = 8.36, n = 25)] in these tagged individuals, suggesting that halibut spend a substantial portion of time at rest and rely on burst swimming interspersed with recovery periods, which is consistent with behaviours expected for flatfish (Duthie, 1982; Tuene & Nortvedt, 1995).

4.1. Individual variability in activity level

Behavioural variability is common across animal species, including fishes, and is important for understanding changes in population dynamics (Baptista et al., 2020; Bolnick et al., 2003; Mittelbach et al., 2014). Although fish behaviour was once seen as simple, modern studies show significant individual differences both within and among individuals over time (Benhaïm et al., 2023; Bolnick et al., 2003; Kristiansen & Fernö, 2007). Activity level has recently been used as a proxy for fish behaviour and personality (Cote et al., 2010), and linked to intrinsic traits such as size at a given life stage (e.g. Ahti et al., 2020; Roy & Bhat, 2018), metabolic rate (Careau et al., 2008; Nespolo & Franco, 2007; Rupia et al., 2016) and state of hunger (Macquart‐Moulin et al., 1991; Miyazaki et al., 2000; Stoner, 2003). In addition to intrinsic considerations, extrinsic factors can play an important role in modulating activity level (Archard & Braithwaite, 2011; Stoner, 2004). Because most fish are ectotherms, many studies have reported an effect of temperature on individual activity (Olchena et al., 2017; Stoner et al., 2006; Volkoff & Rønnestad, 2020), including in Pacific halibut where a positive link between temperature and activity was observed within the preferred temperature range (Stoner et al., 2006). Competition can also modulate activity level (Zhdanova & Reebs, 2005); for instance, an increase in stocking density led to increasing levels of foraging activity in farmed Atlantic halibut (Kristiansen & Fernö, 2007; Stoner & Ottmar, 2004). Because many species of fish are visual predators, light also widely influences fish activity (Reebs, 2002; Stoner, 2004; Zhdanova & Reebs, 2005), including Pacific halibut (Stoner, 2003, 2004). Importantly, individuals may respond differently to the same environmental conditions due to differences in coping capacity or behavioural flexibility (Kristiansen & Fernö, 2007), which aligns with the substantial individual variability observed in our study.

In the present study, the experimental design minimized the influence of intrinsic factors. By tagging large (>120 cm) and most likely mature individuals, the influence of body size or maturity status could not be formally linked to activity level (see Figure S3). Furthermore, because halibut were immediately released after tagging, other intrinsic characteristics such as metabolic rate, condition or age could not be measured, making it impossible to investigate potential links between activity level and intrinsic characteristics. An exception was sex, which was identified by Marshall et al. (2023). The group of tagged individuals included two halibut (17P0166 and 17P0170 identified as males by Marshall et al., 2023), which exhibited much higher and sustained activity levels during winter compared to any other individuals. However, the number of tagged males was too low to attribute this pattern to an effect of sex on activity level. Similar to intrinsic factors, most extrinsic factors, such as halibut population density (e.g. competition, reproduction status) or prey availability experienced by a given individual through the year, were unknown. Temperature, light, depth and acceleration data were recorded at high resolution by PSATs, but characteristics of the bottom water layer of deep channels, where halibut distribute from late fall to spring, are relatively uniform. As a result, during the largest part of the year individuals experienced minimal environmental variation compared to the diversity of activity patterns exhibited (see Figure S4). Interpretation is further complicated by the fact that halibut are known to undertake spawning increases during winter, which are associated with pronounced changes in depth and elevated activity levels (Marshall et al., 2023). During migration periods, high activity is often observed alongside gradual depth changes (Gatti et al., 2020). In contrast, during summer, activity levels may also be high while depth remains largely constant, likely reflecting foraging or hunting behaviour. These overlapping behavioural signatures create a complex pattern that is difficult to disentangle, and it was not possible to effectively link the high individual variability in activity to either intrinsic factors or characteristics of the environment encountered by halibut.

Differences in the behaviour of a given individual relative to others, which remain stable in time, are often referred to as ‘personality’ or ‘behavioural style/syndrome’ (Budaev & Brown, 2011; Dall et al., 2004; Réale et al., 2007). Such patterns have been described in fishes, with activity level considered a main component characterizing animal personality (Budaev & Brown, 2011; Dingemanse et al., 2010; Réale et al., 2007). Several studies have shown that the personality and relative differences in activity level among individuals remain stable across different environments and stress levels (e.g. Conrad et al., 2011; Mittelbach et al., 2014; Rupia et al., 2016). In the present study, drivers of the wide individual variability in activity level of the 25 halibut considered could not be easily identified, which may reflect high individual variability in personality and behavioural repertoire.

4.2. Common patterns in activity level

Similarities and differences in migration patterns and strategies among individuals of a given population are often used to designate different groups or contingents (Conrad et al., 2011; Secor, 2015). Migratory contingents have been identified in a wide variety of fish taxa, ranging from large pelagics like Atlantic bluefin tuna (Thunnus thynnus) (Galuardi et al., 2010) to demersal species like Atlantic Cod (Gadus morhua) (Le Bris et al., 2013) and estuarine species like striped bass (Morone saxatilis) (Secor, 1999). In the case of Atlantic halibut in the GSL, the system‐wide study by Gatti et al. (2020) also suggested the presence of several migration contingents.

In this study, we examined whether the subset of fish geolocated by Gatti et al. (2020) exhibited shared activity patterns based on data from the PSAT accelerometer. Results from the DFA in the present study suggested a peak in activity during winter as the three common trends showed an increase at some point during that season. Previous studies from Gatti et al. (2020) and Marshall et al. (2023) identified winter as the spawning period, suggesting that peaks in activity observed via accelerometry could be linked to spawning behaviour. Halibut are known to exhibit a series of abrupt spawning increases in the water column over a period of several weeks (Le Bris et al., 2018; Marshall et al., 2023; Seitz et al., 2005). From all available electronic tagging data, Marshall et al. (2023) concluded that the spawning season for GSL Atlantic halibut occurs between January and April, with a peak in mid‐February, corresponding to the observed peak in trend A (‘early winter peak, high activity’) that was observed in the current study by 10 February. Trend B (‘late winter peak, moderate activity’), which had activity peaking in early March, may reflect individuals that spawn over a longer period or spawn later in the season, a phenomenon observed in several species (Morgan et al., 2013; Wright & Trippel, 2009), including in flatfishes (Morgan, 2003). Trend C (‘spring–summer increasing activity’), showing an increase from early December, may be linked to early spawning or to a late migration towards the spawning ground, as noted by Le Bris et al. (2018). During spring–summer, halibut return to their feeding grounds. Foraging involves potential changes in depth (Le Bris et al., 2018) and various swimming behaviours (Kristiansen & Fernö, 2007), which may explain the diverse activity levels reflected in the common trends in spring–summer.

Although the DFA from the current study yielded three common trends in activity level of halibut tagged in GSL over a seasonal time scale, the level of individual variability was surprisingly high, making it nearly impossible to identify contingents based on activity level. Such level of individual variability could be attributable to the fact that halibut is a solitary top predator. It does not need to exhibit group‐level antipredator behaviour/activity such as schooling, or group hunting behaviours, or to follow specific migration corridors to the spawning grounds. This high variability instead suggests variation in foraging and migrations behaviours. This may be similar to other non‐schooling top predators such as the swordfish (Xiphias gladius), for which PSAT tagging revealed high variability in vertical movements of diel patterns (e.g. Dewar et al., 2011). For such species, we argue that at the very least, a larger sample of tagged individuals would be needed to reveal common activity patterns and strategies, especially given that the number of tags available in a study is often the primary factor preventing population inferences (Griffiths et al., 2018).

In addition to behavioural considerations, another factor that may blur common trends is the free‐floating nature of tags used in the present study, which may result in the underestimation of activity in specific situations. For instance, when an individual exhibits slow movements such as slow‐pace swimming or gliding (Gleiss et al., 2011; Scott et al., 2016), the PSAT likely maintains its nearly vertical orientation, confounding slow movements with those caused by natural currents.

4.3. Periodic behaviour

Despite constraints linked to free‐floating tags, Nielsen et al. (2018) concluded that acceleration metrics recorded by PSATs could properly detect most types of activities in halibut, particularly periodic behaviours. Fishes are known to follow a circadian cycle (Zhdanova & Reebs, 2005) and express behaviour after a clear periodicity to adapt to a changing environment (Hunter et al., 2004; Scott et al., 2016; Shepard et al., 2006). Many factors can drive the prevalence of periodic behaviours, as well as their period and duration. In various species, periodic activity that is linked to feeding habits occurs approximately at the same time every day (Le Bris et al., 2013; Shepard et al., 2006), including in flatfish (Nielsen et al., 2018; Scott et al., 2016), which tend to lay on the seabed motionless while digesting food after a period of active hunting (Gibson et al., 2015). A daily periodicity in behaviour, whether nocturnal or diurnal, is often associated with circadian cycles in the movements of their prey or vertical distribution in the water column (Baird et al., 2001; Croll et al., 1998). Periodic behaviours can also be attributable to natural cycles in factors other than daylight, occurring over shorter (e.g. tides) and longer (e.g. lunar phases) periods (Cartamil & Lowe, 2004; Hunter et al., 2004; Wilson et al., 1993). In high‐latitude environments, where seasonality is strongly contrasted, periodic activity typically varies with the seasons in response to factors like prey availability, consumption and distribution. (Godo & Michalsen, 2000; Schwalme & Chouinard, 1999; Stensholt, 2001). This structures basic activities such as foraging, migrating and breeding to definite periods over the seasonal cycle (Navarro & Gutiérrez, 1995). For example, periodic activity linked to foraging may decrease or completely stop during the spawning season when individuals exhibit a high proportion of courtship and aggressive behaviours associated to mating (Carvalho et al., 2003; Scott et al., 2016; Stoner & Ottmar, 2004).

In the present study, we observed periodic activity in 15 of the 25 tagged halibut. Moreover, periodic behaviours exhibited by these individuals were observed during various portions of the seasonal cycle. The most common type of periodic activity observed followed a diel periodicity associated with daytime, which was most prevalent in spring/summer. This timing corresponds to the period when individuals were likely to have returned to their presumed feeding grounds (Le Bris et al., 2018). This suggests that a substantial portion of the periodic activity observed in this study may be associated with foraging. In addition to the daytime periodic activity pattern, night‐time periodic activity was observed in nine individuals. However, the timing over the seasonal cycle and the short duration of this night‐time pattern were highly variable, making it difficult to speculate on the drivers of observed night‐time activity.

Focusing on the daytime periodic pattern, the high variability we observed suggests substantial individual‐level variability in feeding strategies among halibut from the GSL. Our results also suggest that a large number of individuals exhibit shifts in foraging activity/strategy through the summer, which may reflect the capacity to adapt to changing prey distribution or shift to different prey taxa to optimize food acquisition. Even during the foraging season, daytime periodic behaviour was observed only in a minority of tagged individuals. Flatfish are known as ambush predators, and it is possible that a high proportion of individuals may adopt a sit‐and‐wait strategy to capture a limited number of large prey (Gibson et al., 2015; Nielsen et al., 2018), which would explain the absence of periodic patterns and relatively low activity levels in some individuals.

5. CONCLUSION

The findings outlined in the present study illustrate that although identifying contingents in behaviour and personality traits in marine fish populations is challenging, the use of acceleration time series recorded by PSATs allows investigation of the behavioural variability that exists in a given population. We detected periodic activity patterns in some individual halibut profiles and revealed unexpectedly high variability in activity level among individuals. Although concurrent analysis of the full suite of PSAT data (depth, temperature, light and acceleration) may yield insights into potential drivers, establishing and describing activity patterns represented a fundamental first step. The observed diversity of activity levels and patterns challenges the common assumption that halibut, like most flatfish, is a passive bottom‐dwelling species (Kristiansen et al., 2004; Nilsson et al., 2010). Instead, our observations align with a growing body of literature supporting the concept that fish exhibit considerable individual variability in their behaviours, activity levels and responses to stress or perceived threats (Budaev & Brown, 2011; Cote et al., 2010; Robert et al., 2013), which opposes the traditional preconceived idea that fish populations consist of individuals that can be considered as ‘clones’.

Our findings are important in the context of halibut stock management as the failure to account for differences in animal personalities may lead to bias in the estimation of population size and structure (Biro, 2013; Biro & Dingemanse, 2008; Merrick & Koprowski, 2017). Individual differences in behaviour and activity form the basis for variability in migrations, dispersal, habitat use and foraging, which may affect population dynamics (Bowler & Benton, 2005; Canestrelli et al., 2016; Cote et al., 2010). Individual differences in personality can also affect the monitoring of exploited populations because the fast‐growing individuals that are generally characterized by bold (active) behaviour may be more likely to be captured, especially by passive fishing gear like longlines (Biro, 2013; Biro & Dingemanse, 2008). This could be relevant for the stock assessment of halibut in the GSL, for which a main index provided by a longline survey relies on active foraging for capture. Understanding individual variability in personality is also important in the context of environmental change, as the adaption potential to habitat change may partly depend on the diversity of behavioural repertoire (Conrad et al., 2011).

AUTHOR CONTRIBUTIONS

All authors collectively developed the project's objectives, contributed to methodology and performed tag deployment/recovery. Marie‐Pier Boulanger carried out data analysis with support from Hugues P. Benoît. Marie‐Pier Boulanger was involved in result interpretation and writing of the manuscript with support from all authors.

FUNDING INFORMATION

This project is part of a collaborative programme between the Institut des sciences de la mer (ISMER) at the Université du Québec à Rimouski (UQAR), the Maurice Lamontagne Institute (IML) of Fisheries and Oceans Canada (DFO), the Université du Québec à Chicoutimi (UQAC) and the Fisheries and Marine Institute of Memorial University of Newfoundland. It was supported by a Strategic Partnership grant from the Natural Sciences and Engineering Research Council of Canada (NSERC) (grant number STPGP 506993‐17), the John R. Evans Leaders Fund of the Canada Foundation for Innovation, the Department of Fisheries and Land Resources of Newfoundland and Labrador, Ressources Aquatiques Québec and the Ministère de l'agriculture, des pêcheries et de l'alimentation du Québec.

Supporting information

Data S1. Supporting information.