Feeding ecology of wild migratory tunas revealed by archival tag records of visceral warming


  • Sophie Bestley,

    Corresponding author
    1. CSIRO Marine and Atmospheric Research Laboratories, Hobart, Tas. 7000, Australia; and
    2. School of Zoology, University of Tasmania, Sandy Bay Campus, Hobart, Tas. 7001, Australia
    Search for more papers by this author
  • Toby A. Patterson,

    1. CSIRO Marine and Atmospheric Research Laboratories, Hobart, Tas. 7000, Australia; and
    2. School of Zoology, University of Tasmania, Sandy Bay Campus, Hobart, Tas. 7001, Australia
    Search for more papers by this author
  • Mark A. Hindell,

    1. CSIRO Marine and Atmospheric Research Laboratories, Hobart, Tas. 7000, Australia; and
    Search for more papers by this author
  • John S. Gunn

    1. CSIRO Marine and Atmospheric Research Laboratories, Hobart, Tas. 7000, Australia; and
    Search for more papers by this author

  • 1

    Based on the length to weight conversion W = 3.13088L2.9058 × 10−5 (Robins 1963).

*Correspondence author. CSIRO Marine and Atmospheric Research, GPO Box 1538, Hobart, Tas. 7001, Australia; E-mail: sophie.bestley@csiro.au


  • 1Seasonal long-distance migrations are often expected to be related to resource distribution, and foraging theory predicts that animals should spend more time in areas with relatively richer resources. Yet for highly migratory marine species, data on feeding success are difficult to obtain. We analysed the temporal feeding patterns of wild juvenile southern bluefin tuna from visceral warming patterns recorded by archival tags implanted within the body cavity.
  • 2Data collected during 1998–2000 totalled 6221 days, with individual time series (n = 19) varying from 141 to 496 days. These data span an annual migration circuit including a coastal summer residency within Australian waters and subsequent migration into the temperate south Indian Ocean.
  • 3Individual fish recommenced feeding between 5 and 38 days after tagging, and feeding events (n = 5194) were subsequently identified on 76·3 ± 5·8% of days giving a mean estimated daily intake of 0·75 ± 0·05 kg.
  • 4The number of feeding events varied significantly with time of day with the greatest number occurring around dawn (58·2 ± 8·0%). Night feeding, although rare (5·7 ± 1·3%), was linked to the full moon quarter. Southern bluefin tuna foraged in ambient water temperatures ranging from 4·9 °C to 22·9 °C and depths ranging from the surface to 672 m, with different targeting strategies evident between seasons.
  • 5No clear relationship was found between feeding success and time spent within an area. This was primarily due to high individual variability, with both positive and negative relationships observed at all spatial scales examined (grid ranges of 2 × 2° to 10 × 10°). Assuming feeding success is proportional to forage density, our data do not support the hypothesis that these predators concentrate their activity in areas of higher resource availability.
  • 6Multiple-day fasting periods were recorded by most individuals. The majority of these (87·8%) occurred during periods of apparent residency within warmer waters (sea surface temperature > 15 °C) at the northern edge of the observed migratory range. These previously undocumented nonfeeding periods may indicate alternative motivations for residency.
  • 7Our results demonstrate the importance of obtaining information on feeding when interpreting habitat utilization from individual animal tracks.


Understanding the movement of animals in time and space and its implications for the abundance and distribution of populations is a pivotal problem in animal ecology, fundamental to conservation and resource management strategies. Migration is often a response to environmental heterogeneity, such as adaptations to seasonal cycles in weather patterns and resource availability (Alerstam, Hedenstrom & Akesson 2003). For immature or nonbreeding individuals, the resources driving movement generally relate to food supply and/or habitat suitability (Baker 1978). Modern telemetry has provided data on individual movements of highly mobile marine animals previously difficult to study due to their speed and the vast range over which they travel (Ropert-Coudert & Wilson 2005; Hays 2008). Direct links between movement patterns and food resources have been difficult to establish, due to a lack of explicit information on food availability, foraging activity and feeding success.

In the oceanic environment, resources are patchily distributed at a range of spatial and temporal scales and predators must successfully locate prey in a three-dimensional habitat (Hindell et al. 2002). Foraging theory predicts that animals should spend more time in areas of high foraging success, and should minimize time spent moving between these areas (Stephens & Krebs 1986). Although the interpretation of individual movement data has proved to be a nontrivial task (Turchin 1998), time spent in an area is commonly used as a convenient way to represent data from multiple individuals or species (Guinet et al. 2001; Bradshaw et al. 2004). Habitat utilization by marine predators is widely assumed to reflect the quality and availability of resources in an area, with areas of high use popularly inferred to be areas of high foraging success (Bailey & Thompson 2006). From this premise ensues an increasing effort to identify oceanic regions of enhanced biological productivity of interest to predators (Sydeman et al. 2006), often with objectives relating to conservation zoning and fisheries management. However, the assumption that areas of highest residency correspond to feeding areas has remained largely untested in the absence of direct feeding data.

Direct observation of feeding is generally not possible for large marine predators as they forage over large and remote areas and commonly while diving. Hence, information on when and how feeding actually occurs is commonly inferred indirectly from behavioural information such as vertical diving (Boyd & Croxall 1996), landing events in seabirds (Shaffer, Costa & Weimerskirch 2001) or movement patterns (Jonsen, Flenming & Myers 2005; Robinson et al. 2007). There are a number of methods for directly establishing feeding rates for marine vertebrates. Motion sensors attached to the jaws can reveal feeding patterns over periods of a few days (Fossette et al. 2008). For cetaceans that use echolocation to locate prey, loggers recording ambient sounds may reveal prey pursuit and capture (Watwood et al. 2006). However, for longer-term records of feeding, perhaps the most widely used technique is to measure stomach or oesophageal temperature (Gales & Renouf 1993; Austin et al. 2006). Yet stomach telemetry is often hampered by the premature ejection of sensors from the stomach.

In bluefin tunas, the visceral temperatures increase markedly in association with digestion (Carey, Kanwisher & Stevens 1984). This is thought to be due to a combination of factors, but probably most important is the heat produced by specific dynamic action (i.e. the result of metabolic heat production during digestion). The elevated temperatures promote increased enzyme activity (Stevens & McLeese 1984) and appear to be the primary mechanism by which tunas digest food much faster than other piscivorous fish. In cage experiments, archival tags incorporating a temperature sensor implanted within the body cavity of southern bluefin tuna (Thunnus maccoyii) and Pacific bluefin tuna (Thunnus orientalis) showed regular patterns of visceral warming and cooling, providing an accurate record of when feeding events occurred (Gunn, Hartog & Rough 2001; Itoh, Tsuji & Nitta 2003). These patterns also occur in data collected from wild fish (Gunn & Block 2001; Itoh et al. 2003; Kitagawa et al. 2004).

For adult animals, long-distance migrations are often associated with travel to breeding sites. This means that tracking often simply reveals shuttling between breeding and foraging sites, rather than specifically animal search for prey patches (Hays et al. 2006; Bailey et al. 2008). Tracks of juveniles, being nonbreeding, might therefore be expected to be more tightly coupled to food resources. We used long-term archival tag records of visceral warming to examine the temporal feeding patterns and seasonal foraging ecology of wild juvenile southern bluefin tuna (SBT) during their migrations in the south Indian Ocean. Specifically, we examined the hypothesis that highly mobile species should spend more time in energetically profitable areas, by examining the relationship between the feeding success of SBT and the time spent in specific areas. The rare combination of long-term feeding and movement data allows for a unique interpretation of habitat utilization by a highly migratory marine predator.


data collection and processing

Archival tag data

During the austral summers of 1998–2000, SBT (n = 200) were caught by pole-and-line in the Great Australia Bight (GAB) and archival tags of model Mk7 (Wildlife Computers, Redmond, WA, USA) were surgically implanted into the peritoneal cavity ventral to the stomach. Tags sampled pressure (depth), ambient light, ambient water and visceral temperatures every 4 min. To date, 51 (25·5%) have been recovered and data retrieved from 47. Due to very early recapture (n = 10), or sensor (n = 4) and tag (n = 4) failures, long-term (> 120 days) data were obtained from only 29.

Location estimation

Daily longitudes were determined by geolocation methods (Hill 1994) using geocontrol software version 2·01·0002 (Wildlife Computers). Nineteen age-3 fish (mean fork length = 99 ± 3 cm, range = 93–111 cm) (Eveson, Laslett & Polacheck 2004) moved west into the south Indian Ocean during their first year at liberty, and these migrants are the focus of this analysis. Latitude was estimated by comparing the surface water temperature recorded by the tag with satellite sea surface temperature (SST) estimates (Teo et al. 2004). Briefly, using the Advanced Very High Resolution Radiometer (AVHRR) weekly global 18 km multichannel SST (MCSST) (night passes) data (http://podaac.jpl.nasa.gov/PRODUCTS/p016.html), a strip centred on the geolocation longitude (±1ºE) was searched from 20°S–60°S. Shown in this analysis are the median positions of all MCSST pixels matching within ±0·2 °C of the median temperature recorded in the surface 5 m during each 24 h. Using this method, the average 90th percentile boundaries of all pixel matches are 1·1 and 0·8 degrees to the north and south of the median position, respectively.

Determining the time of feeding events and relative meal size

The visceral warming recorded by the internal temperature sensor on tags (Fig. 1) provided a record of (i) feeding incidence, and (ii) a means of estimating relative intake mass. The data were viewed over 48 h windows and a basal temperature calculated. The start of a feeding event was identified by either (i) a sharp dip in visceral temperature (associated with the ingestion of cold food and/or water) followed by a steep steady rise above the initial temperature, or (ii) a steep steady rise not preceded by a dip. The simultaneous ambient temperature and depth records were used to evaluate whether observed changes were from moving into cooler water. A feeding event was considered to have ended when the visceral temperature passed the maximum rise and (i) returned to basal, (ii) dropped to a new plateau, or (iii) changed in a manner indicative of the start of another feed.

Figure 1.

Example of archival tag time series showing visceral warming patterns recorded by the internal temperature sensor (bold black line). Circles indicate commencement of a feed, horizontal lines indicate feed duration and vertical lines indicate time at which the maximum heat increment is reached (Tmax). Vertical swimming depth (grey) and ambient water temperature (black) are also shown. (a) First feeding events of SBT97639 17 days post-tagging in the Great Australia Bight during the austral summer; (b) SBT97622 in the central south Indian Ocean (97–98°E, 34–35°S) during the austral winter.

In SBT, there is a robust relationship between total intake size (of prey in grams) and the time taken to reach the maximum heat increment above basal (Tmax, Fig. 1) (Gunn et al. 2001). However this relationship was developed from a single prey type (pilchards), and the type and relative importance of baitfish in SBT diet is known to be regionally variable, for example, from > 80% to < 50% weight (Young et al. 1997). Furthermore, the effects of varying prey type (e.g. fish cf. squid) and energy density or composition (e.g. lipid content) on the slope of this relationship are unknown and likely to be substantial (Olson & Boggs 1986). Therefore, it is important to stress the use of this data as a relative quantitative index of intake size only, and all quantitative predictive models are restricted to using counts or presence/absence of feeding events. Results are reported as mean ± SD across individual fish unless otherwise stated.

data analysis

Temporal feeding patterns

To investigate temporal patterns in feeding, we examined both moon quarter and time of day. Lunar illumination for a given date was calculated using standard astronomical equations and feeding events aggregated into the four periods of dark, waxing, full and waning. Time of day was also aggregated into four periods. Feeding events initiated within 2 h either side of dawn or dusk were defined as such. While these periods each spanned 4 h per 24 h, the day and night lengths varied seasonally.

To test for the effects of moon quarter and time of day on feeding frequency, generalized linear mixed models (GLMM) were used where the random effect was the individual fish and the error distribution was Poisson with a log link function. To determine the best predictive model, the GLMM having the lowest Akaike's information criterion (AIC) was selected. The models were fit by maximizing the log-likelihood using the Laplacian approximation. Therefore, to evaluate the properties of individual coefficients, we sampled from the posterior distribution of the parameters of the best predictive model using Markov chain Monte Carlo methods (n = 20 000, using function mcmcsamp in r package lme4). The prior on the fixed-effects parameters is taken to be locally uniform. Reported are the mean estimate ± SD together with the 95% highest posterior density (HPD) intervals and associated empirical P value. All statistical analyses were carried out using r software version 2·5·1 (R Development Core Team 2007), with the GLMMs fitted in the package lme4 version 0·99875-6 (Bates 2007).

Foraging ecology

To investigate whether the foraging ecology of juvenile SBT changed seasonally, we examined the water temperature and depth data, recorded by the tag, associated with the start of each feeding event. Linear mixed models were used where season was fitted as both a fixed and a random effect. The models were fit using restricted maximum-likelihood estimation (REML). Reported F values and P values are based on Wald tests. In case of a lag between prey capture and the start of visceral warming, the models were refit to the data from the previous timestamp (i.e. the temperature and depth records 4 min before the estimated start time) but there were no substantive differences. The LMMs were fitted in the r package nlme version 3·1-83 (Pinheiro et al. 2007).

Feeding and residency

To examine the relationship between feeding success and time spent in an area, we gridded the daily position estimates at various spatial scales. To take into account the error associated with the geolocation methods, the smallest grid used was 2 × 2° in size, ranging up to a 10 × 10° grid. The number of successful/unsuccessful feeding days was then modelled in response to the total number of days spent per square, using a generalized linear model with a binomial error distribution and a logit link function. This model was fit to the original data aggregated across fishes and a bootstrap validation run to ascertain validity across fishes, in which all fish (and all their data points) were sampled with replacement and the model refit (n = 10 000). To investigate further the pattern for individual fish, we examined GLMMs with a binomial error distribution. Data from squares centred within the GAB coastal summer residency area bounded by the coordinates (127·5°E, 29°S), (127·5°E, 34°S), (140°E, 34°S), (140°E, 42·5°S) were excluded from this analysis. Fish with data available only for the outward migration leg from the GAB, that is, without any oceanic residencies, due to early tag failure (n = 3) or recapture (n = 1), were also excluded.


A total of 6221 days of data were collected from the 19 fish, with individual time series spanning 141–496 days (Table 1). Individuals recommenced feeding between 5 and 38 days after tagging (mean ± SD: 19 ± 10 days). These post-release fasting periods were excluded from all analyses. Feeding events (n = 5194) were subsequently identified on 76·3 ± 5·8% of days (range: 63·5–84·9%), for an overall feeding rate of 0·89 ± 0·08 feeding events per day per individual (range: 0·76–1·16). On feeding days, the feeding rate was 1·17 ± 0·09 (range: 1·05–1·48). The estimated daily intake was 0·75 ± 0·05 kg (range: 0·67–0·85 kg) and the mean feed size was 0·85 ± 0·09 kg (range: 0·63–1·03). The mean duration of the visceral temperature signature was 18·5 ± 1·4 h (range: 14·4–20·9). The minimum interval recorded between feeding events was 40 min; however, there was a distinct diurnal cycle with the median ranging between 18·7–24·1 h. There was considerable variation in the maximum between-feed interval, with individuals recording periods of up to 3·9–24·4 days without feeding (mean ± SD: 11·4 ± 6·5).

Table 1.  Feeding information obtained for 19 wild juvenile SBT based on visceral temperature patterns
SBT IDRecord length (days)Post-surgery fast* (days)Days with feeding (days)Total feedsFeeding rate (feeds day−1)Daily intake (kg)
  • Daily intake calculated from the relationship with time to maximum temperature, Tmax = 0·5845 * Intake (g) (Gunn et al, 2001).

  • *

    Data from this post-surgery period were excluded from all subsequent calculations and analyses.

9762240936258 (69·2%)3240·870·75
9768233221244 (78·5%)2840·910·79
9770824720168 (74·0%)2060·910·78
9772133832221 (72·2%)2510·820·77
9761522731165 (84·2%)1910·970·67
9763949617378 (78·9%)4120·860·69
9771820915152 (78·4%)2251·160·73
9773333219231 (73·8%)2800·890·77
9774125213188 (78·7%)2160·900·84
9800738938235 (67·0%)2660·760·78
9773149011370 (77·2%)4420·920·85
9855333517254 (79·9%)2830·890·80
9855631631181 (63·5%)2220·780·71
9857414120 93 (76·9%)1040·860·76
992674546349 (77·9%)4080·910·70
996263705259 (71·0%)3010·820·70
9962734110267 (80·7%)2940·890·76
996292975248 (84·9%)2830·970·79
9966424615193 (83·5%)2020·870·71
Mean ± SD  19 ± 10(76·3 ± 5·8%) 0·89 ± 0·080·75 ± 0·05
Range141–496  5–38(63·5–84·9%) 0·76–1·160·67–0·85

temporal patterns in feeding

The number of feeding events varied strongly with time of the day (GLMM, all |z| > 24, P < 0·0001; Table 2) with the highest number of feeding events (58·2 ± 8·0%) and feeding rate (0·16 ± 0·03 feeds h−1) occurring near dawn (Fig. 2a). Some feeding activity occurred during the day (26·0 ± 6·4%) but rarely at night (5·7 ± 1·3%, Fig. 2a). Moon quarter was not a significant independent factor influencing the number of feeding events (GLMM, all |z| < 1, P > 0·3), however the moon quarter-time of day interaction was significant (Table 2). Night feeds during the full moon quarter were predicted to be more likely by a factor of 10·7 (± 1·3, 95% highest posterior density (HPD) = 6·2–19·6, P < 0·0001) relative to the dark moon, and by a factor of 5·6 (± 1·4, 95% HPD = 3·1–10·2, P < 0·0001) and 3·3 (± 1·4 SE, 95% HPD = 1·8–6·3, P = 0·0001) during the waning and waxing quarters, respectively (Fig. 2b). A weaker effect of the full moon was a small decline in dawn feeding (by a factor of 0·8 ± 1·1, 95% HPD = 0·7–0·9, P = 0·0001) and increase in day and dusk feeding (by a factor of 1·5 ± 1·1, 95% HPD = 1·2–1·8, P < 0·0001, and 1·6 ± 1·1, 95% HPD = 1·3–2·0, P = 0·0001 respectively). The best model explicitly included a random effect for time of day within fishes, due to small variations in relative feed incidence between day and dusk.

Table 2.  Results for generalized linear mixed models (GLMM) with time of day and moon quarter as factors affecting feeding frequency. The best fit was determined according to the lowest Akaike's information criterion (AIC)
  1. n = 5179, i.e. 5194 total feeds less 15 feeds with TOD unknown due to light sensor failure; scale, the estimated scale parameter (ideally close to 1); d.f., degrees of freedom; LL, log-likelihood; ΔAIC, the difference in AIC from that of the best-fitting model.

Fixed effectsRandom effects     
(a) Time of Day (TOD)Intercept1·5 5 –361·2 732·3 204·2
(b) Moon Quarter (MQ)Intercept3·5 5–1928·93867·73339·6
(c) TOD + MQIntercept1·5 8 –360·5 737 208·9
(d) TOD + MQ + (TOD × MQ)Intercept1·317 –297·6 629·2 101·1
(e) TOD + MQ + (TOD × MQ)Intercept + TOD1·027 –237·1 528·1   0
(f) TOD + MQ + (TOD × MQ)Intercept + MQ1·327 –297·6 649·2 121·1
Figure 2.

Frequency of feeding events by (a) time of day, and (b) moon quarter. Feeding rate (feeds hr−1) by time of day is also shown (dashed line) in (a).

seasonal feeding ecology

Comparison of the seasonal temperature–depth habitat of SBT (Fig. 3b–e) and where feeding occurred (Fig. 3f–i) showed similarity in the core areas, but in general a wider habitat envelope. Individuals foraged in ambient water temperatures ranging from a minimum of 4·9–11·3 °C (9·0 ± 1·4) to a maximum of 19·3–22·9 °C (20·9 ± 0·8). The maximum depth records associated with the start of feeding events ranged between 247–672 m (460 ± 134 m), and all fish recorded feed events within the surface 2 m. In summer, 41·8% (± 11·8), 55·2% (± 9·2) and 3·0% (± 6·2) of SBT feeds were initiated within the surface (0–10 m), epipelagic (10–150 m) and mesopelagic (> 150 m) layers of the water column. In winter, these proportions changed to 17·5% (± 7·9), 58·5% (± 10·4) and 24·0% (± 13·2), respectively.

Figure 3.

Seasonal feeding ecology in relation to depth (m) and temperature (°C). (a) Time series for an individual fish (SBT99629) showing locations where a feed event was initiated within the water column (black circles) during the summer GAB residency (1.), a period of rapid westward migration (2.), and winter-spring within the central (70–90°E) south Indian Ocean (3.). Background shows ambient water temperature (°C) at depth. Remaining panels show seasonal (b–e) habitat use, and (f–i) occurrence of feeding events, by depth (25 m bins) and temperature (1 °C bins) for the austral (b, f) summer (January–March, N = 18 fish, n = 1058 feeds), (c, g) autumn (April–June, N = 19, n = 1716), (d, h) winter (Jul–Sep, N = 19, n = 1419), and (e, i) spring (October–December, N = 15, n = 1000). Data in (b) – (i) are aggregated as average proportions across fishes.

The seasonal differences in SBT foraging ecology were significant with respect to both the ambient water temperature (F = 41·7, P < 0·0001) and depth (F = 26·0, P < 0·0001) in which feeding events occurred (Fig. 3; Table 3). As expected, the temperature records showed a shift from feeding in warm summer (18·0 ± 0·4 °C) to cooler winter (13·6 ± 0·2 °C) temperatures. This coincided with a shift from shallow summer feeding (29·6 ± 4·6 m) to deeper winter foraging (96·6 ± 8·5 m) near the bottom of the epipelagic layer. The highest variance in feed temperatures between fish occurred in autumn and summer, likely reflecting both stronger vertical stratification and a higher range of near-surface temperatures in these seasons (Fig. 3a,b). In contrast, winter showed both the lowest variance in feed temperatures and the highest variance in feed depths, reflecting the more homogeneous temperatures and expanded range of depths SBT foraged in during this season (Fig. 3c).

Table 3.  Parameter estimates from linear mixed models with season as a factor affecting the temperature (°C) and depth (metres) at which feed events occur
ResponseSeasonσfs*EstimateSE95% CI
  • n = 5139, i.e. 5194 total feeds less 55 feeds with temperature unknown due to external temperature sensor failure

  • *The between-fish (f) within-season (s) random effect modelled as a random normal variable ~N(0, inline image).

TemperatureSummer 1·7117·990·4117·19–18·78
Autumn 1·8415·510·4314·68–16·34
Winter 1·0013·560·2413·09–14·02
Spring 1·2614·780·3314·12–15·43

feeding and residency

The length of time fish spent in an area showed no clear relationship to their success in feeding. Three example tracks (Fig. 4) demonstrate how in some instances the aggregated positions of individual fish appeared to correspond with enhanced feeding rates, and in others to correspond with reduced feeding. The data aggregated across fishes showed some spatial structuring outside of the GAB with a high number of days per square recorded around the south-west corner of Australia, in the central south Indian Ocean near (40°S, 80°E), in some areas along the subtropical latitudes (i.e. along 30°S) and in the far western basin (Fig. 5a). However, daily feeding frequency and intake showed generally high success throughout the more southerly latitudes (Fig. 5b,c).

Figure 4.

Three examples of SBT migratory paths with daily positions coloured according to estimated daily intake (kg day−1). Open grey circles indicate no feed events occurred. Tracks of (a) SBT97721 for the period 4 February 1998 to 5 December 1998; (b) SBT97731 for the period 12 March 1999 to 1 July 2000; (c) SBT99267 for the period 23 February 2000 to 7 February 2001. The Great Australia Bight region is shown by the grey polygon.

Figure 5.

Residency and feeding of juvenile SBT in the south Indian Ocean. (a) Occupancy (days), (b) feeding frequency (feeds day−1), (c) daily intake (kg day−1), and (d) all extended fasting periods of 5 days or more. Panels (b) and (c) are shown as anomalies about the overall mean (0·92 feeds day−1 and 0·81 kg day−1, respectively) and exclude squares with less than 7 days data. Data are aggregated by 3° squares across fishes (N = 19, n = 5222). In panel (d) colours identify individual fish (N= 17), and symbols link daily positions within a particular fasting period (n = 428 days over 54 separate periods). For example the dark blue represents SBT97721, with the triangles near (83°E, 30°S) showing a 6-day fast in July 1998, and the squares near (78°E, 30°S) showing a 22-day fast in late August 1998, respectively. Grey circles show all other daily positions where nonfeeding was recorded. Background shows the 1998–2000 winter-spring SST (°C) climatology (compiled from June–November monthly NOAA OI SST V2 data, http://www.cdc.noaa.gov/cdc/data.noaa.oisst.v2.html). The climatological position of the northern subtropical front (Belkin & Gordon 1996) is also shown for reference (black line).

From the data aggregated across fishes, the relationship between the number of successful and unsuccessful feeding days per square and the total number of days per square was determined to be significantly positive at only two of the six scales investigated (the 5 × 5° grid and 7 × 7° grid, Table 4; |z| > 3, P ≤ 0·001). The bootstrap procedure showed significance at these two scales in only 52·9% and 59·2% of the bootstrap samples, respectively (α = 0·01). Therefore, at all scales the bootstrapped parameters were nonsignificant; that is, the confidence intervals for all the parameter estimates straddled zero (Table 4).

Table 4.  Results of generalized linear models (GLM) fit to binomial feeding data at different spatial grids (2 × 2 to 10 × 10 degree squares). Results from bootstrapping across fishes (n = 10 000 samples) also shown
Grid sizeBinomial GLMBootstrap results
Est.SDCIPercentage of P values
d.f.DevianceLLAICEst.SEz valuePr(>| z |)lowerupper< 0·05< 0·01
  • Based on a chi-squared test.

3191449·11–417·43 838·87–0·780·50–1·560·12–0·621·18–2·741·7958·146·5
4118360·38–327·89 659·78 0·750·44 1·710·09 0·620·96–1·202·4854·342·2
5 80297·75–267·62 539·23 1·260·39 3·220·001 0·760·86–1·072·3563·352·9
6 60241·87–221·75 447·50–0·120·39–0·300·76–0·111·00–1·851·9556·544·8
7 47203·98–186·69 377·38 1·060·28 3·790·00015 0·760·65–0·591·9369·759·2
10 25131·15–124·98 253·97 0·030·22 0·150·88–0·010·47–1·010·7840·727·2

However, the mixed models revealed significant patterns at the level of the individual fish. At all scales except one, the preferred model contained both a random effect (intercept) for the individual fish and a random effect (slope) for time spent within a square [all likelihood ratio (LR) > 7·9, P ≤ 0·005, Supporting Information Table S1], the single exception being at the 5-degree grid scale (LR = 3·7, P = 0·05). The further addition of time as a fixed effect was not supported at any scale (all |z| < 1·7, all P ≥ 0·09). Examination of the model predictions showed opposing effects among individual fish: some fish increased feeding probability with time spent in a square, whereas other fish decreased and some showed no effect of time (Supporting Information Fig. S1). These patterns were observed consistently over the scales investigated. The significant fish-to-fish variability in the patterns of feeding success over the time spent within an area explains why a population-level parameterization failed to reveal any significant pattern.

The negative relationships relate to prolonged periods of nonfeeding observed in some fish. While nonfeeding days accounted for 24·0% of all days (Table 1), of these 30·5% were actually associated with extended fasts of between 5 and 24 days (n = 428 days over 54 periods, N = 17 of the 19 fish). These fasting periods occurred throughout the year, within the GAB region during both the first (35·2% fasts, N = 13 fish) and the second (24·1% fasts, N = 5 fish) summer residencies, as well as within the wider south Indian Ocean (40·7% fasts, N = 11 fish). Fasting periods occurred almost exclusively within warmer waters (99·5% of fasting days occurred in waters with daily mean SST > 15 °C). The distribution of these periods therefore tended to be along the northern edge of the migratory range observed in this study (Fig. 5d). Furthermore, from the position data available (≥ 5 days, n = 41 of the 54 periods) the majority of fasting periods were clustered along relatively tortuous portions of the migratory paths (87·8% had a linearity index (LI) < 0·5, LI = net movement/total path length during the fasting period) rather than along relatively straight sections of track (Figs 4 and 5d), that is, coincident with periods of apparent residency.


Seasonal long-distance migrations are thought to be related to resource distribution, but data on feeding success needed to test this presumption has been difficult to obtain for highly migratory marine species. This is the first study to relate a quantitative analysis of feeding frequency to the foraging ecology of a wild migratory fish. The feeding data are consistent with a predominantly visual generalist predator using adaptive foraging strategies in a seasonally variable environment. However, the combined feeding and movement data do not support the hypothesis that activity is concentrated only in areas where resources are more available thereby increasing feeding success. Instead the data show relatively consistent feeding throughout most of their migratory range, but also highlight previously undocumented nonfeeding periods. Given that an animal has an incomplete knowledge of their environment, this may reflect variability in foraging success due to poor predator search strategy, or a poor prey field. Alternatively, these results may indicate that other factors can motivate residency. In light of the rapidly increasing emphasis on telemetry-based studies of mobile marine animals, our results demonstrate the importance of obtaining independent information on feeding when interpreting habitat utilization from individual animal tracks.

The tagging procedure was designed to minimize handling stress and involved minor surgery, with fish returned to the water less than 2 min after landing. Short-term recoveries have established that the surgical wound can completely heal after 2 weeks. Nevertheless, the majority of fish (68%) took more than 2 weeks to re-establish normal feeding behaviour after tagging. This post-surgery period was excluded from all analyses, and we believe it unlikely that the tag might have influenced observed patterns in feeding. Although this possibility cannot be excluded, the long-term nature of the archival tag records (some in excess of 40 months) in combination with the high tag-return rates (including recaptures after more than 9 years at liberty) suggest SBT do fully recover from the stress of handling and surgery.

temporal feeding patterns

The temporal patterns in feeding in relation to time of day and moon phase are consistent with tunas being primarily visual predators (Itoh et al. 2003; Kitagawa et al. 2004). The predominance of dawn feeding had previously been inferred for SBT from stomach fullness as well as fishery targeting strategies (Young et al. 1997). Increased activity and enhanced feeding during crepuscular hours has been widely reported across aquatic and terrestrial taxa (Reebs 2002; Kronfeld-Schor & Dayan 2003) and may be due to a combination of factors. For example, vertically migrant prey have not yet descended (Hays 2003) and/or there is a predatory advantage during changing light conditions. Many marine predators also exhibit changes in behaviour over the lunar cycle (Horning & Trillmich 1999) which is further thought to correlate with light-driven changes in the vertical migration of prey.

Further conclusions about the influence of time of day are limited by the subjectivity involved in identifying the separate ingestion of multiple prey items. Using our current methods, it is likely that a number of individual meals are identified as a single feeding ‘event’, and we cannot distinguish between 1 kg ingested in minutes and 1 kg eaten over a number of hours. Distinct multiple feeding events were identified (12·7 ± 4·4% days, range = 7·0–22·3%) and are probably underestimated, although this may be more important in coastal regions where the main prey are thought to be schooling fish (Young et al. 1997). The relatively long duration of each event, similar to the findings from previous studies (Gunn et al. 2001; Itoh et al. 2003), reflects the physiological nature of the data. The time-scale for an event (~18 h) is comparable to the time to gastric evacuation (Olson & Boggs 1986). However, the daily intake estimates, although only intended as a relative measure, provide very plausible values (Olson & Boggs 1986) for a daily ration of 3.8 ± 0.5% (range: 2.9–4.9%) of body mass per day.1 Therefore, our results may in part represent the efficiency of a predator aiming to fill its stomach once per day, and feeding rapidly to satiation when the opportunity arises.

seasonal foraging ecology

Our findings show juvenile SBT predominantly exploit the epipelagic ocean, with some feeding also occurring within the mesopelagic layer particularly during winter. The deepest recorded feeding event (672 m) also provides evidence that SBT occasionally forage at very low light levels, consistent with previous reports of deep-sea crustaceans and bottom-dwelling fish in their diet. A deeper winter vertical distribution, and by inference feeding depth, has been previously reported for other tunas (Thunnus orientalis; Kitagawa et al. 2004) and predatory fish (Oncorhynchus tshawytscha; Hinke et al. 2005). Studies of marine predators in the southern oceans such as penguins and seals have also reported deeper diving patterns, increased diving effort, expanded foraging ranges and an increased diversity of prey items during winter (Aptenodytes patagonicus; Cherel, Ridoux & Rodhouse 1996; Arctocephalus tropicalis; Beauplet et al. 2004; Eudyptes chrysolophus; Green et al. 2005). Such behavioural flexibility, as observed across a variety of top predators, has widely been interpreted as a response to seasonally reduced epipelagic prey availability and density, ensuring broader diets and higher feeding rates.

linking feeding and residency

Increasingly, quantitative analyses aimed at elucidating foraging strategies are being applied to high-precision movement tracks such as obtained via global positioning system collars (Morales et al. 2004). However, the large errors associated with geolocation methods, and to a lesser extent Argos satellite estimates, present substantial limitations to the application of such track-based analytical methods (Bradshaw, Sims & Hays 2007). The geographical grid scales examined in this study, although somewhat of an artificial imposition, were selected to cover an appropriate range given both the coarse resolution of the data and the spatial scales of the movements being studied. The range also encompassed scales previously identified as characteristic of mesoscale foraging patches for other top predators operating over similar spatial scales, although shorter time-scales (Fauchald & Tveraa 2006).

Our analysis found time spent within an area had no clear relationship with feeding success. Most interestingly, this was primarily a result of the high degree of individual variation with both positive and negative trends observed between individuals. Importantly, these patterns were observed consistently across all the spatial scales examined. This finding provides strong evidence against the interpretation of high-use areas for migratory species, commonly determined from only horizontal and/or vertical movement data in the absence of any independent information on feeding activity, as successful feeding grounds (Robinson et al. 2007). This strengthens previous findings from studies using indirect measures such as body condition (Bailleul et al. 2007) and mass gain (Bradshaw et al. 2004) which found equivocal or no evidence for a relationship between foraging success and spatial usage of areas. Explanations for the observed variation in trends can be (i) time spent within an area is more likely to be a measure of searching activity rather than a direct proxy for foraging success, given that predators have imperfect knowledge of their environment; and (ii) predators do not necessarily feed all the time, and may spend time within areas for alternative reasons.

Although animal movements may be driven by resource distribution, superimposed on this is the fact that they may have incomplete knowledge of their environment, particularly in heterogeneous marine environments. At very broad geographical and seasonal scales, predators may have an awareness of prey distribution (Bradshaw et al. 2004; Houghton et al. 2006), and at fine-scales foraging is likely to be dominated by proximal sensory clues (Sims & Quayle 1998). However, the mesoscale search strategies by which animals locate prey remains a pivotal problem in ecology. To examine whether animals use optimal search strategies, recent studies have used empirical data on both horizontal and vertical movements (Edwards et al. 2007; Sims et al. 2008), as well as simulations (Sims et al. 2006). Our finding that juvenile SBT did not consistently spend longest where their feeding success was highest may reflect that as juveniles, some individuals have poor search strategies (Sims et al. 2006); however, no information on adult foraging is currently available to determine if this is a specific juvenile behaviour. Since our study examined same-age individuals within a narrow size range, therefore presumably similar levels of foraging experience and skill, the high variability in patterns of feeding between individuals are not expected to be related to age or size. Yet as unsexed juveniles, it was not possible to determine whether the differences were sex-related.

The variable patterns of feeding success may reflect real differences in the availability of prey or the prey type targeted by individuals. High individual variability in feeding behaviour, not specific to juvenile cases, has been reported in many other top predators including seals (Halichoerus grypus; Austin et al. 2006), albatross (Diomedea exulans; Weimerskirch et al. 1997) and penguins (A. patagonicus; Putz & Bost 1994), and generally attributed to the patchy and unpredictable prey distribution. It is also a possibility that the high spatial overlap in the locations of poor feeding success are indicative of a relatively poor season(s) within a region of usually predictably high forage (Bradshaw et al. 2004). Future studies should reveal if (i) SBT do show fidelity to particular regions; and (ii) the identified areas of low feeding success are consistent or variable between years.

The second explanation for the observed variation in feeding trends is that large predators do not necessarily feed all the time, and may spend time in particular areas for alternative reasons. Particularly in coastal waters, where SBT exhibit strong schooling behaviour, movements may be driven by social factors rather than individual-based decision-making (Gunn & Block 2001). In an oceanic context, association with floating objects and/or topographic structures has been proposed as a behaviour that increases the encounter rate between isolated migratory fishes (Freon & Dagorn 2000). Alternatively, particular topographic features may be important for other reasons such as resting or for navigational reference (Castro, Santiago & Santana-Ortega 2001). Movements may also be related to predator avoidance, which is often ignored for large species but has been demonstrated in bottlenose dolphins (Tursiops aduncus) and green turtles (Chelonia mydas) where there are high abundances of sharks (Heithaus & Dill 2006; Heithaus et al. 2007). Finally, it has been suggested that when tunas forage within frontal zones, the warm side of fronts may be used as a thermal refuge, enabling them to feed in cold waters while minimizing the cost of staying warm (Gunn & Young 1999; Kitagawa et al. 2004). Thermal conditions are known to play a driving role in the large-scale movements of a range of marine predators including turtles, fish and mammals (McMahon & Hays 2006; Neat & Righton 2007). In our study, the prevalence of nonfeeding periods in warm waters (Fig. 5d) provides some evidence that resting periods may occur within thermal refuges over longer time-scales (of days to weeks) in the south Indian Ocean.

In summary, we have shown the value of integrating direct information on feeding when interpreting habitat utilization from individual animal tracks. Our findings do not show a straightforward relationship between feeding and residency. Current efforts to develop novel methods for determining feeding activity on a wide range of marine species will continue to advance our understanding of foraging ecology and the critical linkages between animal behaviour, movement, environment and energetics.


Tag deployments were funded under the collaborative SBT Recruitment Monitoring Program by the Japan Marine Fishery Resources Research Center (JAMARC), the Commonwealth Scientific and Industrial Research Organization (CSIRO), and the Australian Fisheries Management Authority (AFMA). All procedures were approved by the Australian Primary Industries and Water Animal Ethics Committee. We thank the Australian purse-seine industry, as well as the Australian and international long-line fishermen, for their cooperation in returning tags. We also acknowledge the ongoing contribution of many CSIRO staff and particularly thank K. Tattersall for the feed identification work. Thanks to M. Bravington and S. Foster for statistical advice. This study was supported by a joint CSIRO-UTAS QMS scholarship and a CSIRO Postgraduate Award to S.B. C. Davies, G. Hays and two anonymous reviewers provided constructive comments on the manuscript that were greatly appreciated.