Testing optimal foraging theory models on benthic divers

Animal Behaviour 112 (2016) 127e138

Contents lists available at ScienceDirect

Animal Behaviour journal homepage: www.elsevier.com/locate/anbehav

Testing optimal foraging theory models on benthic divers Dahlia Foo a, *, Jayson M. Semmens a, John P. Y. Arnould b, Nicole Dorville b, Andrew J. Hoskins b, Kyler Abernathy c, Greg J. Marshall c, Mark A. Hindell a a

Institute for Marine and Antarctic Studies, University of Tasmania, Hobart, Australia School of Life and Environmental Sciences, Deakin University, Burwood, Victoria, Australia c Remote Imaging Department, National Geographic Television, Washington, D.C., U.S.A. b

a r t i c l e i n f o Article history: Received 10 April 2015 Initial acceptance 7 May 2015 Final acceptance 22 October 2015 Available online MS. number: 15-00297R Keywords: accelerometry Arctocephalus pusillus doriferus benthic foragers biologging marine predators

Empirical testing of optimal foraging models on diving air-breathing animals is limited due to difficulties in quantifying the prey field through direct observations. Here we used accelerometers to detect rapid head movements during prey encounter events (PEE) of free-ranging benthic-divers, Australian fur seals, Arctocephalus pusillus doriferus. PEE signals from accelerometer data were validated by simultaneous video data. We then used PEEs as a measure of patch quality to test several optimal foraging model predictions. Seals had longer bottom durations in unfruitful dives (no PEE) than those with some foraging success (PEE
1). However, when examined in greater detail, seals had longer bottom durations in dives with more PEEs, but shorter bottom durations in bouts (sequences of dives) with more PEEs. Our results suggest that seals were generally maximizing bottom durations in all foraging dives, characteristic of benthic divers. However, successful foraging dives might be more energetically costly (e.g. digestive costs), thus resulting in shorter bottom durations at the larger scale of bouts. Our study provides a case study of how the foraging behaviour of a central place forager foraging in a fairly homogeneous environment, with relatively high travel costs, may deviate from current foraging models under different situations. Future foraging models should aim to integrate other aspects (e.g. diet) of the foraging process for more accurate predictions. © 2015 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.

The ability to acquire resources is crucial for the survival and fitness of animals. Optimal foraging theory (OFT) is a widely used conceptual framework for explaining and predicting foraging behaviours of animals. It attempts to predict how an animal makes foraging decisions to maximize the net rate of energy intake (also known as the ‘currency’ that is being optimized) by minimizing energy costs while maximizing energy gain under relevant constraints in a particular situation (Pyke, Pulliam, & Charnov, 1977; Stephens & Krebs, 1986). Thus, OFT provides testable predictions that can improve our understanding of how animals make foraging decisions to cope in heterogeneous environments where food availability fluctuates spatially and temporally. For air-breathing diving aquatic animals (hereafter divers), including turtles, marine mammals and seabirds that forage in a three-dimensional environment, OFT is also known as optimal diving theory. Optimal diving theory attempts to model how divers modify their time allocation within a dive. A dive is typically broken

* Correspondence: D. Foo, Institute for Marine and Antarctic Studies, University of Tasmania, Private Bag 1, Hobart 7001, Tasmania, Australia. E-mail address: [email protected] (D. Foo).

into four phases: descent, bottom (time assumed to be spent foraging), ascent and a postdive surface interval (SI), when the animal stays on the surface to replenish its oxygen stores before its next dive (Heerah, Hindell, Guinet, & Charrassin, 2014). Bestley, Jonsen, Hindell, Harcourt, and Gales (2014) broadly classified optimal diving models as either physiological or ecological models. Although this dichotomy has limitations, as the optimal diving models already integrate physiological and ecological constraints to some extent, this categorization is useful as it simply considers one type of constraint to be more dominant than the other. We therefore used this dichotomy in a very general sense, while recognizing that it does not affect the fundamental notion of foraging currency in OFT, which in this case is energy for all foraging models mentioned in this paper. Physiological models place emphasis on oxygen depletion of divers (Houston & Carbone, 1992; Kramer, 1988) because, unlike terrestrial animals, divers are ultimately limited by oxygen when they dive. Thus, physiological models assume that within a dive cycle, divers should maximize their bottom duration (i.e. when divers can gain net energy), while minimizing travel duration (i.e. when divers incur a net cost; predictions 1, 2 in Table 1) and/or

http://dx.doi.org/10.1016/j.anbehav.2015.11.028 0003-3472/© 2015 The Association for the Study of Animal Behaviour. Published by Elsevier Ltd. All rights reserved.

128

D. Foo et al. / Animal Behaviour 112 (2016) 127e138

Table 1 Predictions of optimal diving models and optimal foraging models that were tested on Australian fur seals, including the response variable and covariates used for statistical analysis (for each prediction or model) Prediction

Type

Covariates(s)

Source

1 For relatively long travel durations, foraging time decreases with travel duration 2 Proportion of time spent in the foraging area decreases with travel duration

Physiological Bottom duration

Travel duration

Houston and Carbone 1992

Physiological Percentage bottom duration (¼bottom duration/dive duration) Physiological Dive duration

Travel duration

Houston and Carbone 1992

Dive depth and travel duration

Physiological Prey encountered

Bottom duration

Kramer, 1988, Houston and Carbone 1992, Mori et al. 2002 Kramer, 1988

3 Dive duration increases with dive depth and/or travel duration 4 Resource gain (no. of prey encountered) increases linearly with search time spent at depth 5 Postdive surface interval increases as dive duration increases 6 Optimal stay-time should be greater in more productive patches than in less productive patches (dive scale patch quality); however, optimal stay time should be shorter where the environment (bout-scale habitat quality) as a whole is more profitable 7 For deep dives, bottom duration should be largely invariant, no matter the prey density/patch quality

8 Ascent and descent rates should increase with patch quality if seals are reducing transit time

Response variable

Ecological

Postdive surface interval Dive duration

Thompson and Fedak 2001

Ecological

Bottom duration

Charnov, 1976

Ecological

Bottom duration

Ecological

Dive scale patch quality, bout scale patch quality and dive depth (control)

Prey presence or absence (controlled Thompson and Fedak 2001 for travel duration and depth), and the dive scale prey encounter or prey encounter rate Ascent and descent rates Patch quality represented by prey Thompson and Fedak 2001 encounters or prey encounter rate

extend their dive duration when travel duration increases (prediction 3 in Table 1). Therefore, patch quality should be less important to divers primarily constrained by their physiology (Thompson & Fedak, 2001). Simple physiological diving models assume that divers encounter prey at a constant rate in the prey patch, so the number of prey encounters in a dive should increase linearly with bottom duration (prediction 4 in Table 1; Kramer, 1988). Consequently, longer dive durations, longer bottom durations and/or higher dive rates have been used as proxies for increased foraging success and energy gain (Austin, Bowen, McMillan, & Iverson, 2006) even though this may not necessarily be true (Thums, Bradshaw, Sumner, Horsburgh, & Hindell, 2013; Watanabe, Ito, & Takahashi, 2014). For many species, longer dive durations require a longer time on the surface to reoxygenate (prediction 5 in Table 1; Zimmer et al., 2010) reducing the proportion of time spent diving relative to overall time spent at sea (Elliott, Davoren, & Gaston, 2008a). Ecological models consider ecological factors such as prey density, quality and distribution, which are attributed to ‘patch quality’ (Charnov, 1976; Mori, 1998; Mori & Boyd, 2004; Thompson & Fedak, 2001), as primary constraints in foraging. The marginal value theorem (MVT), a classic and influential concept in OFT, is often used to model how an optimal forager allocates its time within a hierarchical patchy environment (however, see Shepard, Lambertucci, Vallmitjana, & Wilson, 2011 who used it to model physiological currencies), whereby smaller-scale, short-term patches of varying patch quality are nested within larger-scale, long-term habitats. The MVT assumes that an animal foraging in small-scale patches will experience patch depletion effects and therefore predicts that a forager should leave all patches, regardless of their profitability, when the instantaneous extraction rate (i.e. ‘marginal value’) reaches the average overall extraction rate for the habitat as a whole (Charnov, 1976). This leads to two opposing predictions: the patch residence time of a forager should be longer in a higher productivity, small-scale patch, but shorter in a higher productivity, large-scale habitat (prediction 6 in Table 1; see Figure 1 in Watanabe et al., 2014). The MVT can be applied to divers, for which individual dives can be considered a small-scale patch, and a series of dives with

relatively short surface intervals between them (bouts) can be considered large-scale habitat. Most studies have shown support for either the short-term (Austin et al., 2006; Sparling, Georges, Gallon, Fedak, & Thompson, 2007) or long-term (Mori & Boyd, 2004; Thums et al., 2013) predictions of the MVT on captive and wild marine predators, while one has recently shown support for
lie penguins, Pygoboth small- and large-scale predictions for Ade scelis adeliae (Watanabe et al., 2014). The model developed by Thompson and Fedak (2001) uses a simple ‘give-up’ rule based on the diver's ability to assess patch quality, while still emphasizing the importance of maximizing bottom duration in a high-quality patch; their model predicts that shallow divers should terminate a dive early in a poor-quality patch, as travel costs are relatively inexpensive. Conversely, deep divers should maximize bottom duration, regardless of patch quality (prediction 7 in Table 1). In addition, Thompson and Fedak (2001) also predicted that divers should increase ascent and descent rates as patch quality increases (prediction 8 in Table 1). Empirical tests of these model predictions are rare due to the lack of data on prey fields, and therefore patch and habitat quality. Measuring the foraging success of free-ranging divers has largely been limited to using proxies such as dive or bottom duration, body condition (Thums et al., 2013) or animal track-based methods (Dragon, Bar-Hen, Monestiez, & Guinet, 2012a, 2012b), in the absence of evidence of actual prey feeding events. Animal-borne video cameras are one of the few practical methods for directly measuring the prey field. However, they are costly, can be difficult to deploy and have limited recording capacity (Biuw, McConnell, Bradshaw, Burton, & Fedak, 2003; Thums, Bradshaw, & Hindell, 2011). Studies that used them have relatively small sample sizes and short-term records (Heaslip, Bowen, & Iverson, 2014) preventing the testing of foraging theory predictions at larger timescales. Alternatively, accelerometers can measure characteristic head and jaw movements of an animal during prey encounter or captures, and also provide longer data records (Hochscheid, Maffucci, Bentivegna, & Wilson, 2005). When used in combination, short-term video evidence of a diver's foraging behaviour can be used to quantitatively validate the prey encounter events (PEE) of free-ranging predators detected by

D. Foo et al. / Animal Behaviour 112 (2016) 127e138

simultaneous long-term accelerometry data (Guinet et al., 2014; Watanabe & Takahashi, 2013). Australian fur seals (hereafter seals), Arctocephalus pusillus doriferus, inhabit an environment (Bass Strait, Australia) with a relatively uniform bathymetry (Arnould & Kirkwood, 2007). Within the Bass Strait, seals are generalist benthic foragers that feed on a variety of prey types, including bony fish, cephalopods and elasmobranchs at depths of 60e80 m (Deagle, Kirkwood, & Jarman, 2009; Kirkwood, Hume, & Hindell, 2008). Benthic environments are generally more stable and less heterogeneous than pelagic environments. This provides a good opportunity to test foraging theory predictions on seals, as their maximum dive depth is predetermined to relatively similar depths, and their typical dive profiles are consistently simple and U-shaped, making identification of the four dive phases straightforward. Our study therefore aimed to test the model predictions given in Table 1 on seals by using a measure of the prey field obtained from PEEs detected from animal-borne head accelerometers, which were validated by simultaneous video evidence from animal-borne video cameras. METHODS Ethical Note The study was carried out with the approval of the Deakin University Animal Ethics committee under Permit No. A14-2011 and in accordance with the Department of Sustainability and Environment (Victoria, Australia) Wildlife Research Permit No. 10005848. Kanowna Island is part of the Wilsons Promontory Marine National Park and was accessed under permit from Parks Victoria. The study was conducted between May and September (winter) in 2011 and 2012 on Kanowna Island, central northern Bass Strait, southeastern Australia (39.1547S, 146.3108E). The island has approximately 10 700 seals, with an annual pup production of approximately 3000 (Kirkwood & Arnould, 2011). The Australian fur seal is a protected species in Australia and is listed as ‘least concern’ by the International Union for Conservation of Nature. During pup rearing, females are central place foragers, returning to the colony regularly. Thus, they were used in this study as they can be recaptured to retrieve deployed data loggers. As the most important demographic component of the population, it is important to understand the ecology of females to understand the effect of environmental changes on the population. Eight adult female seals provisioning pups were randomly selected in the colony. Individuals were approached stealthily so as to not disturb surrounding animals and captured by placing a modified hoop net (Fuhrman Diversified, Seabrook, TX, U.S.A.) over it so that it faced the closed tapered end of the net. This procedure does not harm the animals. Upon capture, individuals were manually restrained and immediately administered isoflurane anaesthesia delivered via a portable gas vaporizer (StingerTM, Advanced Anaesthesia Specialists, Gladesville, NSW, Australia; Gales & Mattlin, 1998). Sedation was generally attained within 5 min, upon which the animal was removed from the net for easier access. Accelerometers were glued to the seals' heads while the GPS data loggers, video cameras and time-depth recorders (TDRs) were glued in series to the dorsal fur along the mid-line posterior to the scapula using a quick-setting epoxy. Seals were equipped with triaxial accelerometers that measured accelerations in the surge (x), sway (y) and heave (z) axes (ca. 3 g, G6A, 40  28  16.3 mm, Cefas Technology Limited, Suffolk, U.K.), GPS data loggers (63  24  22 mm, 31 g, Fastloc 2 GPS data-logger; Sirtrack, Hamilton, New Zealand), TDRs (68  17  17 mm, 30 g, MK9-TDR,

129

Wildlife Computers, Redmond, WA, U.S.A.) and video cameras (Crittercam, 25 cm length  5.7 cm diameter, 1.1 kg, National Geographic Society, Washington, DC, U.S.A.). The accelerometers, TDRs and GPS data loggers sampled at 20 Hz, 1 Hz and every 5 min, respectively. For all animals, the Crittercams were programmed to start and stop recording when seals descended, and ascended past 40 m, on a 1 h on:3 h off duty cycle to maximize battery life while minimizing the potential for missing foraging dives. In total, all devices attached to the seals represented <2% body mass and <1% cross-sectional surface area and probably had negligible additional hydrodynamic drag (Casper, 2009). Data from the GPS data loggers were not included in this study. Instrumentation procedures were usually completed within 45 min of capture. Following the procedures, individuals were allowed to recover from the anaesthetic and resume normal behaviour, and were regularly observed to recommence suckling soon after. There were no observed effects on the behaviour of the mothers or their pups from this procedure. After one or more foraging trips (i.e. 4e8 days) to sea, the animals were recaptured as previously described, anaesthetized briefly (<10 min) and the devices were removed by cutting the fur beneath them. Full sets of overlapping useable data were successfully recovered from five animals. Video Analyses The ultimate objective for quantifying the prey field was to obtain a measure of prey abundance and prey density (and therefore patch quality); therefore we quantified the number of PEEs and PEE per unit time (i.e. prey encounter rate; PER), respectively. Since seals usually only make a rapid head movement during a prey capture attempt or while ingesting it, a PEE was only considered if there was a prey capture attempt, irrespective of whether the prey was successfully caught or not. A PEE typically consisted of prey detection, chase and prey capture attempt, where the seal moved its head rapidly in a darting motion with its jaw open to capture that prey item. Videos were first synchronized with the TDR data. Each video file (representing a single dive) was then examined for PEEs. The timing, duration and capture outcome of all PEEs were recorded. Dive Analyses The data from the TDRs were extracted from the Wildlife Computers proprietary format using Instrument Helper (Wildlife Computers). Zero offset drift in the depth values for each TDR tag was corrected and subsequent summary dive statistics were calculated using a custom-developed R script (R Development Core Team, 2012). Individual dives were defined as any depth exceeding 3 m from the surface (Gentry & Kooyman, 1986). Sequential dives were assigned to bouts when the surface interval was <10 min. This was determined by survival analysis of all the data pooled (Gentry & Kooyman, 1986). Only bouts with at least three dives were included in the analyses to exclude solitary isolated dives. The last dive before the end of a bout was also excluded from the analyses, as the diving behaviour (especially SI) for that dive would be influenced by behaviours other than foraging. Accelerometer Analysis Identifying prey capture attempt signals Acceleration on all three axes (surge, sway and heave) were processed and analysed using the Ethographer package (Sakamoto et al., 2009) with Igor Pro (6.30 J; WaveMetrics, Portland, OR, U.S.A.). The acceleration record was first synchronized with the

130

D. Foo et al. / Animal Behaviour 112 (2016) 127e138

detected, but there was no actual PEE based on video evidence). PEE signals detected on the sway axis at the SD threshold of 0.35 g had the best trade-off between hit rate and false discovery rate; i.e. high proportion of true positives but low proportion of false positives (see Appendix 2 for more details of how this was determined). The PEE signals detected from the sway axis at the optimal SD threshold (0.35 g) were then quantitatively evaluated and validated in more detail. From the videos, periods between PEEs were counted as a non-PEE (Fig. 1e); thus each recorded event was assigned to one of four categories: true positive (TP; PEE signal detected with a PEE); true negatives (TN; non-PEE and no PEE signal detected); false positives (FP; non-PEE but PEE signal detected); and false negatives (FN; PEE but no PEE signal detected) (Fig. 1e). Subsequently, the efficiency of the PEE signals in identifying actual PEEs from changes in acceleration was assessed from two metrics:

depth record from TDRs (see Appendix 1 for details). The process of identifying rapid head movements (hereafter ‘PEE signals’) during PEEs on each accelerometer axes involved (1) applying a 3 Hz highpass filter (Iwata et al., 2012) to the accelerometry data to remove low-frequency head acceleration due to swimming movement (Fig. 1aec). This highlighted the peaks due to dynamic head accelerations. (2) The standard deviation (SD) was then calculated along a moving window of 1.5 s (Viviant, Trites, Rosen, Monestiez, & Guinet, 2010), which smoothed the time series and highlighted extreme accelerations (peaks; Fig. 1c and d). (3) Subsequently, only the SD values that occurred below 40 m (where the video cameras were set to start recording) were retained (Fig. 1d). At this point, the heave (z) axis was dropped from further analyses, as it did not highlight the dynamic accelerations as well as the other two axes visually. (4) PEE signals were extracted by using the mask function in Ethographer to filter out SD values above a threshold. The thresholds ranged from 0.1 to 0.4. (5) PEE signals often occurred in fairly distinguishable clumps ranging from 1.5 to 3.5 s apart, as smaller peaks often occurred on either side of a larger peak. Hence, PEE signals within 3.5 s of each other were combined into a single signal (Fig. 1e).

(1) accuracy (¼[TP þ TN]/[TP þ TN þ FP þ FN]); (2) weighted accuracy (¼[TP þ TN [TP þ FN]/[TN þ FP]]/ [[TP þ FN] þ [TN þ FP][TP þ FN]/[TN þ FP]]). After the acceleration signal for PEEs was validated, the acceleration signal analysis was extended to the complete acceleration record to identify all possible PEE signals. PEE signals could in reality be part of the same PEE since there can be multiple prey

0 –40 –80

(a)

2

–2 (b) 2 0 –2

(c)

0.8 0.35 threshold

0.4 0 (d)

Original signal combined signal answer

TN

TP

FP

FN

TN

TP TN TP

TN

(e) 30

60

90

120 Time (s)

150

180

Mobile standard deviation

High-pass filtered

0

Combine if < 3.5 s apart

SD of Filtered Y axis acceleration (g) acceleration (g) acceleration (g)

Depth (m)

Signal validation PEE signals were then classified into true positives (i.e. signal corresponded to an actual PEE) or false positives (i.e. signal

210

Figure 1. An example of prey encounter events (PEEs) within a 210 s dive by an Australian fur seal showing (a) the depth profile with corresponding raw swaying head accelerations (y axis) recorded on (b). Data were filtered (c) using a high-pass 3 Hz filter to remove noise associated with swimming and the SD (d) was calculated along a moving window of 1.5 s to highlight the dynamic head movement; (e) the timing of PEEs was inferred from SD peaks above the threshold of 0.35 g (‘Original signal’); original signals were combined if they were less than 3.5 s apart (‘Combined signal’); the combined signals were compared to the ‘answers’, which were actual PEEs, and nonprey events (black solid lines) in the videos, resulting in a true positive, true negative, false negative or false positive (denoted by TP, TN, FN and FP, respectively, at the bottom). Vertical dotted lines represent the time when the seal is above 40 m, when acceleration values were unlikely to reflect feeding events.

D. Foo et al. / Animal Behaviour 112 (2016) 127e138

capture attempts during a single PEE; hence PEE signals that were separated by less than 17 s were aggregated into one PEE. This value was selected from the second inflection point in a survival analysis of all the PEE data pooled. Model Fitting Prey encounter events (a proxy of prey abundance) derived from the complete acceleration record, and the derived prey encounter rates (a proxy of prey density; PEE divided by bottom duration) were used as measures of patch/habitat quality. These were used together with simultaneous diving behaviour parameters to test the predictions given in Table 1. We used linear mixed-effect models (LME) and generalized linear mixed models (GLMM) to model the relationships among the variables listed in Table 1 according to the various model predictions. Only benthic dives (88% of all dives) that were part of a bout were used in model fitting. Count (i.e. PEEs) response variables were fitted using a GLMM with a Poisson error distribution while continuous response variables were fitted using LMEs. GLMMs and LMEs were developed with the R version 3.1.0, using the ‘lme4’ and ‘nlme’ package, respectively. All models included seal ID as a random term. All continuous predictors were scaled and centred before fitting to facilitate model convergence and to be able to compare the respective contribution of the predictors (Zuur, Leno, & Smith, 2007). Continuous response variables were also checked for normality and transformed where appropriate. The significance of parameters included in the models was examined by adding the parameter of interest to a null model (Table 1) and assessing the effect of the addition on the fit of the model using likelihood ratio tests and the change in Akaike information criterion (AIC). For prediction 6 (Table 1), this was instead done by removing the parameter of interest from the full prediction model. Significance levels for the likelihood ratio tests were set at a ¼ 0.05. Model selection was not conducted as we were interested only in the specific effect of the predictor (explanatory) variables of interest on the response variables to test model predictions. For prediction 6 (Table 1), both measures of patch quality, i.e. PEE and PER, were used in separate models. Additionally, we examined the effect of PEEs on the postdive surface interval (SI), and whether the seals were terminating their dives immediately after capturing a prey. We tested the latter by examining the effect of the bottom duration before and after the first PEE. Data are presented as mean ± SD unless otherwise stated. RESULTS Video Observations An average of 3.45 ± 0.96 h of video footage, containing 61 ± 32 individual dives and 118 ± 86 prey encounter events (range 20e252) was obtained per seal. The video samples of dives had a mean duration of 2.05 ± 1.02 min (Appendix Table A1). Overall, 77.8% (individual range 50e90.4%) and 64.3% (individual range 50e71.3%) of the PEEs were confirmed PEEs (prey item was in camera's view) and successful captures, respectively. Almost all of the prey captures were on the sea floor. The mean durations of a prey chase and handling were 13.5 ± 17.5 s and 4.02 ± 5.32 s, respectively. Upon catching a prey item, the seals either consumed the prey at the sea floor, indicated by head jerks while swallowing the prey (69%), or ascended to the surface with it (31%). The identifiable prey species were cephalopods, crustaceans, elasmobranchs
guen et al., 2015). or teleost fishes (Kernale

131

Acceleration Signal Validation Using prey encounter event signals obtained from the sway axis at the optimal threshold (0.35 g), 589 PEEs and 836 non-PEE (from video evidence) were classified into 409 true positives, 54 false positives, 782 true negatives and 180 false negatives, resulting in an accuracy and weighted accuracy of 83.6% and 81.5%, respectively. The accelerometer data, therefore, provided a robust index of relative prey field attributes, which was required to test model predictions. Prey Encounter Events During the Foraging Trip A total of 2320 benthic dives (depth
70 m) were extracted from the entire simultaneous acceleration and TDR data set across the five seals. Of the benthic dives, 5.3% were isolated dives (i.e. not part of a foraging bout) and were excluded from analysis. Summary statistics of dive parameters are given in Table 2. Overall, 5526 PEE signals were detected in 76.4% of the benthic dives across the five individuals. Of that number, 3391 PEEs remained after aggregating PEE signals that were within 17 s of each other (Appendix Table A2). Test of Model Predictions Physiological model predictions The dive data supported physiological predictions 1 and 2: seals shortened their bottom duration, both absolutely (Fig. 2a) and proportionally (Fig. 2b), with increasing travel duration. Seals also extended their dive duration when their travel duration increased (Fig. 3a), but not when their maximum dive depth increased, thus partially supporting prediction 3. The data did not fully support the prediction that PEEs should increase linearly with increasing bottom duration (prediction 4); while PEE and bottom duration did have a positive relationship, there was a relatively poor fit (Fig. 3b). Seals did, however, support the prediction that SI increases with dive duration (prediction 5, Table 3, Fig. 3c). Ecological model predictions Support for the MVT varied according to the parameter used to represent patch quality. Using PEEs as a measure of patch quality, seals showed support for the MVT, with their bottom duration (when controlled for dive depth) in the prey patch increasing with dive (small-) scale patch quality, but decreasing with bout (large-) scale habitat quality (prediction 6a, Fig. 4). Conversely, if prey encounter rate was used as the measure of patch quality, the MVT was not supported as bottom duration decreased with dive scale PER, and bout scale PER had little effect on bottom duration (prediction 6b). When maximum depth and travel duration were fixed, bottom duration was slightly longer when no suitable prey were present (Fig. 5), partially supporting the prediction for deep divers (prediction 7). As predicted, seals increased their ascent rate and descent rate for the subsequent dive when they encountered more prey in the dive (prediction 8, Table 3, Fig. 6). Additionally, their SI decreased when they encountered more prey in the dive (Table 3, Fig. 3d), and their bottom duration after the first PEE increased when their first PEE occurred earlier in the dive (Appendix 3, Fig. A2). DISCUSSION Empirical testing of foraging theory on all free-ranging animals is challenging. However, when achieved, it provided substantial insights into animal ecology. New and emerging technologies are

132

D. Foo et al. / Animal Behaviour 112 (2016) 127e138

Table 2 Description, mean and SD of parameters used in testing model predictions across five Australian fur seals Parameter

Description

Mean

SD

Ascent rate (m/s) Bottom duration (s) Bottom duration after first prey encounter (s) Bout scale prey encounter rate Bout scale prey encounters Descent rate (Nþ1) (m/s) Dive duration (s) Dive scale prey encounter rate Dive scale prey encounters % Bottom duration Postdive surface interval (s) Time to first prey encounter Travel duration (s)

Depth of ascent phase/duration of ascent phase Duration of bottom phase Bottom duration remaining after first prey encounter Total prey encounters in the bout/bout duration Total prey encounters in the bout Depth of descent phase/duration of descent phase Total duration of dive Prey encounters in the dive/bottom duration of the dive Prey encounters in the dive Bottom duration/dive duration Duration at the surface between feeding dives (see Methods) Time from the start of the bottom phase to the first prey encounter Ascent durationþdescent duration

1.62 118.19 72.41 0.004 106 1.61 212.53 0.013 1 0.54 118.50 35.97 95.34

0.20 40.40 46.39 0.002 89 0.17 39.27 0.011 1 0.10 72.86 35.23 10.22

(a)

allowing us better access to data needed to test foraging theory empirically, and this is particularly true for divers.

160

Bottom duration (s)

Longer Bottom Durations Do not Mean Greater Foraging Success

140

120

100

80 (56,66]

(76,86]

(106,116]

(b)

% Bottom duration (s)

0.5

0.4

0.3

0.2

(56,66]

(76,86] (106,116] Travel duration (s)

Figure 2. Box plots of fitted values from the model showing the effect of travel duration (binned into 10 s intervals) on (a) absolute bottom duration (prediction 1, Table 3) and (b) squared-transformed percentage bottom duration (¼bottom duration/ dive duration; prediction 2, Table 3) by five Australian fur seals. Ends of whiskers represent the greatest and lowest values, excluding outliers (i.e. values outside 1.5 times above and below the interquartile range). Upper and lower boundaries of the box represent the upper quartile and lower quartile of values. The solid horizontal line within the box represents the median.

There was weak support for the common prediction that resource gain increases linearly with time spent searching at depth (bottom duration; Kramer, 1988; Thompson & Fedak, 2001). While the number of prey encounters increased slightly with bottom duration, the fitted model did not explain a lot of the variation. This result was similar to that found in harbour seals, Phoca vitulina concolor, using video evidence (Heaslip et al., 2014). This suggests that the functional response of seals might be nonlinear, where prey availability decreases with foraging time due to either prey escaping or consumption (Charnov, 1976; Mori & Boyd, 2004; Viviant, Monestiez, & Guinet, 2014; Watanabe et al., 2014). Thus, the common interpretation that increased bottom duration is an indication of increased foraging success may not always be accurate. The data supported predictions for a deep diver (Thompson & Fedak, 2001) with seals' bottom duration being relatively longer in dives with no PEEs than in dives with PEEs. This indicates that the seals tended to continue searching at the bottom for prey instead of terminating the dive prematurely based on some initial assessment of the patch quality. Similar behaviour has been observed in deep-diving mesopelagic short-finned pilot whales, Globicephala macrorhynchus, which also did not shorten their bottom duration in unfruitful dives (Aguilar Soto et al., 2008). In particular, benthic divers have limited foraging habitat, especially in the Bass Strait where productivity is very homogeneous and relatively depauperate (Arnould & Warneke, 2002); hence they should maximize every opportunity to forage when in the benthic zone. Furthermore, dives with no PEEs would be less energetically costly than dives with PEEs, as they require less swimming effort to chase and capture prey, and hence seals can have longer bottom durations in unsuccessful dives. Consistent with previous studies on Australian fur seals (Hoskins, Costa, & Arnould, 2015), our results suggest that seals were maximizing their bottom duration within the benthic foraging zone. Similarly, other benthic divers, including other species of pinnipeds (e.g. Australian sea lions, Neophoca cinerea: Costa & Gales, 2003; Galapagos sea lions, Zalophus wollebaeki: Villegas-Amtmann, Costa, Tremblay, Salazar, & Aurioles-Gamboa, 2008) and seabirds (e.g. emperor penguins, Aptenodytes forsteri: Rodary, Bonneau, Le Maho, & Bost, 2000; Brunnich's guillemots, Uria lomvia: Elliott, Davoren, & Gaston, 2008b) generally indicate that they are maximizing their bottom durations.

D. Foo et al. / Animal Behaviour 112 (2016) 127e138

3 (b)

240 (a) 230

2.5

220 PEE

Dive duration (s)

133

210

2

1.5 200 1

190 180

(56,66]

(76,86]

(96,106] (116,126]

(6,36]

(66,96]

Travel duration (s)

(126,156]

(216,246]

Bottom duration (s)

5.4 (c)

(d)

4.8

5

4.6 SI (s)

4.7

SI (s)

5.2

4.8

4.5 4.4

4.6 4.3 4.4 4.2 (103,133]

(193,223]

(283,313]

Dive duration (s)

0

1

2

3

4

5

6

PEE

Figure 3. Box plots of fitted values from the model showing the effect of (a) travel duration (controlled for dive depth) on dive duration (prediction 3, Table 3); (b) bottom duration of the dive on number of prey encounter events (PEE) (prediction 4, Table 3); (c) dive duration (prediction 5, Table 3) and (d) number of prey encounter events on log-transformed postdive surface interval (SI) by five Australian fur seals. Continuous variables on the x axes (travel, bottom and dive duration) were put into bins for data representation. Ends of whiskers represent the greatest and lowest values, excluding outliers (i.e. values outside 1.5 times above and below the interquartile range). Upper and lower boundaries of the box represent the upper quartile and lower quartile of values. The solid horizontal line within the box represents the median.

Seals Adjusted Diving Behaviour According to Patch Quality Support for the small-scale patch predictions of the marginal value theorem varied according to the measure of patch quality used. For a given depth, bottom duration increased with dive (patch) scale PEEs but decreased with dive scale PER. In other studies that used PER as a measure of patch quality, mesopelagic southern elephant seals, Mirounga leonina (Guinet et al., 2014) and
lie penguins, Pygoscelis adeliae (Watanabe et al., 2014) had Ade longer bottom durations with increasing PER. This difference may be attributed to the type of habitat used: benthic prey occur in relatively low densities within a habitat, whereas mesopelagic prey tend to occur in higher-density patches, providing a richer food source once located (Chilvers & Wilkinson, 2009). The seals in our study did not terminate a dive immediately when they encountered a prey item early in the dive and were capable of capturing multiple prey within a dive. However, many of their dives resulted in only one or two PEEs, meaning that the search time between PEEs would have been relatively high, and would have thus influenced the
lie calculation of PER during bottom duration. In contrast, Ade penguins, which primarily feed on mesopelagic Antarctic krill that

occur in swarms, would have had relatively little search time between prey encounters within a dive, and thus have a wider range of PEEs (range 0e61). One assumption of the MVT is that foragers deplete the exploited patch at a continuous rate (Wajnberg, Bernhard, Hamelin, & Boivin, 2006). However, patches might have stochastic characteristics; for example, when patches contain discrete resource items, such as those in a benthic environment, the time and energy spent searching and sampling the environment have to be taken into account. Thus, by averaging the resource gain over some time interval (e.g. calculating PER), we assumed that the forager is omniscient when it is unlikely to be, resulting in conclusions that differ from the original predictions of the MVT (Wajnberg et al., 2006). Therefore, prey encounter rate might not be a suitable proxy for prey density or patch quality in this case for seals. There was support for the bout (habitat) scale predictions of the MVT where bottom duration decreased with increasing bout scale PEEs. Similarly, southern elephant seals (Bestley et al., 2014; Thums
lie penguins (Watanabe et al., 2014) have et al., 2013) and Ade shorter bottom durations in high-quality habitats. Therefore, the foraging behaviour of seals showed support for the small-scale

134

D. Foo et al. / Animal Behaviour 112 (2016) 127e138

Table 3 Ranked generalized linear mixed-effects models (GLMM) and linear mixed-effects models (LME) used to test foraging theory predictions

DAIC

Prediction

Follows?

Response

Parameter

AIC

1

Yes

Bottom

NULL Travel duration

22128.08 22030.99

0 97.09

NULL Travel duration

4289.16 4881.045

0 591.885

NULL Travel duration Depth

22040.23 22030.99 22038

0 9.24 2.23

2

Yes

% Bottom

Dive duration 3a 3b

Yes No

4

Yes

5

6a

6b

7

8a

8b

Extra

Extra

Somewhat

Yes

No

Yes

Yes

Yes

SI

PEE

Bottom

Bottom

Bottom

Descent rate (Nþ1)

Ascent rate

SI

Bottom duration after first PEE

LL

P

11061.04 11011.5

***

7.770

0.777

***

0.047

0.002

** *

2.449 1.673

0.777 1.150

2147.58 2444.522 11017.12 11011.5 11015

Slope

SE

1154.414 1099.659

***

0.1038

0.0095

0 44.1

3075.7 3052.6

***

0.136

0.020

0 124.2

11057.09 10993

2.671

1.192

***

77.43 34.72

11017.38 11038.73

*** ***

0.085 7.131

0.011 0.743

22122.19 21917.36

0 204.83

11057.09 10952.68

0.985

1.219

***

21917.5 22115.52

204.69 6.67

10953.75 11052.76

*** **

0.208 11.695

1.144 0.811

Depth (NULL) Travel duration (NULL) Prey present

22027.99

0

11008.99

22016.78

11.21

11002.39

**

1.350 7.980 5.965

1.165 0.784 1.873

NULL PEE

1766.787 2033.919

0 267.132

886.3935 1020.9597

***

0.056

0.003

NULL PEE

1698.399 1869.277

0 170.878

852.1993 938.6383

***

0.046

0.003

NULL PEE

2314.827 2237.251

0 77.576

1154.414 1114.625

***

0.079

0.008

7959.483 7739.293

***

22.633

1.005

NULL Dive duration

2314.827 2207.319

NULL Bottom

6155.4 6111.3

Depth (NULL) Full model Term dropped: Bout PEE Dive PEE

22122.19 21997.99 22044.76 22087.47

Depth (NULL) Full model Term dropped: Bout PER Dive PER

Descent duration (NULL) Time to first prey

15926.97 15488.59

0 107.508

0 438.38

Models were fitted according to the foraging model predictions in Table 1 or were extra tests. Parameters of interest are in bold, while those not in bold are part of the null model. Also shown are the maximum log-likelihood (LL), Akaike's information criterion (AIC), the difference in AIC for each model from the null model (DAIC). Log-likelihood ratio tests and its P values were also used to test the significance of a parameter by removing it from the full model (prediction 6) or adding it to the null model. Slope and SE values of the parameters are from the lowest ranking model. Type of transformation on response variables is also indicated. *P < 0.05; **P < 0.01; ***P < 0.001.

patch and large-scale habitat predictions of the MVT, whereby their foraging behaviour changed in opposite directions according to the spatial scale of the foraging patch/habitat quality. Similar changes in foraging behaviour with increasing spatiotemporal scales were also observed in Antarctic fur seals, Arctocephalus gazella (Viviant et al., 2014). Successful foraging can lead to an increase in recovery oxygen consumption as compared to unsuccessful foraging as energy is required for the digestion and warming of prey (Williams, Fuiman, Horning, & Davis, 2004). This suggests that long and successful foraging dives at the short-term dive cycle might be more energetically costly than unsuccessful ones, resulting in seals having to adapt their diving behaviour by decreasing their overall bottom duration at bout level (Viviant et al., 2014). Interestingly, Hoskins and Arnould (2013) reported that Australian fur seals reduced foraging effort across the day and proposed that it might be due to physiological constraints (digestive costs) or prey availability as the day progresses. Our study did not consider different timescales; hence we are unable to confirm whether diurnal changes in prey availability affected foraging behaviour.

It seems contradictory that the seals made shorter dives when they encountered less prey within a dive, but also displayed behaviours that suggested they were maximizing bottom duration even in poor-quality patches (i.e. longer bottom duration when no preferred prey was present). Jackson (2001) reported that while Brants's whistling rats, Parotomys brantsii, might follow simple central place foraging strategies, factors such as time of day and food plant species also influence their foraging behaviour (Jackson, 2001). If benthic divers should generally prioritize maximizing bottom duration regardless of patch quality, and increased prey availability generally results in increased foraging effort (e.g. increased time in patch due to increased handling times; McAleer & Giraldeau, 2006), then bottom duration may be directly influenced by physiological and ecological constraints in different situations. For example, we observed the seals processing larger prey such as cephalopods at the surface whereas they consumed smaller prey at the foraging zone. Such behaviour has been observed in other freeranging pinnipeds as well (Cornick & Horning, 2003); although this strategy is uncommon, the effect of prey type can influence

D. Foo et al. / Animal Behaviour 112 (2016) 127e138

135

140 100 120 50 Bottom duration (s)

100

0

–50

80 60 40

Bottom duration (s)

–100 (a) –2

–1

0

1

2

20

3 0

Dive–PEE

Absent

150 (b)

Present

Figure 5. Mean bottom duration of five foraging Australian fur seals in response to the effect of suitable prey presence (controlled for dive depth and descent duration; prediction 7, Table 3). Error bars represent the 95% pointwise confidence interval.

100

50

0

–50

–100

–200

–100

0 Bout–PEE

50

100

Figure 4. Partial regression plots (i.e. the relationship between the response variable and a predictor, with the other predictors held constant) from a linear mixed model (bottom duration ¼ dive depth þ dive scale prey encounter events (‘Dive-PEE’) þ bout scale prey encounter events (‘Bout-PEE’)) with seal ID as a random factor (prediction 6, Table 3). Dive depth was added as a control. For example, the x axis is the residuals from a linear mixed model of (a) dive-PEE and (b) bout-PEE against the other predictors. The y axis in (a) and (b) is the residuals from a linear mixed model of bottom duration against all predictors excluding dive depth. Black solid lines represent leastsquares regression lines.

foraging behaviour in unexpected ways. Likewise, guillemots tend to dive sequentially to the same depth when high-quality prey patches are discovered, lending support to the hypothesis that sequential dives are influenced by patch quality rather than internal physiology (Elliott et al., 2008b). Thus, the dominant constraints on the foraging behaviour of benthic divers at the dive scale might be too complex to tease apart. Nevertheless, dive scale foraging behaviours still manifested into bout scale foraging behaviours that supported the predictions of the MVT. Seals Swam Faster During Transit when they Encountered Prey Dive duration increased with travel duration, which also resulted in shorter absolute and proportional bottom duration, concurring with previous findings on Australian fur seals (Hoskins &

Arnould, 2013). Since 70% of dives were to consistent depths, a change in travel duration represents a change in travel rate rather than a change in dive depth (Hoskins & Arnould, 2013); thus seals compensated for longer bottom durations by swimming faster during vertical transit, suggesting that seals were maximizing their bottom duration not only for physiological reasons contrary to some optimal diving theory but also for ecological reasons such as patch quality. Similarly, when controlled for dive depth, the dive duration of harbour seals increased with their travel duration (Heaslip et al., 2014). As predicted by Thompson and Fedak (2001), increased prey encounters per dive resulted in increased ascent rate for the current dive and increased descent rate for the subsequent dive. This is evidence that the seals were reducing transit time when in a good prey patch (Thompson & Fedak, 2001), consistent with the findings in other marine diver studies (Gallon et al., 2013; Hanuise, Bost, & Handrich, 2013; Hoskins & Arnould, 2013; Viviant et al., 2014). Shorter SI was associated with more PEEs but shorter dive duration within a dive. Seals may facilitate the continued exploitation of a good prey patch after a successful foraging dive by having a shorter SI. However, because divers are ultimately constrained by their oxygen balance (oxygen gained on the surface is related to the oxygen used in the subsequent dive), a shorter SI would confer a shorter subsequent dive, which would also be facilitated by shorter transit times to good-quality patches. In guillemots, SI may be considered ‘anticipatory’ in short dives or ‘reactive’ in long dives where birds replenish oxygen according to what they need for the following dive, or what they have used from the previous dive, respectively (Elliott et al., 2008b). Southern elephant seals also reduce their SI after dives with prey encounters as opposed to none (Gallon et al., 2013). However, Bestley et al. (2014) found that increased horizontal foraging movement of multiple seal species was associated with longer SI. Successful foraging dives may have higher energy expenditure and thus require longer SI needed to replenish oxygen stores (Bestley et al., 2014). Therefore, the relative duration of SI after a successful foraging dive may simply reflect different foraging strategies of seals.

136

D. Foo et al. / Animal Behaviour 112 (2016) 127e138

natural conditions, as nuances in foraging behaviour differ between species depending on what and where they forage in the water column (Meynier, Morel, Chilvers, Mackenzie, & Duignan, 2014; € tz, 2010). The PEEs Naito, Bornemann, Takahashi, McIntyre, & Plo extracted from the accelerometry data proved to be a valid measure of patch quality to test foraging theory predictions.

(a) 1.9

Ascent rate (m/s)

1.8

1.7

Conclusions

1.6

1.5

1.4 0

1

2

3

4

5

6

(b)

Lactating female seals showed that their foraging behaviour is complex and may be influenced by physiological and/or various ecological factors. Their foraging behaviour also varied according to the spatial scale of prey patch quality. Overall, the foraging behaviour of lactating female Australian fur seals supports the predictions of some optimal foraging models. Our study provides a case study of how the foraging behaviour of a central place forager foraging in a fairly homogeneous environment, with relatively high travel costs may deviate from current optimal foraging models under different situations. Future optimal foraging models should aim to integrate other aspects (e.g. diet) of the foraging process for more accurate predictions.

1.8 Descent rate (N+1) (m/s)

Acknowledgments We thank the numerous volunteers, in particular Kathryn Wheatley and Beth Volpov, who assisted in the field work throughout the study. Logistical support was provided by Parks Victoria and Geoff Boyd (Prom Adventurer Boat Charters). Data from the accelerometer, TDR and timings of actual prey encounter events will be accessible under the Dryad database. Original video files are unavailable due to copyright.

1.7

1.6

References

1.5

0

1

2 3 4 Prey encounter events

5

6

Figure 6. Box plots of fitted values from the model showing the effect of prey encounter events on (a) ascent rate and (b) descent dive duration (prediction 8, Table 3) within a foraging dive of five Australian fur seals. Ends of whiskers represent the greatest and lowest values, excluding outliers (i.e. values outside 1.5 times above and below the interquartile range). Upper and lower boundaries of the box represent the upper quartile and lower quartile of values. The solid horizontal line within the box represents the median.

Validation of Prey Encounter Detection with Accelerometry Data Prey encounter event signals could detect PEEs with a relatively high level of accuracy. Although relatively low, the occurrence of false negatives was none the less greater than that of false positives, which indicated that the detection of PEEs using accelerometers was relatively conservative. False negative PEEs (i.e. PEEs in the videos that the accelerometer failed to detect) were mostly unsuccessful foraging events, and occasionally occurred when seals captured large prey in a swift and smooth process with no rapid head movement. In contrast, head accelerometers attached to
lie penguins produced many false positive PEEs when they Ade were foraging for benthic prey as opposed to pelagic prey, owing to the penguins' head movements when searching the seabed (Watanabe & Takahashi, 2013). Therefore, it is important to validate the acceleration signals for different species, especially under

Aguilar Soto, N., Johnson, M. P., Madsen, P. T., Díaz, F., Domínguez, I., Brito, A., et al. (2008). Cheetahs of the deep sea: deep foraging sprints in short-finned pilot whales off Tenerife (Canary Islands). Journal of Animal Ecology, 77(5), 936e947. Arnould, J. P. Y., & Kirkwood, R. (2007). Habitat selection by female Australian fur seals (Arctocephalus pusillus doriferus). Aquatic Conservation-Marine and Freshwater Ecosystems, 17, S53eS67. Arnould, J. P. Y., & Warneke, R. M. (2002). Growth and condition in Australian fur seals (Arctocephalus pusillus doriferus) (Carnivora: Pinnipedia). Australian Journal of Zoology, 50(1), 53e66. Austin, D., Bowen, W. D., McMillan, J. I., & Iverson, S. J. (2006). Linking movement, diving, and habitat to foraging success in a large marine predator. Ecology, 87(12), 3095e3108. Bestley, S., Jonsen, I. D., Hindell, M. A., Harcourt, R. G., & Gales, N. J. (2014). Taking animal tracking to new depths: synthesizing horizontal-vertical movement relationships for four marine predators. Ecology, 96, 417e427. Biuw, M., McConnell, B., Bradshaw, C. J. A., Burton, H., & Fedak, M. (2003). Blubber and buoyancy: monitoring the body condition of free-ranging seals using simple dive characteristics. Journal of Experimental Biology, 206(19), 3405e3423. Casper, R. M. (2009). Guidelines for the instrumentation of wild birds and mammals. Animal Behaviour, 78(6), 1477e1483. Charnov, E. L. (1976). Optimal foraging, marginal value theorem. Theoretical Population Biology, 9(2), 129e136. Chilvers, B. L., & Wilkinson, I. S. (2009). Diverse foraging strategies in lactating New Zealand sea lions. Marine Ecology Progress Series, 378, 299e308. Cornick, L. A., & Horning, M. (2003). A test of hypotheses based on optimal foraging considerations for a diving mammal using a novel experimental approach. Canadian Journal of Zoology, 81(11), 1799e1807. Costa, D. P., & Gales, N. J. (2003). Energetics of a benthic diver: Seasonal foraging ecology of the Australian sea lion, Neophoca cinerea. Ecological Monographs, 73(1), 27e43. Deagle, B. E., Kirkwood, R., & Jarman, S. N. (2009). Analysis of Australian fur seal diet by pyrosequencing prey DNA in faeces. Molecular Ecology, 18(9), 2022e2038. Dragon, A. C., Bar-Hen, A., Monestiez, P., & Guinet, C. (2012a). Comparative analysis of methods for inferring successful foraging areas from Argos and GPS tracking data. Marine Ecology Progress Series, 452, 253e267. Dragon, A. C., Bar-Hen, A., Monestiez, P., & Guinet, C. (2012b). Horizontal and vertical movements as predictors of foraging success in a marine predator. Marine Ecology Progress Series, 447, 243e257.

D. Foo et al. / Animal Behaviour 112 (2016) 127e138 Elliott, K. H., Davoren, G. K., & Gaston, A. J. (2008a). Increasing energy expenditure for a deep-diving bird alters time allocation during the dive cycle. Animal Behaviour, 75, 1311e1317. Elliott, K. H., Davoren, G. K., & Gaston, A. J. (2008b). Time allocation by a deepdiving bird reflects prey type and energy gain. Animal Behaviour, 75, 1301e1310. Gales, N. J., & Mattlin, R. H. (1998). Fast, safe, field-portable gas anesthesia for otariids. Marine Mammal Science, 14(2), 355e361. Gallon, S., Bailleul, F., Charrassin, J. B., Guinet, C., Bost, C. A., Handrich, Y., et al. (2013). Identifying foraging events in deep diving southern elephant seals, Mirounga leonina, using acceleration data loggers. Deep-Sea Research Part IiTopical Studies in Oceanography, 88e89, 14e22. Gentry, R. L., & Kooyman, G. L. (1986). Fur seals: Maternal strategies on land and at sea. Princeton, NJ: Princeton University Press. Guinet, C., Vacquie-Garcia, J., Picard, B., Bessigneul, G., Lebras, Y., Dragon, A. C., et al. (2014). Southern elephant seal foraging success in relation to temperature and light conditions: insight into prey distribution. Marine Ecology Progress Series, 499, 285e301. Hanuise, N., Bost, C. A., & Handrich, Y. (2013). Optimization of transit strategies while diving in foraging king penguins. Journal of Zoology, 290(3), 181e191. Heaslip, S. G., Bowen, W. D., & Iverson, S. J. (2014). Testing predictions of optimal diving theory using animal-borne video from harbour seals (Phoca vitulina concolor). Canadian Journal of Zoology, 92(4), 309e318. Heerah, K., Hindell, M., Guinet, C., & Charrassin, J.-B. (2014). A new method to quantify within dive foraging behaviour in marine predators. PLoS One, 9(6), e99329. Hochscheid, S., Maffucci, F., Bentivegna, F., & Wilson, R. P. (2005). Gulps, wheezes, and sniffs: how measurement of beak movement in sea turtles can elucidate their behaviour and ecology. Journal of Experimental Marine Biology and Ecology, 316(1), 45e53. Hoskins, A. J., & Arnould, J. P. Y. (2013). Temporal allocation of foraging effort in female Australian fur seals (Arctocephalus pusillus doriferus). PLoS One, 8(11), e79484. Hoskins, A. J., Costa, D. P., & Arnould, J. P. Y. (2015). Utilisation of intensive foraging zones by female Australian fur seals. PLoS One, 10(2), e0117997. Houston, A. I., & Carbone, C. (1992). The optimal allocation of time during the diving cycle. Behavioral Ecology, 3(3), 255e265. Iwata, T., Sakamoto, K. Q., Takahashi, A., Edwards, E. W. J., Staniland, I. J., Trathan, P. N., et al. (2012). Using a mandible accelerometer to study fine-scale foraging behavior of free-ranging Antarctic fur seals. Marine Mammal Science, 28(2), 345e357. Jackson, T. P. (2001). Factors influencing food collection behaviour of Brants' whistling rat (Parotomys brantsii): a central place forager. Journal of Zoology, 255, 15e23.
guen, L., Dorville, N., Ierodiaconou, D., Hoskins, A. J., Baylis, A. M. M., Kernale Hindell, M. A., et al. (2015). From video recordings to whisker stable isotopes: a critical evaluation of timescale in assessing individual foraging specialisation in Australian fur seals. Oecologia, 1e14. Kirkwood, R., & Arnould, J. P. Y. (2011). Foraging trip strategies and habitat use during late pup rearing by lactating Australian fur seals. Australian Journal of Zoology, 59(4), 216e226. Kirkwood, R., Hume, F., & Hindell, M. (2008). Sea temperature variations mediate annual changes in the diet of Australian fur seals in Bass Strait. Marine Ecology Progress Series, 369, 297e309. Kramer, D. L. (1988). The behavioral ecology of air breathing by aquatic animals. Canadian Journal of Zoology, 66(1), 89e94. McAleer, K., & Giraldeau, L. A. (2006). Testing central place foraging in eastern chipmunks, Tamias striatus, by altering loading functions. Animal Behaviour, 71, 1447e1453. Meynier, L., Morel, P. C. H., Chilvers, B. L., Mackenzie, D. D. S., & Duignan, P. J. (2014). Foraging diversity in lactating New Zealand sea lions: insights from qualitative and quantitative fatty acid analysis. Canadian Journal of Fisheries and Aquatic Sciences, 71(7), 984e991. Mori, Y. (1998). The optimal patch use in divers: optimal time budget and the number of dive cycles during bout. Journal of Theoretical Biology, 190(2), 187e199. Mori, Y., & Boyd, I. L. (2004). The behavioral basis for nonlinear functional responses and optimal foraging in Antarctic fur seals. Ecology, 85(2), 398e410. Mori, Y., Takahashi, A., Mehlum, F., & Watanuki, Y. (2002). An application of optimal diving models to diving behaviour of Brünnich's guillemots. Animal Behaviour, 64(5), 739e745. €tz, J. (2010). Fine-scale Naito, Y., Bornemann, H., Takahashi, A., McIntyre, T., & Plo feeding behavior of Weddell seals revealed by a mandible accelerometer. Polar Science, 4(2), 309e316. Pyke, G. H., Pulliam, H. R., & Charnov, E. L. (1977). Optimal foraging e a selective review of theory and tests. Quarterly Review of Biology, 52(2), 137e154. R Development Core Team. (2012). R: A language and environment for statistical computing. Vienna, Austria: R Foundation for Statistical Computing.

137

Rodary, D., Bonneau, W., Le Maho, Y., & Bost, C. A. (2000). Benthic diving in male emperor penguins Aptenodytes forsteri foraging in winter. Marine Ecology Progress Series, 207, 171e181. Sakamoto, K. Q., Sato, K., Ishizuka, M., Watanuki, Y., Takahashi, A., Daunt, F., et al. (2009). Can ethograms be automatically generated using body acceleration data from free-ranging birds? PLoS One, 4(4), e5379. Sato, K., Mitani, Y., Cameron, M. F., Siniff, D. B., & Naito, Y. (2003). Factors affecting stroking patterns and body angle in diving Weddell seals under natural conditions. Journal of Experimental Biology, 206(9), 1461e1470. Shepard, E. L., Lambertucci, S. A., Vallmitjana, D., & Wilson, R. P. (2011). Energy beyond food: foraging theory informs time spent in thermals by a large soaring bird. PLoS One, 6(11), e27375. Sparling, C. E., Georges, J. Y., Gallon, S. L., Fedak, M., & Thompson, D. (2007). How long does a dive last? Foraging decisions by breath-hold divers in a patchy environment: a test of a simple model. Animal Behaviour, 74, 207e218. Stephens, D. W., & Krebs, J. R. (1986). Foraging theory. Princeton, NJ: Princeton University Press. Thompson, D., & Fedak, M. A. (2001). How long should a dive last? A simple model of foraging decisions by breath-hold divers in a patchy environment. Animal Behaviour, 61(2), 287e296. Thums, M., Bradshaw, C. J. A., & Hindell, M. A. (2011). In situ measures of foraging success and prey encounter reveal marine habitat-dependent search strategies. Ecology, 92(6), 1258e1270. Thums, M., Bradshaw, C. J. A., Sumner, M. D., Horsburgh, J. M., & Hindell, M. A. (2013). Depletion of deep marine food patches forces divers to give up early. Journal of Animal Ecology, 82(1), 72e83. Villegas-Amtmann, S., Costa, D. P., Tremblay, Y., Salazar, S., & Aurioles-Gamboa, D. (2008). Multiple foraging strategies in a marine apex predator, the Galapagos sea lion Zalophus wollebaeki. Marine Ecology Progress Series, 363, 299e309. Viviant, M., Monestiez, P., & Guinet, C. (2014). Can we predict foraging success in a marine predator from dive patterns only? Validation with prey capture attempt data. PLoS ONE, 9(3), e88503. Viviant, M., Trites, A. W., Rosen, D. A. S., Monestiez, P., & Guinet, C. (2010). Prey capture attempts can be detected in Steller sea lions and other marine predators using accelerometers. Polar Biology, 33(5), 713e719. Wajnberg, E., Bernhard, P., Hamelin, F., & Boivin, G. (2006). Optimal patch time allocation for time-limited foragers. Behavioral Ecology and Sociobiology, 60(1), 1e10. Watanabe, Y. Y., Ito, M., & Takahashi, A. (2014). Testing optimal foraging theory in a penguin-krill system. Proceedings of the Royal Society B: Biological Sciences, 281(1779), 20132376. Watanabe, Y. Y., & Takahashi, A. (2013). Linking animal-borne video to accelerometers reveals prey capture variability. Proceedings of the National Academy of Sciences of the United States of America, 110(6), 2199e2204. Williams, T. M., Fuiman, L. A., Horning, M., & Davis, R. W. (2004). The cost of foraging by a marine predator, the Weddell seal Leptonychotes weddellii: pricing by the stroke. Journal of Experimental Biology, 207(6), 973e982. Zimmer, I., Wilson, R. P., Beaulieu, M., Ropert-Coudert, Y., Kato, A., Ancel, A., et al. (2010). Dive efficiency versus depth in foraging emperor penguins. Aquatic Biology, 8(3), 269e277. Zuur, A. F., Leno, E. N., & Smith, G. M. (2007). Analysing ecological data. Springer.

Appendix 1. Synchronization of acceleration and depth records For each seal, the acceleration record was synchronized with the depth record (from TDRs) by calculating pitch (body orientation) and matching it to the dive depth profiles. Pitch was derived by isolating the static component (low-frequency readings due to gravity) in the surge (x) acceleration axis from the dynamic component (high-frequency readings due to animal movement) using a 0.1 Hz low-pass filter (Sato, Mitani, Cameron, Siniff, & Naito, 2003). The pitch reflected the device angle of the logger itself and thereby reflected the animal's lateral posture. Descents were represented as negative pitch and ascents by positive pitch values (Sato et al., 2003). Appendix 2. Determination of optimal acceleration axis and threshold for detecting prey capture attempts A TP was assigned if at least 50% of a combined PEE signal overlapped with the duration plus 4.5 s to the start and end of an ‘answer’ (Fig. 1e); otherwise, it was considered an FP. Next, for each

D. Foo et al. / Animal Behaviour 112 (2016) 127e138

axis and threshold combination, the hit rate (¼number of ‘answers’ that corresponded with at least 1 TP/total number of ‘answers’) and false discovery rate (¼FP/(FP þ TP)) were determined. The optimal axis and threshold combination was the one that maximized the sum of the hit rate and precision (¼1  false discovery rate), which was the sway (y) axis acceleration at the threshold of 0.35 g (Fig. A1); only its associated PEE signals were used for further analysis.

Appendix 3. Testing whether seals terminate a dive immediately after a single prey encounter event

100 Bottom duration after first prey (s)

138

Hit rate + precision

2

1.5

1

0.5

80 60 40 20 0 –20

0 0.1

0.2 0.3 Threshold

(0,10]

0.4

Figure A1. The sum of hit rate and precision (¼1  false discovery rate), potentially ranging from 1 (a random guess) to 2 (ideal signal), plotted against the threshold in the head-only acceleration (g) on the surge (x) and sway (y) axes for the signal of prey encounters (with prey capture attempt). Open circles represent the surge (x) axis; solid circles represent the sway (y) axis. The arrow points to the optimal threshold (0.35 g) determined on the sway (y) acceleration axis.

(60,70] (130,140] Time to first prey (s)

Figure A2. Box plots of fitted values from the model showing the effect of the time to the first prey encounter event (controlled for descent duration) on the bottom duration after the first prey of five Australian fur seals. Ends of whiskers represent the greatest and lowest values, excluding outliers (i.e. values outside 1.5 times above and below the interquartile range). Upper and lower boundaries of the box represent the upper quartile and lower quartile of values. The solid horizontal line within the box represents the median.

Appendix 4. Summary of deployment information and predicted prey encounter events from accelerometry data Table A1 Female Australian fur seal identity, deployment and recovery dates, mass, total video, acceleration and TDR record durations, number of video recorded dives, total number of prey encounter events in the videos and number of successful prey captures from all types of prey encounter events in the videos Seal ID

Deployment

Recovery

Mass (kg)

Total video duration (h)

Video recorded dives (N)

Recorded prey encounter events

Recorded successful prey captures

Total acceleration duration (h)

W1855 W1859 W1873 W1881 W1905 Mean±SD Total

15 16 26 15 17

21 May 2011 20 May 2011 1 June 2011 25 July 2011 7 June 2012

50.5 54.5 88 55.5 78 65.3±16.7

2.24 4.67 3.99 2.79 3.56 3.45±0.96 17.3

57 89 93 53 13 61±32 305

84 94 252 139 20 118±86 589

42 67 173 84 13 76±61 379

144 83 148 192 104 134±42.3 671

May 2011 May 2011 May 2011 June 2011 May 2012

Table A2 Summary of the number of prey encounter events retrieved from the accelerometry data

Seal W1855 W1859 W1873 W1881 W1905 Total Overall mean Overall SD

Total

Per day

Mean 289 617 744 1543 198 3391

Mean 58 206 149 257 50 719

678 533

144 91

Per hour SD 57 63 102 80 53

Mean 7 11 12 20 6 55 11 6

Per dive SD 5 5 9 13 4

Mean 1 1 2 3 2 9 2 0

SD 1 1 1 1 1