SEARCH

SEARCH BY CITATION

Keywords:

  • Atlantic salmon;
  • ideal despotic distribution;
  • optimal foraging;
  • patch choice;
  • sampling

Summary

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  • 1
    Models linking the behaviours of individual animals, their positions within socially complex groups and spatio-temporal variation in resource distribution offer a promising base for predicting population responses to changing environments. The ideal free and despotic distributions and their derivatives are particularly influential in this regard.
  • 2
    Due to the difficulties of conducting work in the wild, for some groups of animals such models are often based on observations of animals in small-scale systems under conditions that are well controlled, but unnaturally simple.
  • 3
    Using an experimental system based on field observations of home range size and variation in food availability, the present study tested whether models derived using small-scale laboratory observations are valid for juvenile Atlantic salmon in more natural conditions.
  • 4
    Contrary to predictions, we found no differences in behaviour between the control fish (which experienced consistently rich feeding patches) and the experimental fish (which experienced unpredictable 10-fold changes in patch quality).
  • 5
    Also contrary to predictions, in the variable condition, salmon used high quality patches (which were an order of magnitude better than low quality patches) only marginally (5%) more than would be expected if they were to forage at random. There was significant variation in foraging strategies between individual fish, with 28% of the population making non-random use of foraging patches.
  • 6
    The only apparent systematic relationship between social rank and use of foraging patches was that fish that were both dominant and made many moves between feeding locations tended to leave rich patches less frequently than they left poor patches.
  • 7
    Despite the low correlation between patch quality and movement, there was substantial movement of fish among patches. Forty-four per cent of moves followed aggressive interactions and most others were spontaneous, with no obvious motivating factor apparent.
  • 8
    The study exposes a discrepancy between expectations derived from the basic concepts of patch choice theory and the behaviour of Atlantic salmon in the conditions pertaining in the present study.
  • 9
    It is suggested that this discrepancy may arise both from the fact that applicability of patch choice models may be very sensitive to the stability of differences in patch quality and from uncertainties about the costs of habitat sampling.

Introduction

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

Development of models that examine the relationship between the behavioural choices of individual animals, their growth, reproduction and survival and the ecology of populations is an area of intense current theoretical endeavour (Sutherland 1996; Clarke & Mangel 2000; Giraldeau & Caraco 2000; Green & Stamps 2001; Krause & Ruxton 2002). Two aspects of behaviour that have been included in several such models are an individual's ability to track profitable feeding patches in a variable environment and the effects of social interactions on foraging decisions (Tregenza, Hack & Thompson 1996). For example, the Ideal Despotic Distribution (Fretwell 1972) considers how movement of animals is constrained by intraspecific competition in an environment with a spatially patchy but static resource and has been used to predict future behaviour of both individuals were populations (Calsbeek & Sinervo 2002; Tome 2003; Zimmerman, La Haye & Gutierrez 2003).

Ruxton, Armstrong & Humphries (1999) extended existing models based on the Ideal Despotic Distribution with patches of fixed quality by exploring how animals of different social status distribute themselves among arrays of discrete patches that fluctuate in quality in time and space. Like similar models (e.g. Bernstein, Kacelnik & Krebs 1988; Abrams 2000; Gill, Sutherland & Norris 2001), this assumes that foragers seek to track profitable feeding patches, but that their ability to do this depends on social rank. Testing these assumptions requires detailed information on how animals of known rank respond to fluctuating resource distribution and we report here on a study designed to provide such information for an extensively used model system, the juvenile Atlantic salmon (Salmo salar L.).

Atlantic salmon spawn in streams and rivers during autumn and winter and their eggs hatch during the following spring. On dispersal from the nest, young fish take up residence in streams where they remain for 1 to 7 years before many of them migrate to sea (Metcalfe & Thorpe 1990). Survival and growth in fresh water are important determinants of the number of fish migrating to sea and subsequently returning as adults. In many streams, the foraging areas of juvenile salmon are separated by large boulders, so in these cases it is appropriate to model the habitat in terms of discrete feeding patches.

In simple laboratory tanks, aggressive, dominant individuals tend to monopolize food and space (Metcalfe et al. 1989; Gotceitas & Godin 1992) and to grow faster relative to subordinates, in spite of higher metabolic rates (Metcalfe, Taylor & Thorpe 1995). In a laboratory flume, Atlantic salmon parr adjusted their choice of local foraging site within a fixed resource distribution such that rank correlated with patch quality (Fausch 1984). Similarly, in natural pools where food availability tends to be consistently highest towards the upstream ends, ranks of salmonid parr matched quality of spatial position (Hughes 1992; Nakano 1995). Groups of Amago trout (Oncorhynchus masou ishikawe) offered a choice between two feeding patches of different profitability made preferential use of the better quality patch and after 4 weeks their distribution was as predicted by the Ideal Despotic Distribution (Hakoyama & Iguchi 2001). Therefore, the existing data on juvenile salmon in simple, laboratory conditions with spatially fixed food distributions are consistent with rank-related tracking of changes in resource distribution as predicted by this model.

However, observational studies in the wild have produced some problematic results. Using Passive Integrated Transponder (PIT) tags in a natural stream it has been shown that, although large juvenile salmon have a high degree of overall site attachment (Armstrong et al. 1994), on a finer scale they move between several feeding sites dispersed over c. 1–10 m lengths of stream (Armstrong, Huntingford & Herbert 1999; Martin-Smith & Armstrong 2002). Furthermore, it has been shown recently that in natural riffles the relative availability of drifting food can change over a time-scale of a few hours or less, independent of variation in water discharge (Martin-Smith & Armstrong 2002). The concept of individual fish concentrating their foraging in a single patch of fixed quality may not apply in these circumstances. Additionally, PIT tag studies have shown that the home ranges of individual fish may overlap extensively (Armstrong et al. 1999; Martin-Smith & Armstrong 2002) and several studies of juvenile salmon in natural streams have shown that dominant fish may not be at a growth advantage (Martin-Smith & Armstrong 2002; Sloman & Armstrong 2002; Harwood et al. 2003). The concept of dominant, despotic fish monopolizing favourable feeding areas, thereby enjoying better growth than subordinates, may also not apply.

To resolve these issues it is necessary to examine whether and how salmon of known rank track variable food supplies in natural conditions. This requires a test arena of suitably large scale, but also sufficient control for observed behaviours to be matched unambiguously with variations in food distributions. To achieve this objective, we used replicated sections of an artificial stream that were large enough to accommodate typical home ranges of juvenile salmon in the natural stream used in recent studies (Armstrong, Shackley & Gardiner 1994; Armstrong et al. 1999; Martin-Smith & Armstrong 2002) and a computer-based feed delivery system to generate frequent but unpredictable changes in patch quality, as in the riffle areas of natural streams. The study addressed the following questions for fish foraging in small groups:

  • • 
    Do juvenile Atlantic salmon make preferential use of profitable patches?
  • • 
    If not, what causes a fish to leave a patch in which it has been feeding?
  • • 
    Does social rank influence the ability or willingness of juvenile salmon to track profitable feeding sites?

Materials and methods

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

general protocol

The general procedure was to capture salmon parr that had been reared in the wild, tag and mark these fish, and then release them in groups of four into sections of a glass-sided indoor stream within which the relative qualities of feeding patches were controlled using a computer-operated feeding system. Gross movements of fish between feeding patches were recorded automatically using a PIT tag detecting system and finer-scale behaviour was recorded by direct visual observation. The fish were then removed to measure their growth in the arenas and introduced into a standard procedure for assessing dominance ranks. In total, three runs were conducted, each using four sections of the indoor stream (Table 1). Each run included a single control in which feeding patches were similar and constant over time, and three treatment replicates within which patch qualities varied in time and space. Thus, there were nine treatment replicates and three controls. Details of each component of the trials are given below.

Table 1.  Weights (mean and range) of fish used in each run and section of experiment. Dates of each run are given in the first column
Run (dates)SectionFish weight (g), mean (range)
1 (03–17 August 2001)A14·7 (13·4–16·0)
B15·5 (13·7–17·0)
C15·6 (13·8–17·5)
D13·7 (13·1–14·9)
2 (17–31 August 2001)A18·1 (17·2–19·9)
B18·2 (15·4–21·5)
C18·2 (16·0–21·5)
D17·6 (16·8–18·5)
3 (31 August−14 September 2001)A16·6 (14·6–17·9)
B17·3 (16·6–17·6)
C16·8 (15·0–18·6)
D17·8 (15·3–20·1)

the experimental arenas

Experiments were conducted in a large flow-through artificial stream, comprising a continuous channel 50 m in length and 1·55 m wide, at the Fisheries Research Services Freshwater Laboratory field station at Almondbank, Perthshire. Plate glass windows running along one side allowed observers, shaded by a hide, to watch fish without disturbing them. The channel was lined with a butyl rubber membrane covered by a layer of coarse gravel, with white plastic markers delineating a grid of squares 30 × 30 cm. Lighting was provided by 400-W Philips SON Agro bulbs (6000 lux), suspended 1·8 m above the gravel surface at 1·8 m intervals along the length of the stream, set to an ambient photoperiod. A drum filter (mesh size 0·5 mm) removed much of the natural drift from the input water, diverted from the River Almond. Mean water depth (21 cm ± 1·4 SD) and velocity (22 cm s−1 ± 3 SD) were within the preferred range of Atlantic salmon parr (Heggenes, Baglinière & Cunjak 1999).

The stream was divided into four replicate sections, each measuring 7·3 m × 1·55 m; the layout of a single section is shown in Fig. 1. Neighbouring sections were separated by stainless-steel screens (mesh size 5 mm) projecting 40 cm above water level. Each section was landscaped as four patches separated by rows of boulders; water depth above the boulders was approximately 10 cm. Three shelters (sheets of grey PVC 20 × 15 cm) were positioned against the windows in each patch. A flat cobble downstream of the feeder outlet (see below) was provided as a feeding station.

image

Figure 1. The layout of one section of the artificial stream, showing four discrete feeding patches separated by rows of boulders. Within each patch, the area enclosed by the dotted line was classified as central; the area outside the dotted line was marginal.

Download figure to PowerPoint

the fish

Our subjects were 1 + Atlantic salmon that had been stocked in spring 2000 as fry in a tributary of the River Braan, Scotland, above an impassable waterfall and with no natural salmon population. Fish were caught by electro-fishing at the start of each of three runs (see Table 1 for dates) and transported to the Almondbank field station. After being held in a tank for no more than 24 h, they were anaesthetized (Benzocaine in alcohol), weighed (to 0·1 g) and given a subcutaneous mark with alcian blue dye (a standard, benign marking technique for juvenile salmon, e.g. Kelly 1967). A PIT-tag was inserted into the body cavity through a small incision sealed with a 50 : 50 mix of CicatrinTM antibiotic powder (Wellcome Foundation Ltd, London, UK) and OrahesiveTM Protective Powder (ER Squibb & Sons, Hounslow, UK). After recovery, four fish were introduced to each of the four sections of the artificial stream, giving a density of 0·35 fish m−2. This is a natural density for salmon of this age in a lower tributary of the river system from which they were collected (Egglishaw & Shackley 1977). The sizes of fish in the four sections were matched as far as possible within runs (Table 1). After 2 weeks, the fish were removed by electro-fishing, re-weighed and re-marked prior to the start of dominance ranking.

food and feeding regimen

An under-gravel food supply pipe emerged in the centre of each patch, 16 cm above the gravel surface. Food suspended in water (dead chironomid larvae, a natural drifting prey of juvenile salmon) emerged from the pipe outlet, from which it drifted downstream. Uneaten food was caught in a net (width 50 cm; mesh 1 mm) 70 cm downstream of the outlet, thus creating discrete feeding patches. A coarse mesh (10 mm) positioned across the front and back of the net prevented access by fish to prey items caught in the net. The nets and end screens were cleaned daily prior to the commencement of the feeding period.

Food delivery to each patch was controlled by a computerized feeder system (MacLean et al. 2003). Food was delivered daily from 10.30 h to 15.30 h. This feeding time was divided into 10 30-min periods. During each 30-min period, food availability in each patch was set independently at one of two levels: rich (one delivery every 30 s of, on average, 0·86 larvae) or poor (one food delivery every 300 s). In each run, all the patches in one section (the control section, which occupied a different position in each run) were set at the rich level throughout. In the other three sections (the experimental sections), patch quality was variable. Over the whole run, every patch was at the rich level for approximately 50% of the time, but changes between rich and poor levels occurred randomly at the start of any 30-min feeding period and were therefore unpredictable. The total amount of food delivered per patch per day at the rich level approximated to the known requirements of experimental fish, based on the known composition of chironomid larvae and on rations recommended in fish farming feed tables for salmon of the same size and at the experimental temperatures (I. D. McCarthy, personal communication).

the pit tag system and analysis of pit records

To monitor the location and activity of the fish, a flatbed PIT detector (active area 97 cm × 33 cm) was buried under a thin layer of gravel in the centre of each patch, with the feeder outlet positioned in the centre of the detector. Thus, fish registered on the detector when in a position to intercept food items. The detectors were linked to a computer via a bank of PIT decoders. Whenever a fish was within range of a detector, date, time, detector number and unique identity code (from the PIT tag implanted in the fish's body cavity) were saved to a computer file. To reduce multiple records from fish holding station continuously above detectors and therefore to keep data files to a manageable size, detectors were programmed to register each tag only once during each 15-s period. Inactive fish did not hold position above detectors, but sheltered around the edges of the patch. Data gathered by the PIT system were used to assess overall activity levels and differential use of patches with rich and poor feed delivery rates.

Differential use of feeding patches was examined in two ways, as follows.

1. Based on time spent in patches receiving high and low levels of feed delivery

To assess activity during the time when food was available, the feeding period was divided into 5-min intervals (to match the rate of food delivery at the poor level). For each 5-min interval, if a fish ever came within range of a detector, it was considered to be actively foraging in that patch, as opposed to taking shelter on the edge of a patch. If fish apportion their foraging randomly irrespective of feed level, then the proportion of time they spend foraging at the rich level should be dictated by the overall availability of rich patches in the relevant section over the whole course of the run. Thus:

  • Expected proportion of time at rich level (expected time proportion) = (Total number of rich periods provided in patches A + B + C + D) (Total number of periods × 4)

This expected time proportion varied between 0·45 and 0·51 because the number of rich periods in each section was subject to a random process.

For each individual, a G-test was used to compare the actual number of 5-min intervals spent at the rich and poor levels with the expected number, where:

  • Expected no. of rich intervals = Expected time proportion × total no. of active intervals

and

  • Expected no. of poor intervals = Total no. of active intervals − expected no. of rich intervals.

The sequential Bonferroni correction for 36 cases was used to identify statistically significant results.

To quantify the tendency of individual fish to concentrate foraging in rich patches, the proportion of active intervals that each fish spent at the rich level (observed time proportion) was expressed as a deviation from the expected time proportion (based on random patch use, see above), to give a time preference index. A positive score on this index represented a preference for patches at the rich level, while a negative score represented a preference for patches at the poor level. A score of zero indicated patch use that was random with respect to feed delivery level. A one-sample t-test was used to assess, for the population as a whole, the extent to which this time preference index differed from zero. In order to identify any effect of activity level on preference for rich patches, the total number of active 5-min intervals for each fish was expressed as a proportion of the total number of 5-min intervals available during the feeding period, to give an activity index. The relationship between time preference and activity indices was then investigated using linear and non-linear regression.

2. Based on movement away from patches at the two feed levels

On the basis of the focal observations, movements between patches that were part of a bout of movement (i.e. occurring ≤ 3 min after a previous move, see Results) were deemed to be unrelated to foraging and were not included in subsequent analyses of foraging-related movements. On this basis, a fish was considered to be foraging in a given patch if it had spent at least 3 min there; it was deemed to have moved from that patch at the time of its last PIT registration on it before registering elsewhere. Individuals that moved away from a given category of patch more often than expected were identified using G-tests. These compared the number of times each fish left patches in the rich state (summed over the entire run) with the number of times expected had they moved away without reference to food availability. To prevent differences in time proportion from confounding the analysis, the expected number of moves was calculated by multiplying the total number of moves made by each fish by the proportion of time each spent on patches in the rich condition (the observed time proportion). Four fish were omitted from the analysis as they performed a small (< 10) number of moves. The sequential Bonferroni criterion for 32 cases was used to identify statistically significant results.

To quantify the tendency of individual fish to leave patches receiving the rich level of food delivery, the proportion of moves from rich patches (leave proportion) was expressed as a deviation from the expected proportion, based as above on the total number of moves made by each fish and the proportion of time each spent on patches in the rich condition. This was labelled the leave preference index. A positive score represented a tendency to leave patches in the rich condition, a negative score represented a tendency to leave patches in the poor condition, while a score of zero indicated a tendency to leave without respect to feed delivery level.

The total number of moves performed during the whole run by a given fish was expressed as the number of moves per hour of feed presentation, to give movement rate. To investigate the effect of this overall movement rate on the tendency to leave preferentially patches at the rich or poor feed delivery level, the relationship between the leave preference index and movement rate was examined using an analysis of covariance, with rank as a fixed factor and movement rate as the covariate.

focal observations and analysis of observational data

Focal observations were used to assess food intake at the varying rates of food delivery and to identify the precursors of movements between patches. Focal observations were conducted during the feeding periods (from 10.30 h to 15.30 h) throughout each run on days when the water was sufficiently clear to identify individual fish from their dye marks. Hence, the number of observations per section varied from 10 to 17. Sections and fish within each section were selected for focal observation randomly, without replacement. The behaviour of the focal fish was recorded over a 20-min period using an event-recorder (FIT-system, Smile Design, Zurich) installed on a palm-top computer (PalmPilot, 3Com). The locations of the focal fish, and any fish in the same patch, were recorded continuously as sheltering (in a shelter or parallel to and within two body widths of the edge of a rock), marginal (in the marginal section of a feeding patch, see Fig. 1) or central (in the central section of a feeding patch, see Fig. 1).

Feeding, aggression and movement between patches were recorded as point activities. Aggressive interactions included unreciprocated attacks and reciprocal bouts, as described by Kalleberg (1958), and scares (no overt aggression but one fish fled or adopted a clearly subordinate posture on the approach of a second individual). The outcome of aggressive bouts was usually decisive; the loser lowered its fins into a subordinate posture, while the winner retained an aggressive posture with fins raised (see Kalleberg 1958). In the few cases where the outcome was unclear, no winner or loser was assigned. The identities of the winner and loser and their locations were recorded at the end of the interaction. Despite the use of a drum filter to remove natural prey items, the fish were sometimes observed feeding on drift and benthic prey other than the chironomid larvae supplied by the feeder system. However, chironomid larvae were by far the most common prey items, accounting for 74% of overall food intake. A feeding event was recorded whenever a fish ate or inspected but did not ingest any class of prey item. When the focal fish moved to a new patch, observations of others remaining in the old patch were discontinued.

To compare the effect of food availability on intake rates, intake rates (the number of chironomids ingested per minute observed) were calculated for each individual across all the focal observations for the run, subdivided according to the rate of food delivery and location of the fish (sheltering, marginal or central, see above). Because the presence of competitors could have influenced feeding behaviour, comparisons of intake rates at the different rates of delivery are based on data gathered when fish were not sharing a patch with an active competitor. Fish that were observed for a total of < 5 min in each location or food level were excluded from analysis.

Sequence analysis was used to identify the precursors of movements between patches. The dependence of each act (aggression, movement or feeding) on the act immediately preceding it in the behavioural sequence was examined (a first-order Markov model) by constructing a transition matrix using data from all the focal observations. X2 analysis and post-hoc pairwise comparisons were used to identify cells in the transition matrix that accounted for a significant χ2 result. G-tests were used to compare the frequency of aggression-related movements and spontaneous movements (see Results) between control and experimental treatments, between the two levels of food availability within the experimental treatment, and between patches with zero, one or more active fish. The expected numbers of moves for each of these conditions were adjusted according to the total length of observations under each condition.

dominance ranking

Because the frequency of aggression during runs was low, and some pairs of fish were rarely or never seen interacting, ranks were assigned after each run by observing the fish from all sections together in a single section of the stream divided into two patches similar in layout to those used during the experiment. Interactions among fish were recorded over the following two weeks for a total of at least 10 h. Food was delivered ad libitum during observations. A dominance matrix was constructed on the basis of the outcome of aggressive interactions. Relationships between pairs of fish were considered to be undecided if fewer than three interactions were seen; otherwise the individual that won most encounters was considered to be the dominant one of the pair. A linear dominance hierarchy including all 16 fish was constructed using the method of de Vries (1998). Overall, dominance relationships were clear and highly polarized. The four fish from each original experimental section were assigned a rank (1–4 where 1 is the most dominant) according to their relative positions within this hierarchy. There was good agreement between dominance status assigned on the basis of these ranking sessions and the outcome of aggressive encounters observed during the run; of 81 pairwise encounters observed during experimental runs, 80% were won by the fish ranked as more dominant in the post-run ranking session.

Results

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

food intake in rich and poor patches

The rate of food intake differed significantly between the rich and poor levels. Fish that occupied a central position within a patch had on average an almost 10-fold higher rate of food intake in the rich condition compared to the poor condition (median intake of fish in the central position of rich patches = 0·87 items min−1 and in poor patches = 0·10 items min−1; Mann–Whitney U = 6, n = 27, P < 0·001). Fish had lower food intake rates when occupying a marginal position (median intake in the marginal position in rich patches = 0·06 items min−1 and in poor patches = 0·00 items min−1; Mann–Whitney U = 41, n = 24, P= 0·078) and never fed while sheltering.

differential use of rich and poor patches in the experimental condition

Overall, fish tended to use the upstream and downstream patches most frequently, the number of fish making greatest use of the four patches (A–D) from upstream to downstream being A: 23, B: 7, C: 4 and D: 13, with one fish preferring patches B and C equally (χ2 = 17·58, P < 0·005). There were very few significant effects of patch quality on patch use. Overall, fish spent only marginally more time foraging at the rich as opposed to the poor level (mean time preference index = 0·05 ± 0·15; one-sample t-test with test value = 0, t = 2·998, P < 0·005), despite the large difference between the two levels of food availability.

The majority (72%) of fish (group A in Fig. 2) behaved randomly with respect to patch quality; G-tests showed no significant differences between observed and expected numbers of intervals spent in, or moves from, rich patches. The remaining fish fell into four groups on the basis of the nature and direction of their preferences. Some behaved non-randomly with respect to foraging time, spending either more time (group B in Fig. 2) or less time (group C in Fig. 2) than expected at the rich level, but showed no tendency to leave patches of either type preferentially. Others behaved non-randomly with respect to leave rates, either leaving rich patches more often (group D in Fig. 2) or less often (group E in Fig. 2) than expected, but not spending more time than expected on patches at either the rich or the poor level.

image

Figure 2. Preferences for rich and poor levels of food supply in juvenile Atlantic salmon foraging in an artificial stream, using two measures of preference. Time proportion (the proportion of available time spent foraging in patches supplying food at the rich level, expected levels 0·45–0·51) is shown on the x axis. Leave proportion (the proportion of total movements that fish made away from patches supplying food at the rich level) is shown on the y axis. The diagonal line shows the expected leave proportion for each value of the time proportion. Fish are divided into five groups (A–E) on the basis of G-tests comparing the observed numbers of intervals in, or moves from, rich and poor patches with expected numbers assuming random use of food levels (i.e. no preference). See text for details. Symbols represent four ranks of fish, from the most dominant in a group (rank 1) to the most subordinate (rank 4).

Download figure to PowerPoint

Regression analysis showed absolutely no relationship between activity and the time preference index (Fig. 3. Linear inline image = −0·03, NS; logarithmic inline image = −0·01, NS; quadratic inline image = −0·03, NS; cubic inline image = 0·06, NS), although variation in time preference index decreased as activity index increased. With one exception, activity indices for those fish whose behaviour differed from random fell well within the range for the whole sample (Fig. 3). Thus fish that showed a significant preference for patches at the rich level were not otherwise atypical. There was, however, a significant negative relationship between movement rate and leave preference index (Fig. 4; F1,27= 7·06, P = 0·013); thus fish that moved more overall tended to leave poor patches preferentially.

image

Figure 3. Tendency to spend time foraging in rich patches (time preference index) against overall activity level (activity index) in juvenile Atlantic salmon foraging in an artificial stream. Fish that spent significantly more time than expected foraging at the rich level (positive values for the time preference index) or the poor level (negative values for the time preference index) are circled.

Download figure to PowerPoint

image

Figure 4. Tendency to leave rich patches (leave preference index) plotted against movement rate in juvenile Atlantic salmon foraging in an artificial stream. Fish that left rich patches (positive values for leave preference index) or poor patches (negative values for leave preference index) significantly more often than expected are circled.

Download figure to PowerPoint

reasons for leaving patches

The transition matrix (Table 2) of the behavioural sequences recorded during focal observations indicated a strong dependence of current acts on previous acts (χ2 = 656, 4 d.f., P << 0·001), with a clear tendency for each type of action (aggression, moving or feeding) to be repeated. Performance of aggression and movement were associated mutually and positively in the sequential records and both were associated negatively with subsequent feeding. Feeding was associated with a decreased subsequent probability of both aggression and movement. There was a strong tendency for one movement to be followed by further movement, resulting in bouts that involved between two and seven patch changes, with the fish moving typically between sheltering and/or marginal areas of the stream. The second of two successive moves nearly always (93% of occasions) occurred within 3 min of the first. Consequently, all moves that occurred ≤ 3 min after a previous move were considered to be part of a movement bout. Thus, three classes of movement between patches were identified: those that were part of an ongoing bout of movement, those resulting from an aggressive interaction, and spontaneous movements.

Table 2.  First-order contingency table of Atlantic salmon behaviour sequences. Cells in the table show the number of times aggression, movement or feeding (the preceding behaviour) was followed by the same categories of behaviour (the following behaviour) during 178 focal observations of 20-min duration. The number of times each behaviour was the first to follow the start of observation is also shown but was excluded from χ2 analysis
 Following behaviour
AggressionMovementFeeding
Preceding behaviour
Aggression5430  48
Movement2341  23
Feeding61251307
Start of observation3124 142

Of the fish that left a patch after an aggressive interaction, 88% left within 30 s and 97% within 1 min 15 s. These movements were classed as aggression-related moves. All other movements between patches (i.e. those that were not immediately preceded by aggression or another move and hence had no identifiable proximate cause) were classified as spontaneous moves. Only movements at the start of bouts (i.e. aggression and spontaneous moves) were considered in further analysis. After reclassification of moves and the exclusion of 43 movements within bouts there remained 77 moves, of which 34 (44%) were aggression-related and 43 (56%) were spontaneous.

Aggression strongly promoted movement between patches. The loser of an aggressive interaction usually left the patch (26 of 34 cases), and in most cases (21 of 26) did not return within the time-scale of the focal observation. Winners were less likely to leave a patch (only eight of 34 cases) and more likely to return when they did (five of eight cases). The frequency of both aggression-related and spontaneous moves was unrelated to food delivery level (aggression-related: Gadj = 0·92, 1 d.f., NS; spontaneous: Gadj = 0·94, 1 d.f., NS). Overall, fish were most likely to leave patches containing two or more active fish, because of the high incidence of aggression-related moves (Gadj = 20·55, 1 d.f., P < 0·001).

After leaving a patch as a result of a spontaneous move, fish often moved to a sheltering position (45% of 33 cases where the location was known at the end of the subsequent movement bout). Focal animal records showed that social rank was not associated with the number of spontaneous moves (Gadj = 0·78, 3 d.f., NS) or with the overall number of moves (Gadj = 1·54, 3 d.f., NS).

the effects of social rank on patterns of space use

The frequency of moves resulting from lost aggressive interactions was greater in lower-ranking fish (Gadj = 8·67, 3 d.f., P < 0·05), while the frequency of moves occurring after a win was lower (G-test not valid as expected values < 5). Thus polarization of behaviour between fish assigned different ranks was marked in terms of the immediate outcome of aggressive interactions as reflected in the focal observations. However, few rank-related differences were found in patterns of patch exploitation as reflected in the PIT tag records. Fish of different ranks did not differ in the location of their preferred patch; for example, of the 23 fish making greatest use of the upstream patch, nine were ranked 1 (the most dominant), four were ranked 2, three were ranked 3 and seven were ranked 4 (the most subordinate). Rank 4 fish were over-represented among those that behaved non-randomly with respect to patch quality (Fig. 2) and both fish in group E (leaving patches at the rich level less often than expected) were dominant.

Fish of different ranks did not differ in time preference index (median = 0·07, 0·06, 0·06 and 0·03 for ranks 1, 2, 3 and 4, respectively; Kruskal–Wallis H = 2·01, d.f. = 3, P = 0·571). There was no significant interaction between movement rate and rank as predictors of leave preference (Fig. 4, ancova: F3,24 = 0·34, NS). Subsequent analysis of main effects showed a significant negative effect of movement rate (see above). Thus fish that moved more overall were more likely to leave poor patches preferentially. There was no significant effect of status when the full scale of ranks 1–4 was used (F3,27 = 2·42, P = 0·088). However, when fish were classified as dominant (rank = 1) or subordinate (ranks 2–4), a significant effect on leave preference of both movement rate (ancova: F1,29 = 6·82, P = 0·014) and status (F1,29 = 5·25, P = 0·029) was identified, with no interaction (F1,28 = 0·28, P = 0·604). Movement was a better predictor of leave preference for dominant fish (inline image = 72·2%, n = 8, P = 0·005) than for subordinates (inline image = 6·3%, n = 24, P = 0·125).

differences between experimental and control treatments

There were no detectable differences in patterns of patch use between the control treatment (in which fish experienced constant high quality patches) and the experimental treatment (in which they experienced randomly changing patch quality). Thus fish spent no more time foraging overall (as measured by the activity index) in the experimental treatment than in the control treatment (Mann–Whitney U = 212, n = 48, NS), despite the fact that overall food availability was higher in the latter. Fish in the experimental and control treatments showed no differences in movement rate (Mann–Whitney U = 187·5, n = 48, NS). The frequency of both aggression-related and spontaneous moves were unaffected by treatments (aggression-related moves: Gadj = 0·00, 1 d.f., NS; spontaneous moves: Gadj = 1·72, 1 d.f., NS).

Discussion

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

In this study we set out to determine whether the ability to track variable resources and the strong effect of social status on foraging behaviour that underpin Ideal Despotic models and that have been demonstrated clearly in juvenile Atlantic salmon under simple laboratory conditions are also evident in more complex conditions. Starting with the question of whether these fish make preferential use of profitable patches, our data show that on average the fish allocated just 5% more of their foraging activity to patches at the rich feed delivery level than those at the poor level. This was the case even though food was almost 10 times more abundant on rich patches, with concomitant differences in rates of food intake.

Far from making strongly preferential use of profitable patches, most individuals behaved randomly with respect to patch quality. Some used the feeding patches in a manner that differed statistically from random, but the nature of the non-randomness was not consistent across fish. Given the large and functionally significant difference in food availability between rich and poor levels and the number of replicates, it is unlikely that the experimental design was inadequate to permit discrimination of an effect. The complete lack of differences in behaviour between the control fish (which experienced consistently rich feeding patches) and the experimental fish (which experienced unpredictable 10-fold swings in patch quality) also suggests strongly that fish were not adjusting their behaviour to variable patch quality on the spatial and temporal scale used in this study. In addition, as all patches delivered food at the rich level approximately 50% of the time, and most fish spent less than 50% of the available time foraging (Fig. 3), even fish with low movement rates had the potential to bias their activity to the rich food levels. The fact that juvenile Atlantic salmon do not bias their foraging activity towards rich patches contradicts a core assumption of Ideal Despotic models (including that of Ruxton et al. 1999), namely that individuals switch among foraging sites on the basis of food availability.

This raises the question of what causes juvenile Atlantic salmon to leave patches on which they have been feeding, given that they did not leave in response to a change in food availability. Our focal animal observations showed that, in addition to a strong behavioural momentum (i.e. fish tended to continue doing what they were doing), in many cases (44%) when fish left a patch this was in response to an aggressive interaction, with the loser usually the one to leave. In the majority of cases, movement away from a patch was defined as spontaneous, in the sense that we could see no predictors of leaving.

In terms of the influence of social rank on the ability or willingness of juvenile salmon to track profitable feeding sites, in spite of clear and consistent behavioural polarization between dominant and subordinate individuals and in spite of the fact that losing an aggressive interaction can sometimes prompt a fish to move between feeding patches, we found surprisingly few effects of rank. The two fish that left rich patches less than predicted on a random model were both dominants, and one index of preferential use of rich patches (leave preference) was related significantly to movement rate in dominant but not subordinate fish. At the most, therefore, those dominant fish that have preferences for high-quality patches may be free to express these, whereas frequently moving subordinates are not. The fact that differences in social status that are clearly evident in juvenile salmon foraging in (almost) natural conditions have so little effect on use of feeding patches where these are so different in profitability is contrary to our strong expectations based on previous studies (e.g. Fausch 1984; Gotceitas & Godin 1992; Hughes 1992; Hakoyama & Iguchi 2001) and to the assumptions of many models based on the Ideal Despotic Distribution.

Several key features of the present study, including several that are comparable to natural conditions, may explain these unexpected results. In the first place, densities were lower than those used in most laboratory studies. In addition, the fish had several feeding patches to choose among rather than just two (as in the study of Hakoyama & Iguchi 2001, for example) and these patches were isolated visually (in contrast to the study of Gotceitas & Godin 1992, for example). Additionally, and perhaps most importantly, patch quality was variable rather than fixed, as in the study of Fausch (1984), who found that dominance rank of juvenile Atlantic salmon in an artificial stream closely matched patch quality when the spatial distribution of food was predictable over time. Given patch qualities that are stable over time, sampling among patches would allow animals to obtain necessary information on the distributions of resources and competitors before becoming attached to the best defensible site. Even if initial sampling were expensive (in terms of predation risk or attacks from conspecifics, for example), the pay-off would be substantial for most fish. When patch options can be compared visually, as in the experiment of Gotceitas & Godin (1992), sampling costs are likely to be low and hence tracking the quality of variable patches worthwhile. Decreased habitat stability and/or increased visual complexity reduce the marginal gains from a bout of sampling and make it necessary to sample more frequently to maintain foraging potential. In such conditions, the cost of sampling may well start to outweigh its benefits; an effect of predation risk, and hence the cost of sampling, on selectivity has been described in a number of taxa in the context of foraging (e.g. Mayer & Valone 1999; Leaver & Daly 2003) and mate choice (e.g. Forsgren 1992; Godin & Briggs 1996).

Little information is available on variations in patch qualities in natural streams, although on a time-scale of several hours in riffle habitats there is considerable variation (Martin-Smith & Armstrong 2002 and unpublished), probably arising from patchiness in drift from the benthic invertebrate community and aerial input and from the activities of upstream competitors, which exert shadow competition (Lubin et al. 2001) by filtering food from the water column. Thus it is likely that there is intense short-term variation in local food abundance, which means that efficient short-term tracking of patch qualities may not be possible.

It is possible that the fish in the present study were attempting to track processes fluctuating over longer time-scales than those used in our manipulations that partly determine patch quality, for example seasonal changes in discharge affecting local water velocities (Nislow, Folt & Parrish 1999). There is evidence from a simple laboratory experiment that some individual salmon track local velocity changes; however, in that case the majority of the population chose to remain locally site-attached (Kemp, Gilvear & Armstrong 2003). Some of the unexplained spontaneous movements observed could be associated with such long-term sampling.

Information accrued by sampling need not be food availability per se but could include other knowledge, for example, which other known individual con- and hetero-specific competitors remain alive within the home range. Knowledge of competitors may be as important as that concerning other environmental factors in enabling individual fish to assess risk and to improve their efficiency of habitat use (Koops & Abrahams 1999).

There are trade-offs inherent in the design of any scientific study. In our case, the principal trade-off was between studying fish in an environment that was as close as possible to that typically experienced in the wild, and using an environment where we could actually obtain the measurements needed to test our hypotheses. The detailed visual observation of feeding behaviour of juvenile salmonids necessary for this study could not be obtained from fish in a real stream without the imposition of very substantial lighting and video technologies, which would probably have affected fish behaviour significantly. Furthermore, control of food supply is essential to monitor responses of fish to changes in patch quality over time. Hence, we decided that the optimal solution was an artificial stream in a glass-sided channel with every effort made to provide as realistic a setting for the fish as possible, while retaining control of food provision. Another trade-off that we needed to tackle was between the realism of the system and the number of replicates. Because we gave the fish a very substantial area in which to range and interact, this necessarily reduced the sample sizes that we were able to achieve. While our sample sizes compare favourably with similar studies, they are not of the exhaustive nature needed to completely eliminate all reasonable concerns about the possibility of us failing to reject null hypotheses simply as a consequence of low statistical power. Hence, we accept that our arguments based on observations of no effects ought to be considered as suggestive rather than absolutely definitive.

This study provides an illustration of the potential problems in extrapolating from simple laboratory studies to make predictions of population processes within more complex environments. With respect to salmon in particular, and probably also many other animals, two areas of uncertainty emerge that need to be understood more thoroughly before further progress can be made. First, there is a need to understand in much more detail exactly how animals perceive mortality risk and integrate it into foraging decisions. It is well established that foraging is sensitive to predation risk (Lima & Dill 1990), and various mechanisms have been identified by which the presence of predators could potentially influence prey behaviour. For example, in various species individuals can identify chemical cues produced both by potential predators and by injured conspecifics (Kats & Dill 1996; Wisenden 2000) and learn to recognize areas in which they themselves (Csanyi 1985; Huntingford & Wright 1989) or their conspecific companions (Brown & Laland 2003) have received predatory attacks. However, there is little understanding of how these various cues are integrated to give an overall perception of risk by individual animals. This is in spite of the fact that there is an extensive literature on the mechanisms that underlie antipredator behaviour in animals (Kavaliers & Choleris 2001) and that the outcome of predictive models can be extremely sensitive to this factor (Lima 1998; Mitchell & Lima 2002). Secondly, the magnitude and frequency of fluctuations in relative local patch qualities need to be understood in more detail. To date, the definition of measurements of drift available to juvenile salmon has been c. 3–6 h, commensurate with obtaining an adequate sample of invertebrate drift and keeping the effects of disturbance during placement of collecting equipment small relative to the sample period. New techniques and approaches are needed to monitor the short-term variations in prey availability that may drive the behaviours of predators and explain discrepancies between studies in simple and more natural environments. Furthermore, such short-term variations need to be integrated with measurements at a range of time-scales to make predictions of how animals could sample the environment effectively.

This study points to some general principles in developing behavioural ecology. Small-scale laboratory studies in simplified environments are useful for identifying the behavioural capabilities that dictate what animals can potentially do in natural conditions. However, assumptions should be scrutinized and predictions tested in near-natural conditions, even when this requires the development of sophisticated ways of conducting manipulations in the field and controlled but complex laboratory tests.

Acknowledgements

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References

This work was funded by the UK's Natural Environmental Research Council (GR3/12626). Many thanks are due to M. S. Miles, S. Keay and J. Muir for their technical support and advice throughout the experiment and to two anonymous referees for detailed comments on an earlier version of this paper.

References

  1. Top of page
  2. Summary
  3. Introduction
  4. Materials and methods
  5. Results
  6. Discussion
  7. Acknowledgements
  8. References
  • Abrams, P.A. (2000) The impact of habitat selection on the spatial heterogeneity of resources in varying environments. Ecology, 81, 29022913.
  • Armstrong, J.D., Huntingford, F.A. & Herbert, N.A. (1999) Individual space use strategies of wild Atlantic salmon parr. Journal of Fish Biology, 55, 12011212.
  • Armstrong, J.D., Shackley, P.E. & Gardiner, R. (1994) Redistribution of juvenile salmonid fishes after localised catastrophic depletion. Journal of Fish Biology, 45, 10271039.
  • Bernstein, C., Kacelnik, A. & Krebs, J.R. (1988) Individual decisions and the distribution of predators in a patchy environment. Journal of Animal Ecology, 57, 10071026.
  • Brown, C. & Laland, K.N. (2003) Social learning in fishes: a review. Fish and Fisheries, 4, 280288.
  • Calsbeek, R. & Sinervo, B. (2002) An experimental test of the ideal despotic distribution. Journal of Animal Ecology, 71, 513523.
  • Clarke, C.W. & Mangel, M. (2000) Dynamic State Variable Models in Ecology. Oxford University Press, Oxford.
  • Csanyi, V. (1985) Ethological analysis of predator avoidance by the paradise fish Macropodus opercularis L. Part 1. Recognition and learning of predators. Behaviour, 92, 227240.
  • De Vries, H. (1998) Finding a dominance order most consistent with a linear hierarchy: a new procedure and review. Animal Behaviour, 55, 827843.
  • Egglishaw, H.J. & Shackley, P.E. (1977) Growth, survival and production of juvenile salmon and trout in a Scottish stream 1966–75. Journal of Fish Biology, 11, 647672.
  • Fausch, K.D. (1984) Profitable stream positions for salmonids: relating specific growth rate to net energy gain. Canadian Journal of Zoology, 62, 441451.
  • Forsgren, E. (1992) Predation risk affects mate choice in a gobiid fish. American Naturalist, 140, 10411049.
  • Fretwell, S.D. (1972) Populations in a Seasonal Environment. Princeton University Press, Princeton.
  • Gill, J.A., Sutherland, W.J. & Norris, K. (2001) Depletion models can predict shorebird distribution at different spatial scales. Proceedings of the Royal Society London, Series B, 268, 369376.
  • Giraldeau, L.A. & Caraco, T. (2000) Social Foraging Theory. Princeton University Press, Princeton.
  • Godin, J.G.J. & Briggs, S.E. (1996) Female choice under predation risk in the guppy. Animal Behaviour, 51, 117130.
  • Gotceitas, V. & Godin, J.-G.J. (1992) Effects of location of food delivery and social status on foraging-site selection by juvenile Atlantic salmon. Environmental Biology of Fishes, 35, 291300.
  • Green, C.M. & Stamps, J.A. (2001) Habitat selection at low population densitites. Ecology, 82, 20912100.
  • Hakoyama, H. & Iguchi, K. (2001) Transition from a random to an ideal free to an ideal despotic distribution: the effect of individual difference in growth. Journal of Ethology, 19, 129137.
  • Harwood, A.J., Armstrong, J.D., Metcalfe, N.B. & Griffiths, S.W. (2003) Does dominance status correlate with growth in wild stream-dwelling Atlantic salmon (Salmo salar)? Behavioral Ecology, 14, 902908.
  • Heggenes, J., Baglinière, J. & Cunjak, R. (1999) Spatial niche variability for young Atlantic salmon (Salmo salar) and brown trout (S. trutta) in heterogeneous streams. Ecology of Freshwater Fish, 8, 121.
  • Hughes, N.F. (1992) Ranking of feeding positions by drift-feeding Arctic grayling (Thymallus arcticus) in dominance hierarchies. Canadian Journal of Fisheries and Aquatic Sciences, 49, 19941998.
  • Huntingford, F.A. & Wright, P.J. (1989) How sticklebacks learn to avoid dangerous feeding places. Behavioural Processes, 19, 181189.
  • Kalleberg, H. (1958) Observations in a stream tank of territoriality and competition in juvenile salmon and trout (Salmo salar L. & S. trutta L.). Report of the Institute of Freshwater Research Drottingholm, 31, 5598.
  • Kats, L.B. & Dill, L.M. (1996) The scent of death: chemosensory assessment of predaiton risk by prey animals. Ecoscience, 5, 361394.
  • Kavaliers, M. & Choleris, E. (2001) Anti-predator responses and defensive behaviour: ecological and ethological approaches for the neurosciences. Neuroscience and Biobehavioral Reviews, 25, 577586.
  • Kelly, W.H. (1967) Marking freshwater and marine fish by injected dyes. Transactions of the American Fisheries Society, 96, 163175.
  • Kemp, P.S., Gilvear, D.J. & Armstrong, J.D. (2003) Do juvenile Atlantic salmon track local changes in water flow? Rivers Ecology and Management, 19, 569575.
  • Koops, M.A. & Abrahams, M.V. (1999) Assessing the ideal free distribution: do guppies use aggression as public information about patch quality? Ethology, 105, 737746.
  • Krause, J. & Ruxton, G.D. (2002) Living in Groups. Oxford University Press, Oxford.
  • Leaver, L.A. & Daly, M. (2003) Effects of predation risk on selectivity in heteromyid rodents. Behavioural Processes, 64, 7175.
  • Lima, S.L. (1998) Nonlethal effects in the ecology of predator–prey interactions − what are the ecological effects of anti-predator decision-making? Bioscience, 48, 2534.
  • Lima, S.L. & Dill, L.M. (1990) Behavioural decisions made under the risk of predation − a review and prospectus. Canadian Journal of Zoology, 68, 619640.
  • Lubin, Y., Henschel, J.R. & Baker, M.B. (2001) Costs of aggregation: shadow competition in a sit-and-wait predator. Oikos, 95, 5968.
  • MacLean, A., Miles, M.S. & Armstrong, J.D. (2003) A computer activated food delivery system for dispensing natural prey items among complex arrays of patches. Journal of Fish Biology, 63, 16101614.
  • Martin-Smith, K. & Armstrong, J.D. (2002) Growth rates of wild stream-dwelling Atlantic salmon correlate with activity and sex but not dominance. Journal of Animal Ecology, 71, 413423.
  • Mayer, M.D. & Valone, T.J. (1999) Foraging under multiple costs: the importance of predation, energetic and assessment error to a desert forager. Oikos, 87, 571579.
  • Metcalfe, N.B., Huntingford, F.A., Graham, W.D. & Thorpe, J.E. (1989) Early social status and the development of life-history strategies in Atlantic salmon. Proceedings of the Royal Society of London, Series B, 236, 719.
  • Metcalfe, N.B., Taylor, A.C. & Thorpe, J.E. (1995) Metabolic rate, social status and life-history strategies in Atlantic salmon. Animal Behaviour, 49, 431436.
  • Metcalfe, N.B. & Thorpe, J.E. (1990) Determinants of geographical variation in the age of seaward migrating salmon, Salmo salar. Journal of Animal Ecology, 59, 135145.
  • Mitchell, W.A. & Lima, S.L. (2002) Predator–prey shell games: large-scale movement and its implications for decision-making by prey. Oikos, 99, 249259.
  • Nakano, S. (1995) Individual differences in resource use, growth and emigration under the influence of a dominance hierarchy in fluvial red-spotted masu salmon in a natural habitat. Journal of Animal Ecology, 64, 7584.
  • Nislow, K.H., Folt, C.L. & Parrish, D.L. (1999) Favourable foraging locations for young Atlantic salmon: application to habitat and population restoration. Ecological Applications, 9, 10851099.
  • Ruxton, G.D., Armstrong, J.D. & Humphries, S. (1999) Modelling territorial behaviour of animals in variable environments. Animal Behaviour, 58, 113120.
  • Sloman, K.A. & Armstrong, J.D. (2002) Physiological effects of dominance hierarchies: laboratory artefacts or natural phenomena? Journal of Fish Biology, 61, 123.
  • Sutherland, W.J. (1996) From Individual Behaviour to Population Ecology. Oxford University Press, Oxford.
  • Tome, D. (2003) Nest site selection and predation driven despotic distribution of breeding long-eared owls. Journal of Avian Biology, 34, 150154.
  • Tregenza, T., Hack, M.A. & Thompson, D.J. (1996) Relative competitive success of unequal competitors changes with overall density. Oikos, 77, 158162.
  • Wisenden, B.D. (2000) Olfactory assessment of predation risk in the aquatic environment. Philosophical Transactions of the Royal Society, London, 355B, 12051208.
  • Zimmerman, G.S., La Haye, W.S. & Gutierrez, R.J. (2003) Empirical support for a despotic distribution in a California spotted owl population. Behavioural Ecology, 14, 433437.