Behavioral Ecology Vol. 8 No. 4: 448-455
Experiments and a model examining learning
in the area-restricted search behavior of
ferrets (Mustela putorius faro)
David G. Haskell
Section of Ecology and Systematics, Cornell University, Ithaca, NY 14855-2701, USA
L
earning has been shown to be a key component of many
aspects of animal foraging behavior (Dukas and dark,
1995; Dukas and Visscher, 1994; Kamil and dements, 1990;
Stephens and Krebs, 1986). An aspect of foraging behavior
that has not been empirically investigated from the standpoint
of learning are the "area-restricted searches" (ARSs) that foraging organisms perform after encountering a prey item or
upon entering an area of high food concentration. An ARS
involves an increase in the rate of turning of the search path
(sinuosity), resulting in a concentration of search effort in the
area around the previously encountered prey item (Bell, 1991;
Croze 1970; Tinbergen et aL, 1967). ARS behavior has been
reported in a wide variety of plants (Slade and Hutchings,
1987; Sutherland and SdUman, 1988) and animals (Bell, 1991;
Benedix, 1993; Brett, 1991; Mone and Fournier, 1994) and
has been invoked as an important determinant of populationlevel distribution patterns of predators and prey (Kareiva and
Odell, 1987). Despite its seeming significance, the study of
ARS suffers from a paucity of experimental evidence that
searches are performed in an efficient way. In particular, ARSs
have been assumed to be invariant behaviors that are not
modified in response to a foraging organism's experience of
the spatial dumping of prey (Andren, 1991; Taylor 1976).
The few theoretical papers that have addressed ARS suggest
that learning and flexibility may be an important aspect of
ARS behavior, yet this suggestion has yet to be tested empirically (Benhamou, 1992, 1994; Knoppien and Reddingius,
1985; Krakauer and Rodrfguez-Girones, 1995). Previous empirical work on ARS has focused mainly on proximate questions such as the description of foraging paths, although some
authors have hypothesized that the behavior is adaptive (Beukema, 1968; Bond, 1980; Smith, 1974a,b). I investigated the
role of experience in modifying spatial aspects of predators'
search paths. I used a spatial simulation model to make qualD. G. HaskeH is currently at (he Department of Biology, University
of the South, Sewanee. TN 37383, USA.
Received 28 August 1996; accepted 29 December 1996.
1045-2249/97/J5.00 O 1997 International Society for Behavioral Ecology
itative predictions about search behavior in a laboratory system using ferrets {.Mustela putorius furo) foraging for oil-drop
"prey items." By manipulating the spatial dumping of prey
and recording the responses of the predators, I examined two
hypotheses. First, does the path length and sinuosity of the
predator's search respond, via learning, to changes in prey
dumping? Second, does this response conform to the predictions of an opdnudity model?
Simulation
Methods
The simulation model is designed to predict how the path
length and sinuosity (rate of turning) of ARSs should change
according to variation in prey dumping. Three prey items
were placed in a triangle on a grid. I used seven different prey
dumpings that differed from each other in the distance between prey items. The most dumped pattern had two grid
spaces on the side of the triangle between the prey items. The
most spread out pattern had 14 spaces between prey items
(the other prey items were placed on the grid as near to equilateral as the discrete grid would allow). A predator foraged
across the grid. The predator started at one (chosen at random) of the three prey items and moved around the grid in
one of four search tactics. The model therefore focuses on
the search behavior of predators once a prey item has been
encountered. Search tactics differed only in the sinuosity of
the search path. Tactic I was the most sinuous: the predator
had an equal probability of moving in all eight possible directions (unbiased random walk or Brownian motion). In tactic
II die predator moved either straight forward, diagonally forward or to the right or left relative to its previous motion, and
all five possible directions had an equal chance of being chosen. In tactic HI the predator moved only straight forward or
diagonally forward relative to its previous motion, and alf
three possible directions had an equal chance of being chosen. Tactic IV was the same as tactic HI except that the predator had a two-thirds probability of moving straight forward,
the remaining two directions each had a one-sixth probability
Downloaded from http://beheco.oxfordjournals.org/ at Pennsylvania State University on April 27, 2014
Area-restricted searches have been described u important components of the foraging behavior of many organisms. It is unclear,
however, whether individual foragers can use learning to fine-tune their searches, or even whether these searches are efficiently
performed. I used a simulation model to make qualitative predictions about search behavior in a laboratory system. The simulation model indicates that the sinuosity and path length of searches strongly affect search efficiency. The model predicts that,
for a rate-maximizing forager, path length should increase and search sinuosity should decrease as prey become less dumped.
Foraging animal* may therefore be selected to learn the path length and sinuosity of searches in response to changing degrees
of dumping of prey. These predictions were tested in a laboratory system involving ferrets (Mustela putorius furo) foraging for
oil-drop "prey items." Search paths changed in a graded manner to experimental manipulations of the dumping of prey. As
predicted by the model, ferrets learned to perform longer and less sinuout search paths as prey became less dumped. This
study provides the first evidence that area-restricted search behavior is learned and can be fine-tuned to efficiently exploit
different spatial distributions of food. Key words: area-restricted search, behavior, learning, Mustela putorius furo, opnmality,
spatial model. [Behav Ecol 8:448-455 (1997)]
Htskell * learning and torch behavior
449
- I (high sinuosity)
80
60-II
40-
-m
20"
of being chosen. Sinuosity therefore decrease* from tactic I
to IV. The grid was large enough that the predator never encountered the grid edge.
As the predator moved across the grid, I recorded whether
a prey item was encountered on each move. This process was
repeated 2 X 10* times for each combination of search tactic
and prey dumping (there are 28 combinations: 4 sinuosities,
7 prey dumpings). For each search tactic/prey dumping combination, I then summed across all 2 X 10* records to produce
an estimate of the probability distribution of prey encounters
on each move that the predator makes (2 X 10* repetitions
for each combination produced smooth plots). For example,
if the predator started in a closely dumped group of prey and
moved in a straight line, the distribution would show a high
- peak for the first few moves as the predator moved through
a food-rich area and had a high probability of finding food.
The plot would then drop off sharply to zero as the predator
left the dump and moved into a foodless region.
For each prey dumping/search tactic combination, I calculated the length of the search path (number of steps) that
yielded the highest expected rate of return (number of prey
per step). The calculation was performed by converting the
expected prey intake plots into cumulative plots (prey intake
versus path length) and finding the path length that produced the largest prey intake/path length value. Thus, the
measure of search efficiency in this model is rate maximization within the patch (rate in this model is based on path
length, not time; the model is analogous to rate-maximization,
with travel time equal to zero, in the time-based rate-maximization patch models; see Stephens and Krebs, 1986). I used
this simple measure because the goal of the simulation was to
produce qualitative predictions for the ferret-oil drop system.
The predictions are fairly robust because repeated simulations
with other prey patterns (squares, hexagons) have produced
similar predictions, and including a between-patch travel time
when computing the rate-maximizing number of steps does
not change the qualitative predictions of the model.
Rtsulis and mttrpretatim
The best combination of search tactic and search path length
varies with prey dumping. For a rate-maximizing predator
that optimizes both search sinuosity and search path length,
the best path length increase* with distance between prey
(Figure 1). The best search sinuosity decreases as distance between the prey increases (Figure 1). The two extreme search
sinuosities were never optimal in this modeL The most sinu-
15
tue
rttT^^^^^y*^ft^^^^ff
DFCQAtor
ihould increase search path
length and decrease search sinuosity. The slight decrease in
optimal path length between
the third and fourth points is
due to the switch in optimal
search sinuosity that occurs at
this point
ous (I: unbiased random walk or Brownian motion) researched its own path too frequently, and the lean sinuous
(IV) left prey chimps too fast. Thus the simulation predicts
that when faced irith relatively spread out prey, a predator
should use longer searches and engage in less turning than
when prey are dumped. This result makes intuitive sense and
confirms the suggestions made in Chandler's (1969), Smith's
(1974b) and Bond's (1980) verbal models. The only other
simulation analysis of optimal sinuosity and search path length
in a grid of prey items (Benhamou, 1992) found that, although switching ARSs and less sinuous searches did increase
search efficiency, the optimal level of ARS path sinuosity and
length hardly varied across all the prey groupings used in the
simulation. However, Benhamou modeled highly dumped
prey. The difference between the results of this study and
those of Benhamou, therefore, suggest that the benefits of
learning may be dependent on the degree and scale of prey
dumping.
In addition to calculating optima, the simulation can be
used to examine the efficiency of search tactics away from the
optimum by generating fitness surface*. Examination of fitness surfaces around optima can inform us about how natural
selection would affect the evolution of behavior (Mangel,
1991; Mangel and Ludwig, 1992; McNamara and Houston,
1986). In my model, for example, if the fitness surface was
very sharp, predators would have to forage with precisely the
right path length and sinuosity to reap the high rate of return
at the optimum. If the surface was domed, or nearly flat, a
predator would only need to search at roughly the right sinuosity and path length to gain the high rate of return near
the optimum.
Figure 2 shows how the rate of return for a predator foraging in one prey grouping varies with both search path
length and sinuosity. The optimal search tactic occurs at a
path length of 24 and a sinuosity value of II. The fitness surface drops more sharply for shorter than optimal searches
than for longer than optimal searches. Thus in my simulation,
predators are heavily penalized for underestimating optimal
path lengths but, depending on the quality of other food
patches, they would pay a lower cost for overestimation. Variation in path length produces similarly shaped fitness surfaces
in the other six prey arrangements.
The fitness surface for sinuosity is harder to interpret First,
continuous measures of sinuosity cannot be derived from
movement on a discrete grid. Thus, although my four levels
of sinuosity span a wide range of search methods, they do not
Downloaded from http://beheco.oxfordjournals.org/ at Pennsylvania State University on April 27, 2014
IV (low sinuosity)
5
10
Distance between prey
Results of theriin11^*^*^mod*
cL As prey get farther apart,
450
Behavioral Ecology VoL 8 No. 4
OJM
OJB
OJE
1*1*"
• Tactic IV Octst sinuous)
10
20
30
40
50
60
Length of search path
Figure X
Rate of return plotted against all combinations of path length and
search «inuo«ty for the third prey distribution (prey 6 grid units
apart). See text for details.
represent four evenly spaced points on a continuum. Despite
these caveats, there is a similar shape of fitness surface for all
seven prey distributions. The rate of return falls most sharply
for lower than optimal sinuosity levels. For the prey arrangements modeled here, natural variation in sinuosity should
therefore err, if at all, in the direction of ovemnuosity.
The observation that fitness surfaces for search paths are
rounded, rather than sharp, accords with Benhamnou's similar (1992) findings and suggests that even slightly inaccurate
learning of ARSs will increase the searching efficiency of a
foraging animaL This might feriHta^ the evolution of ARS
learning mechanisms, given that simple and imprecise learning rules may yield similar benefits to more complex and accurate rules. In addition, for all the prey dumping examined,
selection might favor tactics that skew any natural variation in
search path length upward. This would avoid the higher costs
of performing too short a search. Alternatively, if natural variation in the path length of search tactics cannot be skewed,
natural selection might favor path lengths slightly above the
optima found by this model. In this case most path lengths
would be slightly longer than optimal, but the animal would
rarely perform searches that were shorter than the optima,
thus avoiding the high costs of undersearching described
above [see Mountford's (1968) discussion of brood-size evolution for a more detailed explanation of an analogous effect].
Overall, my simulation results suggest that the benefits of
ARSs vary according to the form that the searches take. Thus,
these searches can be investigated from an optunality standpoint analogous to that used in other foraging models (Stephens and Krebs 1986; see also Benhamou, 1992; Krakauer
and Rodrfguez-Girones, 1995). More specifically, the model's
results suggest that an ability to match the path length and
sinuosity of search paths to the spatial dumping of prey would
enable foraging animals to increase their search efficiency.
Methods
All experiments involved four adult ferrets (Musteia putorius
JUTO) that foraged for prey items in an linoleum-covered arena
(325 cm X 312 cm). The arena was marked into a grid (10-cm
spacing) to facilitate the measurement of foraging paths. The
Downloaded from http://beheco.oxfordjournals.org/ at Pennsylvania State University on April 27, 2014
0
• Tactic I (most sinuous)
• Tactic n
ferrets were captive-raised and had all previously been handfed individual drops of the oil "prey" used in the experiments, but they had not experienced oil drops in the arena.
The arena had 75-cm wide exits on two sides, which the ferrets
could pass through at any time in the trial. The ferrets had
day-to-day access to the arena from their living quarters (9 X
7 m), but when trials were run, only one ferret at a time was
allowed in the arena. Prey items used in these experiments
were drops of vitamin supplement oil ("Iinatone" from Lambert Kay, New York, USA) placed in the arena. This "prey"
was chosen because ferrets have a strong affinity for this food
(Morton and Morton,. 1985), search actively for more drops
once one drop has been discovered, and are unable to detect
the oil drops from a distance. Each drop consisted of approximately 0.1 ml of oiL The pipette used to produce oil drops
was not graduated, therefore drop volume may have varied
somewhat, although visual inspection of drops suggested that
this variation was minimal, When not engaged in a trial, the
ferrets had ad libitum access to high protein dry pellet food
(kitten food from lams Company, Dayton, Ohio, USA) and
water in their living quarters.
Experiment 1. The first experiment tested whether ferrets
would change the path length of their ARS paths when the
spatial clumping of their prey was changed. Every 24 h each
ferret was presented with oil droplets placed in an equilateral
triangle on the arena floor. For 20 days the ferrets were presented with drops 10 cm apart. Each ferrets' foraging trial
began by placing the ferret at one of the three drops, then
observing and recording the path length of the subsequent
search. The trial ended when the ferret lifted its nose from
the arena floor and left the arena. Occasionally the ferrets
would stop, lift their noses, then resume searching. On these
occasions I recorded the entire search path (before and after
the pause). The arena was washed with soap and water after
each ferret's foraging trial. Triangles of drops were placed at
randomized angles with respect to die edge of the arena and
in different parts of the arena for each trial. After 20 days the
drops were placed 100 cm away from each other. Foraging
trials and distance measurements were repeated as described
above.
Experiment 2. The second experiment tested the simulation
model's predictions about search behavior in response to different prey clumping. Ferrets foraged individually for equilateral triangles of oil drops as described above, except the trials
were conducted every 12 h. The experiment involved six treatments among which the distance between prey items differed
(distances between prey for treatments 1-5: 11 cm, 22 cm, 33
cm, 44 cm, 55 cm, respectively). The sixth treatment returned
to the most clumped situation of the first treatment. This sixth
treatment tested whether search path lengths could be reduced again after increasing across five treatments. Within a
treatment each ferret initially experienced four "familiarization" trials with that treatment's prey clumping. For the next
six trials the prey dumping remained the same, and data on
the ferret's foraging path was recorded. After each treatment
either a "control trial" (see below) was conducted or the next
treatment began. Data collection for this experiment, and all
controls, was conducted by an independent observer who was
ignorant of the experiment's purpose and the model's predictions. The observer watched the ferrets as they foraged and
used the grid-marks on the arena to record the search path
on gridded paper.
Controls. The interpretation of the ferrets' search behavior
depends on whether the ferrets can use cues other than memory to match their search patterns to the spatial dumping of
prey. I therefore conducted four controls. The first control
determined the distance from which ferrets can detect the
location of oil drops. Two drops of oil were placed 40 cm
Haskell • Learning and search behavior
451
•S
8.
20
30
40
50
5000
10
20
30
40
10
20
30
40
5000
is8
10
20
30
40
50
Days since start of experiment
Figure S
Distance searched by four ferrets against days since start of experiment Filled arrows indicate the date that oil drops were moved farther
apart Open arrows indicate, for each ferret, the first encounter with the new prey clumping.
apart. A ferret was placed at one drop and observed while
foraging for the second drop. In the majority of cases the
second drop was found by a ferret moving around the arena
and eventually walking to the tide of the second drop, detecting the location of the drop, turning, and eating. Detection is characterized by a sudden halt of walking, increased
inhalation through the nostrils and orientation toward the
prey item. I recorded the distance between the ferret and the
drop when this detection occurred. Fifteen trial* were conducted for each ferret
The second control was conducted after treatments 3-6 of
experiment 2 and involved washing the whole arena with 70%
ethanol, then rewashing with soap and water. Ethanol was
used to supplement the soap and water to make doubly sure
that no odor cue* were present Each ferret was then presented with Just one drop of oil and the subsequent foraging
path was recorded. If ferrets are using odor cues to locate prey
from a distance or if the ferrets are counting prey items, we
would expect the search patterns performed in this control
to be independent of the ferrets' experience with prey clumping. If memory is influencing foraging path, then the treatment preceding the control should affect the form of the
search.
The third control took place after treatment 5 (oO drops
spread out) of experiment 2. The arena was washed with alcohol arid soap after each trial as described above. Over the
next six control trials, the ferrets were alternately presented
with two types of prey dumping. One was identical to treatment 5 (drops spread out), the other was identical to treatment 6 (drops close together). If ferrets are using odor cues
to detect oil drops from a distance, we would expect the path
length and sinuosity of the search to alternate with the prey
clumping. If memory influences behavior, we would expect
only searches appropriate to treatment 5.
The last control was conducted after treatment 6 of experiment 2 and was identical to the third control except that, if
memory influences foraging, we would predict that the ferret*
should conduct searches appropriate to treatment 6.
The records of each ferret's foraging path from all the treatments and controls were analyzed in the same way. I used a
rolling cartographer's measurer to trace, and hence measure,
the length of each ferret's path. On five randomly chosen
records I repeated the path length measurement four n""*
The distance measured in these repeats never differed by
more than 2-5% from die original measurement To obtain a
measure of the sinuosity of foraging paths I followed the foraging path of each ferret across the gridded paper. Each time
the path crossed a gridline, I compared the direction of the
path at mat point to the direction of the path at the last gridline crossing. Each subdivision of the grid has four sides,
hence each gridline crossing was scored as either a left turn,
a right turn, a continuation, or a reversal of the foraging path.
I used the proportion of moves that are continuations as my
metric of sinuosity—a low proportion indicates a sinuous
path. This is a conservative measure because diagonal straight
paths are scored as a mixture of left and right turns and continuations.
RMSUUS and
interpretation
Experiment 1. All four ferrets learned to perform short ARS
paths for the very clumped prey (Figure 3). The ferrets sig-
Downloaded from http://beheco.oxfordjournals.org/ at Pennsylvania State University on April 27, 2014
1
10
452
Behavioral Ecology VoL 8 No. 4
2500
2500
A
12501
1
22
33
D
1250
1
44
11
55
1
22
•
•
•
•
33
I
44
55
2500
1250
0
1 1 2 2 3 3 4 4 5 5
0
1 1 2 2 3 3 4 4 5 5
Distance between prey items (cm)
Figure 4
Search path length plotted against distance between oQ drops for ferrets A-D. Open points indicate search path lengths for the control in
which one drop of oil was presented on an alcohol- and soap-washed arena.
nificandy increased the path length of their ARS when the
interprey item distance increased [Figure 3; Mann-Whitney U
for ferrets A-D, respectively: U » 17.5, 61.5, 60, 19; p (corrected for ties) - .0001, .9001, .0001, .0001; test groups: all
distances from start of experiment until ferret discovered new
dumping arrangement versus all distances after ferret discovered new dumping arrangement]. Individual ferrets differed
in how soon they altered their behavior after the prey dumping was changed. This was due to variation between the ferrets
in how soon they discovered the new prey clumping arrangement (Figure 3). To discover the new prey clumping arrangement, a ferret had to encounter a second oil drop accidentally
as it moved out of the arena following consumption of the
first drop.
Experiment 2. All four ferrets showed a significant increase
in their search path lengths across the first five treatments
(Figure 4; linear regression of log-transformed path length
against distance between oil drops for ferrets A-D, respectively, Fl:u = 145.4, 121.3, 83.2, 162.9; p = .0001, .0001, .0001,
.0001). The ferrets also showed an increase in the proportion
of moves that were "continuations" across the first five treatments (Figure 5; regression of square-root arcsine-transformed propwra»m against distance between oil drops for ferrets A-D, respectively, FIXI - 59.45, 47.45, 33.09, 48.14 ; p .0001, .0001, .0001, .0001). Thus, sinuosity decreased as prey
became more spread out These changes in sinuosity were
observed despite the conservative nature of the measure of
sinuosity used (diagonal straight paths score as a mix of
straight moves and left and right turns). In the sixth treatment, when the drops were dose together again, the ferrets
reverted to short, sinuous search paths. These search paths
were slightly longer than the searches they performed in treatment 1 but statistically indistinguishable from the search path
lengths in treatment 2 [ANOVA on log-transformed distances
searched in first, second and sixth treatments: FtK " 32.65,
p - .0001; Scheffe F contrasts: F(first versus second treatments) - 29.7, p < .05; F(futt versus sixth) - 17.7, p < .05;
J^second versus sixth) = 1.53, p > .05]. Power calculations
were conducted for these, and subsequent nonsignificant ANOVAs; with die sample size used in this experiment, die probability of the ANOVA detecting die average difference between treatment means in all tests was greater than 0.8 (ANOVA pooled across ferrets because there were no significant
differences among ferrets). The sinuosity of the searches in
treatment 6 reverted to that of treatment 1 (ANOVA on
square-root arcsine transformed proportions on moves that
were straight ahead in first, second, and sixth treatments: Fw
• 14.95, p — .0001; Scheffe contrasts: i^first versus second
treatments) = 7.36, p < .05; F(&nt versus sixdi) = 1.06, p >
0.05; /{second versus sixdi) - 14.00, p < .05; ANOVA pooled
across ferrets because their were no significant differences
among ferrets). Thus, ferrets are able to both increase and
decrease path sinuosity and length.
The experimental arena was large enough that, in die majority of cases, edges did not interfere with the ferrets' searches. When ferrets did encounter an edge, they moved away and
Downloaded from http://beheco.oxfordjournals.org/ at Pennsylvania State University on April 27, 2014
11
i
i
•
s
i
•
B
HiikeD • Learning and search behavior
453
0.6
B
i
0.3
•
n
11
22
0
11
22
33
44
55
11
•
1
•
•
•
22
33
44
I
55
55
Distance between prey items (cm)
Figure 5
Proportion of gridline crossings that continued in the same direction as the previous craning plotted against distance between oil drops for
ferrets A-D. Sinuosity decreases as the proportion increases.
continued their search. If edge-effects bad been artificially
limiting the path length or increasing the sinuosity of searches, one would expect these effects to reduce or eliminate differences between treatments. Thus, although the experimental system could not fully mirror the ideal "edgeless" condition of the model, it is unlikely that the results of the experiment were biased by any edge effects.
I suggest that the between-treatment differences in behavior
were caused by the ferrets learning to modify the path length
and sinuosity of their searches based on their experience of
prey clumping. There are four alternative hypotheses for the
observed patterns, which my controls were designed to address. The first alternative (control 1) is that the ferrets could
detect all three prey items from the start of each trial and
traveled from one to the other. However, the distances from
which ferrets detected the location of individual oil drops
were 13 cm ± 0.2, 1.4 cm ± 0.2, 1.4 cm ± 0.2, and 1.4 cm
± 0.2 (mean±SE for each ferret). Thus, even after consuming
an oil drop in the most clumped treatment, ferrets should not
have been able to detect the location of the remaining oil
drops. In addition, the ferrets' foraging paths did not lead
from one prey item to another, but frequently moved around
the arena before encountering the next prey item (Figure 6).
The second alternative explanation (control 2) for the results is that the ferrets could count the number of prey encountered and adopted a "forage until three prey are encountered" tactic Three observations suggest that this was
not the case. First, ferrets continued searching for a significant proportion of time after encountering the third prey
item (proportion of total search path length that occurred
after the third prey item was encountered for four ferrets "
38%; there was no trend for this proportion to decrease
through the experiment; see also Figure 6). Second, if ferrets
were counting, when ferrets found fewer than three prey
items, the path length should be unusually long because the
ferrets were searching for the unfound drops. In fact, no relationship exists between number of prey found and path
length [ANOVA comparing path length when two prey were
found (n « 16) to when three prey were found (n •• 102),
I.iie *" 1-12, p " .293; cases where one prey was found were
excluded because they were so rare (n " 2); ANOVA pooled
across ferrets because their were no significant differences
among ferrets]. Third, when presented with one oil drop
(control 2), the ferrets searched fora distance appropriate to
the treatment that they had just experienced rather than for
a fixed distance (Figure 4; all but two search distances in this
treatment fell within the 95% confidence intervals of path
lengths for each ferret in the preceding treatment; for the
two exceptions, one path length fell above the confidence interval and the other fell below).
The third alternative (controls 3 and 4) is that although the
ferrets cannot detect individual prey items from a distance,
they may be responding to the potentially different concentrations of "oil odor" in each treatment The results of con-
Downloaded from http://beheco.oxfordjournals.org/ at Pennsylvania State University on April 27, 2014
0
i
•
1
i
454
Behavioral Ecology VoL 8 No. 4
10 OB
trols 3 and 4 eliminate this alternative. In control 3 the ferrets
performed long, straight searches despite being presented
with highly dumped oil drops (all controls but one fell within
the 95% confidence intervals for each ferret in treatment 5;
the exception was longer than the 95% upper limit). In several instances a ferret consumed one drop in the cluster, then
performed a long search before rediscovering the cluster and
consuming a second drop. In control 4 the ferrets performed
short, sinuous searches despite the spread out nature of the
oil drops (9 out of 12 controls fell within the 95% confidence
intervals for each ferret in treatment 6; the T^mainfng controls well within the corresponding intervals for treatment 2).
The last alternative explanation is that the ferrets employed
the tactic "search for X distance units, or until another drop
is found, after consuming an oil drop" in all treaaneats. This
alternative concerns the distance searched, but not sinuosity.
The results of control 2 eliminate this alternative. In addition,
this alternative hypothesis would predict a positive relationship between the number of prey encountered and the path
length, whereas no such relationship exists. In all controls,
therefore, I failed to find evidence that behavior of the ferrets
was determined by any factors other than the information
learned by the ferret in previous foraging trials.
DISCUSSION
Ferrets change their ARS behavior in response to different
experiences of the spatial dumping of prey. The changes in
foraging behavior affect both the path length and die rate of
turning (sinuosity) of the search path. These two aspects of
searching behavior responded in a graded manner to changes
in prey distribution, demonstrating that the learning ability
of the ferrets allows fine-tuning of their ARS behavior. There
is a good fit between the predictions of the rate-maxunizatlon
model (Figure 1) and the results of the experiment (Figures
4, 5), suggesting that predators' behavior has been modified
by natural selection to allow efficient exploitation of dumped
distributions of prey. Two additional observation! are required
to corroborate this adaptive hypothesis. First, the learning
L-A. Ciraldeau, R. E. Johnston, M. Mangel, J. P. McCarty, H. K. Reeve,
P. W. Sherman, S. A. Vance, D. W. Winkler, and several anonymous
itvltwm provided helpful suggestions for experimental design and
comment! on the manuscript. S. A. Vance and R. Varmits helped with
the experiments.
REFERENCES
Andren H, 1991. Predation: an overrated factor for over-diipenion of
birds' nests. Anim Behav 41:1063-1069.
Bell WJ, 1991. Searching behaviour the behavioural ecology of finding resources. New York: Oiapman and Hall.
Benedix JH, 199S. Area-restricted search by the plains pocket gopher
(Gtontjs bunamu) in taDgrass prairie habitat. Behav Ecol 4318324.
Benhamou S, 1992. Efficiency of area-concentrated searching behavior in a continuous patchy environment. J Theor Biol 159*67-81.
Benhamou S, 1994. Spatial memory and searching efficiency. Anim
Behav 47:1423-1433.
Beukema JJ, 1968. Predation by the three-spined stickleback (GasttrattUM acuitdtus L): The Influence of hunger and experience. Behaviour 31:1-126.
Bond AB, 1980. Optimal foraging in a uniform habitat: die search
mechanism of the green bcewing. Anim Behav 28:10-19.
Brett RA, 1991. The ecology of naked mole-rat colonies: burrowing,
food, and limiting factors. In: Biology of the naked mole-rat (Sher-
Downloaded from http://beheco.oxfordjournals.org/ at Pennsylvania State University on April 27, 2014
Example of ferret foraging pathi (a) when oQ drop* were 11 cm
apart and (b) when oil drops were 55 a n apart
ability of animals needs to be related to die variability of the
distribution of their prey in die wild. Second, die costs and
benefits measured in studies of ARS have to be shown to affect
die fitnesses of wild individuals. Stephens and Krebs (1936)
discuss diese issues as they relate to patch foraging models.
The experiments described in diis paper represent a conservative test of die potential flexibility of ARSs. The ferrets
used were raised outside of dieir natural environment and
were well-fed throughout die trials. In addition, die experiments were conducted on a small scale relative to die movement of M. putorius in the wild (Weber, 1989a). All these
factors might lead one to anticipate a lack of responsiveness
to prey distribution and an absence of fine-tuning of ARSs.
The costs of "suboptimal" searching behavior would be higher for wild animals foraging over much larger areas without
recourse to a regular supply of food. I would therefore predict
timflar learning abilities in more natural situations, although
other variables, such as mate searching, may complicate the
interpretation of natural movements. ARSs have been reported from radiotracked MusUia putorius in die wild (Lode,
1991; Weber, 1989b), and die role of learning awaits investigation.
Looking beyond die predator's perspective, my results also
have implications for how prey should arrange diemselves to
minimize predanon risk. 'Verbal (Tinbergen et aL, 1967) and
mathematical (Taylor, 1976) models have been used to argue
that prey should arrange themselves regularly in space (overdispersed) so as to avoid being encountered during die
ARSs of predators. These models, however, viewed ARS as a
fixed behavior that predators would engage in whatever die
prey item encountered, and whatever die past experience of
die foraging predator. My results suggest that this assumption
is unwarranted. If predators can modify their searching tactics
based on their experience of how prey is arranged, while die
prey can also modify their arrangement to mjnimiw predadon risk, die optimal degree of prey clumping is not intuitively dear. The situation is a spatial game between predators and
prey. The Nash equilibrium in diis game (die "stalemate"
where bodi players in die game have maximized their payoff
to die extent allowed by die other player) will be determined
by factors such as die costs incurred by foraging predators, as
well as die constraints, such as overall density, acting on die
prey.
Hasten • learning and search behavior
McNamara JM, Houston AI, 1986. The common currency for behavioural decisions. Am Nat 127:558-378.
Mone H, Foumier A, 1994. Effects of snail-conditioned water on survival and host-searching behavior of schistosome mir»rt«n» Anim
Behav 48:1-8.
Morton C, Morton F, 1985. Ferrets: purchase, care, nutrition, diseases,
behavior, and breeding. New \fork: Barren's JMnr»ti»Ti«i Series.
Mountford MD, 1968. The significance of litteraize. J Anim Ecol 37:
363-367.
Slade AJ, Hutchings, MJ 1987. The effects of nutrient availability on
foraging in the ckmal herb Gitduma ktdtmaa.} Ecol 75.-95-112.
Smith JNM, 1974a. The food searching behavior of two European
thrushes. I: Description and analysis of search paths. Behaviour 48:
276-302.
Smith JNM, 1974b. The food searching behavior of two European
thrushes. It The adaptiveness of the search patterns. Behaviour 49:
1-61.
Stephens DW, Krebs JR. 1986. Foraging theory. Princeton, NewJersey:
Princeton University Press.
Sutherland WJ, SuHman RA, 1988. The foraging tactics of plants. Oikos 52:239-244.
Taylor J, 1976. The advantage of spacing-out. J Theor Biol 59:485490.
Tinbergen N, Impekoven M, Franck D, 1967. An experiment on spacing-out as a defense against predation. Behaviour 28:307-321.
Weber D, 1989a. Foraging in polecats, Mustda putorius U, of Switzerland: the case of a specialist anuran predator. Zeit Saugeberk 54:
377-392.
Weber D, 1989b. Observations of activity, movements and ranges of
polecats, Mustafa putorius I_ Rev Suisse Zool 96:841-862.
Downloaded from http://beheco.oxfordjournals.org/ at Pennsylvania State University on April 27, 2014
man PW, Jarvis JUM, Alexander RD, eds). Princeton, New Jersey:
Princeton University Press, 137-184.
Chandler AEF, 1969. Locomotory behavior of first instar larvae of
aphidophagous Syrphidae (Diptera) after contact with aphids.
Anim Behav 17:673-678.
Croze H, 1970. Searching image formation in carrion crows. Berlin:
VerlagPaul Party.
Dukas R, Clark CW, 1995. Searching for cryptic prey: a dynamic mode l Ecology 7&1320-1326.
Dukas R, Visscher PK, 1994. lifrtim* learning for foraging honey
bees. Anim Behar 48:1007-1012.
Kami! AC, Clements KC 1990. Learning, memory and foraging behaviorJn: Contemporary issues in comparative psychology (Dewsbury DA, ed). Sunderiand, Massachusetts: sjp'urr ^««~-4«f«- 7—v>
Kareiva P, OdeD G, 1987. Swarms of predators exhibit prey taxis if
individual predators use area-restricted search. Am Nat 130:233270.
Knoppien P, ReddmgiusJ, 1985. Predators with two modes of searching: a mathematical model. J Theor Biol 114373-301.
Krakauer DC Rodrfguez-Girones MA, 1995. Searching and learning
in a random environment. J Theor Biol 177:417-419
Lode T, 1991. Environment, exploitation and space organisation of
two European musteKdr stone marten and polecat. Vie Milieu 41:
29-38.
Mangel M, 1991. Adaptive walks on behavioral landscapes and the
evolution of optimal behavior by natural selection. ETOI Ecol 5:3039.
Mangel M, Ludwtg D, 1992 Definition and evaluation of the fitness
of behavioral and developmental programs. Annu Rev Ecol Syst 23:
507-36.
455