Content uploaded by Shane A Richards
Author content
All content in this area was uploaded by Shane A Richards on Oct 18, 2015
Content may be subject to copyright.
Linking Angling Catch Rates and Fish Learning under
Catch-and-Release Regulations
PAUL J. ASKEY,* SHANE A. RICHARDS,
1
AND JOHN R. POST
Department of Biological Sciences, University of Calgary,
2500 University Drive Northwest, Calgary, Alberta T2N 1N4, Canada
ERIC A. PARKINSON
British Columbia Ministry of Environment, University of British Columbia, Vancouver,
British Columbia V6T 1Z4, Canada
Abstract.—Many recreational fisheries are subject to varying degrees of catch-and-release fishing through
regulations and conservation-minded anglers. Clearly, releasing a proportion of the catch improves
conservation of the fishery, yet it is not clear how the released catch contributes to angling quality. If fish
change their behavior to lower their individual catchability after they have been caught, then angler catch rates
may not be proportional to fish density. Therefore, even catch-and-release fisheries could exhibit poor angling
quality if there is sufficiently high angler effort. We tested this idea by experimentally fishing five small lakes
that contained rainbow trout Oncorhynchus mykiss in the interior of British Columbia. We found that with
sustained effort of 8 angler-hours d
1
ha
1
and complete release of the catch, catch rates quickly dropped
within 7–10 d. Given the individual capture histories of tagged fish, the most parsimonious catchability model
incorporated learning and heterogeneity into intrinsic catchability. The best-fit parameter values suggest that
the population contained a group of highly catchable fish that were quickly caught and then learned to avoid
hooks. There was a seasonal decrease in catchability that was independent of angling; however, it was not
sufficient to explain the data. Our results indicate that catch rates may decline because of high angling effort
even when the number of fish remains constant. Therefore, management goals that go beyond conservation
issues and attempt to maximize angler satisfaction must account for effort density on a recreational fishery.
Stringent regulations and conservation-minded an-
glers have made catch-and-release fishing increasingly
common in North America (Barnhart 1989; Cooke and
Suski 2005). Catch-and-release fisheries are positive
for conservation-oriented management goals, because
intentional, legal harvest mortality is eliminated.
However, managers often must balance conservation
issues with angler satisfaction and provide quality
angling opportunities. The extent to which angler catch
rates are improved by regulations that impose partial or
complete catch and release is unclear. If catch rates are
directly related to fish density, then angler catch should
increase in proportion to the number of fish saved from
harvest. However, catch-and-release fisheries differ
from harvest fisheries in that the fished population
consists of fish that have never been caught as well as
fish that have been caught and released. Thus, whether
catch per unit effort (CPUE) is proportional to density
depends on the intrinsic assumption that the catch-
ability of fish that have been caught before and fish that
have never been caught fish is equal.
Biologists have frequently observed seasonal de-
clines in CPUE that supersede the decline in sport fish
abundance due to harvest (Aldrich 1939; Beukema
1970; Hackney and Linkous 1978; van Poorten and
Post 2005). Catch per unit effort is the product of fish
density (number per area) and the capture efficiency of
anglers (area swept per angler time); therefore,
excessive decrease in CPUE indicates a s easonal
decrease in capture efficiency. It has been hypothesized
that this pattern may arise because previously caught
fish learn to avoid hooks. Decreased catchability of
previously captured fish has been tested for several
different sport fishes (Beukema 1970; Hackney and
Linkous 1978; Tsuboi and Morita 2004; Young and
Hayes 2004). The conditioning hypothesis has been
supported in most of these experiments. However,
some of the tests produced no evidence of learning, and
the effect appears to vary depending on species and
experimental conditions.
Several other processes have been postulated to
cause seasonal decreases in catchability. Martin (1958)
hypothesized that a rapid decrease in CPUE was caused
by differential vulnerability to capture among individ-
ual fish. The more vulnerable fish are r apidly
* Corresponding author: pjaskey@ucalgary.ca
1
Present address: Biological and Biomedical Science,
University of Durham, South Road, Durham DH1 3LE, UK.
Received January 31, 2006; accepted May 3, 2006
Published online November 30, 2006
1020
North American Journal of Fisheries Management 26:1020–1029, 2006
Ó Copyright by the American Fisheries Society 2006
DOI: 10.1577/M06-035.1
[Article]
harvested, which leaves a less vulnerable pool of fish
and a corresponding decrease in CPUE. Cox and
Walters (2002) presented a theoretical framework in
which fish populations are composed of two pools of
individuals that are defined as vulnerable or invulner-
able to angling. In their example, fish may move
between these defined states by a behavioral change in
reactivity to lures (independent of learne d ho ok
avoidance) or by physically moving between shallow,
fishable shoals and deepwater, unfishable habitats.
They showed theoretically that when an invulnerable
pool of fish is present, increasing effort can lead to
decreased catch rates, despite a near-constant fish
density. Finally, catchability may decrease because of
seasonal environmental changes that are independent
of angler dynamics (van Poorten and Post 2005).
Seasonal changes in temperature and resource avail-
ability may affect the feeding behavior of fish and thus
their susceptibility to anglers.
Seasonal decreases in catchability are common in
recreational fisheries; however, the mechanisms un-
derlying this pattern re main unclear. Evidence is
accumulating to suggest that learned hook avoidance
is a common behavioral response among sport fishes.
However, the most convincing evidence comes from
the laboratory or small experimental ponds. How this
individual-level behavior scales up to entire recrea-
tional fisheries is still poorly understood. In this study,
we used several whole-lake experimental fisheries to
investigate mechanisms for decreased catchability. We
used the individual capture histories of tagged fish to
infer the processes leading to changes in observed
catchability. These processes were then modeled as
time series superimposed against the observed fishery
dynamics.
Methods
Experimental lakes.—Our study was conducted on
four lakes that contained naturalized populations of
rainbow trout Oncorhynchus mykiss and were located
on the Bonaparte Plateau north of Kamloops, British
Columbia (5189
0
43.43
00
N, 120823
0
26.34
00
W; altitude ¼
1,500 m; Table 1). One of the lakes was divided into
two sections, which created a total of five experimental
units. Four of the experimental lakes were subjected to
low angler effort and used to detect environmental
effects on catchability. We subjected one small lake to
intensive fishing effort in order to investigate the
effects of angling pressure on catchability.
To maximize our data collection regime, it was
necessary to create a small, fishable lake that had a
high density of catchable-sized fish. This was accom-
plished by quarantining a 1.1-ha section of the 3.2-ha
Pantano Lake. The natural bathymetry of the lake
consisted of two basins separated by a narrow (width ’
12 m) and shallow (mean depth ¼ 0.5 m) section. To
divide the lake, we constructed a fence of rebar and 6-
mm mesh wire fencing across the shallow area between
the two basins. The small basin was named Little
Pantano and the large basin was named Big Pantano.
We used Little Pantano, which had no creeks flowing
in or out, for the high-effort angling experiment. Fish
were captured with fyke nets (hoop diameter ¼ 0.5 m;
mesh diameter ¼ 6 mm) from both basins and a nearby
lake. We graded the catch for the largest individuals
(minimum fork length ¼ 150 mm) to be tagged and
measured. The fish were released into Little Pantano
after a 24-h recovery period in a net pen. In total, 159
indi vidual ly tagged fish were released into Little
Pantano. In addition, we stocked small size-classes of
fish that were batch marked with fin clips as part of a
separate study. Specifically, we released three size-
classes of fish: 100 small (mean fork length ¼ 80 mm),
79 medium (mean fork length ¼ 117 mm), and 80 large
(mean fork length ¼ 147 mm). Use of several size-
classes allowed us to assess size selectivity of the
angling gear and to observe the rate at which each
group was recruited into the fishery.
Between September 29 and October 3, 2004, we
used gill nets and fyke nets to sample Little Pantano.
Over five nights, the sizes and numbers of gill nets
used per hectare were identical to those used by Post et
al. (1999) and Askey et al. (in press). This standardized
gill-net method has been found to be non-size-selective
for taggable-sized fish (.150 mm). We also fished one
to three fyke nets per night to capture small fish ( ,100
mm) as part of a separate study. The efficiency of our
netting effort was assessed by the recovery rates for
three size-classes of fish that had been stocked 1 week
before netting. After making a temporary clip on the
upper tip of the caudal fin, we released 60 small (mean
fork length ¼ 117 mm), 40 medium (mean fork length
¼ 156 mm), and 45 large (mean fork length ¼ 191 mm)
fish into all lakes for the mark–recapture experiment.
Furthermore, we used fyke nets to capture fish in each
TABLE 1.—Physical description of experimental lakes in
British Columbia used to assess catchability of rainbow trout,
and total angler effort for each lake for the entire open-water
season (1 angler-day is 4 angler-hours).
Lake
Area
(ha)
Maximum
depth (m)
Total effort
(angler-days/ha)
Little Pantano 1.1 3 72.2
Big Pantano 2.1 4 2.9
Today 6.5 11 1.5
Stubby 6.2 9 1.4
Spook 4.4 4 1.5
CATCH RATES AND FISH LEARNING 1021
lake in the week before netting; these fish were clipped
and released, which added 52 fish to the mark–
recapture effort.
Capture efficiency for the gill nets and fyke nets of
tagged fish was estimated from the proportion of
marked fish recovered (fork lengths of 180–310 mm,
equivalent to range in size of tagged fish) as p ¼ m/n
where p is the probability of capture, n is the number of
marked fish released, and m is the number of marked
fish that were recaptured.
Angling procedures.—To standardize our angling
treatment as much as possible, we restricted the angling
to three individuals who used similar fly-fishing gear.
We tested fly patterns on a nearby lake that was not
part of the experiment and chose two general patterns,
which were used for all angling. The patterns were tied
on number 14 hooks; one imitated a general nymph
and the other was a leech pattern.
Little Pantano was fished every day for 30 d from
June 13 to July 12, 2004. Two anglers fished from a
single boat for 4 h daily (presented as 1 boat-day of
effort), ending approximately 30 min before dusk. The
same lake was then fished for 8 d in early August and 3
d in early September. Captured fish were brought to the
boat and held in a large bin that contained a small
amount of water; tag and length data were then
collected. Fish were placed into a small recovery bin
beside the boat; and within an hour, they were moved
into one of two large net-pens, where they were kept
overnight. This was done to control for delayed
hooking mortality.
Big Pantano and three other nearby lakes were
subjected to very low angler effort (Table 1). This was
done so that we could measure if seasonal changes in
catchability had occurred independent of angling
pressure. The lakes were fished once per month for a
2-h period by each angler; anglers used the same gear
that was used to fish in Little Pantano. These fishing
events were organized to coincide with the start and
end of the 30-d period on Little Pantano and the August
and September visits to that lake. All the lakes in our
study were accessible to anglers on foot only, which
limited the risk of angling pressure by other parties.
During the entire summer, we witnessed only a single
hike-in party of two individuals on Lakes Today and
Stubby; this observation is included in the Table 1
angler effort.
Analysis of catch data.—The main goal of our
experiment was to test for learned hook avoidance by
fish in recreational fisheries. However, there are at least
three mechanisms that may independently affect catch
rates over the fishing season: (1) growth of individuals
into more vulnerable size-classes (Cox 2000; Parkinson
et al. 2004), (2) environmental and/or ecological
factors (e.g., changes in temperature or insect activity)
that affect foraging behavior (van Poorten and Post
2005), and (3) apparent mortality, w hich is the
summation of death and tag loss in mark–recaptures.
Each of these factors must be incorporated into our
experimental design.
We have explicitly incorporated size into the
probability of capture, because the size-dependence
of vulnerability to angling is well known (Cox 2000;
van Poorten and Post 2005). The relative vulnerability
(v) is assumed to be a sigmoid function of fish length l,
and is expressed as
v
l
¼ p
max
l
m
l
m
þ L
50
m
; ð1Þ
where p
max
is the maximum vulnerability for large
fish, L
50
is the length at which fish are at 50% of full
vulnerability, and m is the slope at that point. To scale
v to the relative vulnerability for an individual, p
max
is
set to 1. The individual vulnerability model was fit to
the proportion of marked fish recovered per 1-cm
size-group in the first 5 d of the fishing experiment on
Little Pantano. These parameters were then fixed for
fitting the model to capture histories as described
below.
A problem exists where environmental effects on
catchability occur simultaneously with learned hook
avoidance and their effects may be difficult to separate.
We therefore divided our angling experiment into two
parts: (1) a set of lightly fished lakes to assess the
seasonal trend in catchability and (2) a single, intensely
fished lake to assess fish learning. This approach
allowed us to first test whether a purely seasonal trend
in catchability exists. Existence of such a trend would
allow us to incorporate it into the multiple mark–
recapture efforts on Little Pantano and to isolate
learning effects from environmental effects on catch-
ability.
The effort on our four lightly fished lakes was very
low (Table 1), so that a negligible proportion of the fish
population had encountered an angling experience at
any given time. Thus, any changes in catchability over
time were because of temporal changes in environ-
mental or ecological factors. Let d be a parameter to
describe the seasonal environmental influence on catch
rates. It could be simply a function of temperature or a
function of multiple environmental factors, which can
be modeled as a function of time:
d
t
¼ f ðtÞ: ð2Þ
Thus, the expected catch rate EC
il
for angler i on lake l
can be modeled as the seasonal average catch rate for
1022 ASKEY ET A L.
angler i on lake l (l
il
) modified by the seasonality
parameter:
EC
il
¼ ld
t
: ð3Þ
The incorporation of the l parameter puts all lakes on a
relative scale, because we are interested in the relative
change in catch rates over the season. Our results are
not sensitive to differences in absolute catchability,
which vary between lakes because of fish density or
lake characteristics. This time- or temperature-depen-
dent expected catch rate is incorporated into the
Poisson log-likelihood function (LL) of a single
observed catch (C)as
LLðCjl
il
; d
t
Þ¼Clog
e
ðl
il
d
t
Þl
il
d
t
X
C
x¼1
log
e
ðxÞ:
ð4Þ
The maximum likelihood for the entire data set of
observed catches given a seasonality effect is
LLðdatajdÞ¼
X
N
x¼1
LLðC
x
jl
x
; dÞ: ð5Þ
We maximized this likelihood for the catch data, where
d was a function of time or temperature. We omitted
fish from the catch data that would not have been
vulnerable (,150 mm) on day 1 to control for
recruitment of catchable fish over the season. Estimates
for d on days not fished was estimated by linear
interpolation. As a result, for any given day of the
season, the baseline catchability could be adjusted by
multiplying by the estimated d.
Individual catchability analysis.—Suppose I fish are
marked on day 0 and released into a lake. On N
occasions, the fish are recaptured. The first N-1
recaptures are by angling and the last recapture is by
net. The day of recapture n is denoted t
n
. Let x
i,n
¼ 1if
fish i (i ¼ 1toI) is recaptured on day t
n
, and x
i,n
¼ 0if
the fish is not recaptured. Let X be the N by I matrix
describing the recapture data.
Now consider a single fish. Let c
n
be the probability
the fish is caught during recapture day n (we will
consider c
n
in more detail below). Let s
n
be the
probability that the fish survives to time of the nth
recapture, given that it was alive at the time of the (n
1)th recapture. If the fish experiences a time-indepen-
dent instantaneous mortality rate m(t) at time t, then the
survival probabilities can be calculated using
s
n
¼
expðmt
1
Þ if n ¼ 1
expðmðt
n
t
n1
ÞÞ if n . 1
:
ð6Þ
The probability of observing the N recapture data
associated with a single fish, given the probabilities of
recapture and survival, is
Prðx
1
; ::: x
N
jc
1
::: c
N
; s
1
::: s
N
Þ
¼
P
N
n¼1
s
n
½x
n
c
n
þð1 x
n
Þð1 c
n
Þ if l ¼ N
P
l
n¼1
s
n
½x
n
c
n
þð1 x
n
Þð1 c
n
Þ
if l , N
3
X
N
j¼lþ1
ð1 s
j
Þ P
j1
n¼lþ1
s
n
ð1 c
n
Þþ P
N
n¼lþ1
s
n
ð1 c
n
Þ
"#
;
8
>
>
>
>
>
>
>
>
>
>
<
>
>
>
>
>
>
>
>
>
>
:
ð7Þ
where l is the last recapture (i.e., x
n
¼ 0 for all n . l). If
the fish is never recaught, l ¼ 0. Assuming the fish
recaptures are independent, the probability of observ-
ing all the data X is the product of the above
probability for all fish. The log-likelihood function
(LL) is log
e
[Pr(X)], which is incorporated into our
model selection criteria, Akaike’s information criterion
(AIC), as
AIC ¼2LL þ 2k; ð8Þ
where k is the number of estimated parameters in the
model. Akaike’s information criterion values are used
to select the most parsimonious model by penalizing
the model fit (LL) by the number of parameters used.
Thus, the optimal models possess the minimal AIC
values and the relative parsimony of other models is
evaluated by the differences between AIC values,
DAIC (Burnham and Anderson 2002; Richards 2005).
We used AIC to test a suite of biologically plausible
models that may describe temporal patterns in catch
rates. A null model describes catchability as a constant
or dependent on environmental factors; however, we
incorporated two additional reasons for decreased catch
rates: (1) learned hook avoidance and (2) heterogeneity
in intrinsic catchability. To incorporate these factors,
the probability of capture (c) for a single fish is
manipulated. The probability of capture is actually a
composite of several factors, expressed as
c
n
¼ 1 e
q
n
v
n
E
n
; ð9Þ
where c is the probability of capture on day n, q is the
catchability coefficient (area swept per angler per unit
time), E is the effort (angler time per area), and v
accounts for the size selectivity of fishing gear
(equation 1).
For a given effort and length on day t, the probability
of capture for an individual depends on the catchability
coefficient (q). In the simplest case, catchability is
constant:
CATCH RATES AND FISH LEARNING 1023
q ¼ q
0
: ð10Þ
However, q may be dependent on the number of times
an individual fish has been captured previously
(denoted tc)
q ¼ q
tc
: ð11Þ
A potentially more parsimonious version of this idea is
to describe q as a cont inuous function of times
previously caught
q ¼ f ðtcÞ: ð12Þ
Since catchability cannot be negative, a logical
function is the negative exponential,
q ¼ q
0
½e
ðbtcÞ
; ð13Þ
where q
0
is the catchability for fish that have never
been captured and b is a parameter that describes the
decline in catchability for an individual fish that has
been previously captured. The above models were then
substituted into equation (9) for capture probability.
A second possibility is that heterogeneity in q occurs
within the population because individual fis h are
intrinsically more or less catchable regardless of their
capture history. This could arise from behavioral
differences in foraging activity or diet preference. To
model this scenario, we assume the existence of
discrete fish classes (Pledger et al. 2003), each with
its own catchability parameters. Each individual has an
unknown probability p
i
of being in class i. We return to
equation (7) for the probability of an individual fish
capture history and sum the Prob(x
1. . .N
js
1. . .N
,c
1. . .N
) 3
Prob(class ¼ i) over all possible fish classes. We tested
the simple case of two fish classes with unknown
proportions (p and 1 – p) in each class, which leaves
the following probability:
Prðx
1::: N
jc
1:::N
; s
1::: N
; pÞ
¼ p 3 Prðx
1::: N
jc
1:::N
; s
1::: N
Þ
þð1 pÞ 3 Prðx
1::: N
jc
1:::N
; s
1::: N
Þ:
We tested another set of models that incorporated the
existence of two classes of fish and catchability based
on capture history.
The final data set for model selection was a matrix of
157 fish (two hooking mortalities were omitted) and 43
capture events with 1 or 0 values. The final capture
event was fall gillnetting, which was used to confirm
survival for the captured fish and estimate abundance
as mentioned above. A second matrix of equal
dimensions was created by using individual fish fork
lengths. Lengths on days when fish were not measured
were estimated using linear interpolation.
Results
Size Selectivity
The proportion of marked fish captured in the first 5
d of fishing varied with mean length (1-cm size bins).
The maxi mum likelihood fit of the size -selective
vulnerability function yielded parameter estimates of
m ¼ 7.95 and L
50
¼ 211.4 (Figure 1), which are similar
to the parameter values found in other studies of
rainbow trout (Cox 2000; van Poorten 2003). The
estimated parameters confirmed that all individually
tagged fish were vulnerable to angling (minimum size
¼ 150 mm), although they had not reached a fully
vulnerable size (i.e., 0 , v , 1). These parameter
estimates were set as constants in the model selection
process.
Seasonal Trends in Catch per Unit Effort
On our four control lakes, the catch rates for fish that
were vulnerable to capture since day 1 showed a
decreasing trend (Figure 2). The trend was not
temperature driven; catch rates did not recover in the
fall when water temperatures decrease. A temperature-
driven model fit the data poorly (LL ¼104.75). We
chose to describe the trend with a four-parameter, time-
dependent model that fit a mean catch rate adjustment
for each fishing period (LL ¼92.43). The data could
be described by a simple linear decrease, but the goal
FIGURE 1.—The relationship between rainbow trout fork
length and vulnerability to experimental angling in British
Columbia lakes during 2004. Size of the points is proportional
to number of marked fish within the 10-mm size bin (smallest
point ¼ 1; largest point ¼ 57). Solid line is the maximum
likelihood fit of equation (1), where data and model have been
scaled so that the maximum vulnerability for large fish is 1.
1024
ASKEY ET A L.
was to fit the seasonal fluctuations as accurately as
possible to create a baseline for the mark–recapture
analysis.
Little Pantano Catch per Unit Effort
Catch rates were initially quite high on Little
Pantano: approximately 16 tagged fish/boat-day were
caught for the first 5 d (Figure 3a). However, the CPUE
declined rapidly to approximately 5 tagged fish/boat-
day by day 15 and remained low for the rest of the 30-d
trial. Catch rates for the tagged fish remained low when
fishing resumed after a break of 23 d.
In addition to the tagged fish, the lake also contained
fin-clipped fish that were initially too small to be fished
but recruited into vulnerable size-classes as they grew.
These fish became more prevalent in the catch over
time and made up most of the catch in the second half
of the summer, after the break (Figure 3b). Continual
recruitment throughout the summer of the small, batch-
marked fish prevented a dramatic drop in catch rates
for the overall population (Figure 3c).
Our mark–recapture data indicated that we recap-
tured 51.6% of fish larger than 180 mm (all tagged fish
were in this size range) after angling had ceased. We
used fyke nets and gill nets for the recapture; given the
netting efficiency, we estimated 41% of the fish tagged
on day 0 remained present with tags at the end of the
experiment. There were three processes by which fish
were removed from the experiment: (1) hooking
mortality, (2) natural mortality, and (3) tag loss.
Hooking mortality was estimated based on observation
of deaths within the 24-h recovery bins. There were
nine mortalities from hooking injuries; however, only 2
were from the 159 fish marked on day 0. Fourteen fish
with visible tag scars were captured in gill nets, which
gives an estimated tag loss of 30%. These fish are
considered mortalities in the data analysis, because no
information can be collected from them (post tag loss).
Thus ‘‘ mortality’’ estimates in model fitting (Table 2)
include tag loss and death during the season. Tag loss
was not a problem for the fitting of individual capture
histories; only seven fish with tag scars were caught,
and all such captures occurred from day 54 on.
Individual Catchability and Model Selection
The first set of models that were fit to the data set of
individual capture histories, focused on the potential
influence of learned hook avoidance by varying
catchability with previous capture experience. The
DAIC values suggested that the abrupt drop in catch
rates could not be described by a constant catchability
adjusted for seasonal effects (Table 2; Figure 4).
FIGURE 2.—Standardized rainbow trout CPUE (CPUE 3
mean CPUE
1
; CPUE in units of fish per angler-hour) for four
lakes in British Columbia subjected to low angling effort over
the summer (day 1 ¼ June 13, 2004; black filled circles ¼ Big
Pantano Lake, open circles ¼ Lake Today, open triangles ¼
Lake Stubby, diamonds ¼ Spook Lake). The black line is the
best-fit model for the seasonal trend (d).
FIGURE 3.—Rainbow trout catch rates (CPUE in fish/boat-
day) in Little Pantano Lake, British Columbia, over the entire
fishing period (June 13–September 3, 2004). The top panel
shows catch rates for fish that were tagable from day 1
(minimum size ¼ 180 mm). The middle panel shows catch
rates for batch-clipped fish that were too small to individually
tag and that were essentially invulnerable to angling on day 0.
These fish recruited into the fishery by growth. The bottom
panel shows catch rates for all fish present. One boat-day
equals two anglers fishing for 4 h from a single boat.
CATCH RATES AND FISH LEARNING 1025
Models that incorporated experience-dependent catch-
ability models were found to be more parsimonious
(Table 2). The parameter estimates indicated that fish
became less catchable if they had been previously
captured. Furthermore, when specific catchabilities
were fitted to the model for capture history, it was
found that fish catchability continued to decrease with
additional capture experiences. Model 5 had the
optimal fit and depicted catchability as a negative
exponential function of times caught. However, none
of the models that were based on learned hook
avoidance alone were flexible enough to mimic the
sharp initial decrease in catch rates seen in the data
(Figure 4). All models that were based on a single class
of fish underestimated the catch rates seen in the
beginning of the angling experiment
Simply dividing the population into two classes
(with regards to intrinsic catchability) is not helpful, as
model fitting produced equal catchabilities between
classes (q
a
¼ q
b
) or the existence of a single class (p ¼
1). However, models that incorporated learning with
heterogeneity in catchabilities were able to better fit the
trend seen in the catch data. The most parsimonious
model separated the population into two classes based
on intrinsic catchability and both classes exhibited the
same learned hook avoidance function (Table 2; Figure
4). The parameter estimates for this model indicated
that about 32% (p ¼ 0.322) of the entire population
were highly catchable fish that quickly learned to avoid
hooks. Model fitting sugg ested that both classes
learned at a similar rate; a sixth parameter (b
b
) was
not justified (DAIC ¼ 0.7; Table 2). The change in AIC
from the best-fit model to the next best single-class
model was greater than 20, which indicates that the
model including heterogeneous intrinsic catchability is
substantially better than the learning-only model
(Burnham and Anderson 2002; Richards 2005).
TABLE 2.—Summary of rainbow trout catchability models tested, including associated parameter values and fitting
performance. Parameters and abbreviations are as follows: qa ¼ catchability coefficient for class a; qb ¼ catchability coefficient
for class b; tc ¼ times caught previously; ba and bb are slope parameters; l ¼ catch rate; p ¼ probability of being in class i; k ¼
number of estimated parameters; LL ¼ log likelihood; AIC ¼ Akaike’s information criterion; DAIC ¼ difference in AIC between
the given model and the model with the lowest AIC value.
Model specification
Model number Equation Description
1 q
t
¼ q Constant q
2 q
t
¼ q
tc
q changes after first capture event
3 q
t
¼ q
tc
q changes for capture events 1 and 2
4 q
t
¼ q
tc
q changes for capture events 1–3
5 q
t
¼ q
0
3 e
(b3tc)
q is a continuous function of times caught
6 q
t
¼
qa class ¼ a
qb class ¼ b
Two classes of fish with constan t qs
7 q
t
¼
qa
0
3 e
ðb
1
3 tcÞ
class ¼ a
0 class ¼ b
Continuous learning with an invulnerable class
8 q
t
¼
qa
0
3 e
ðb 3 tcÞ
class ¼ a
qb
0
3 e
ðb 3 tcÞ
class ¼ b
Two classes of fish with continuous learning
9 q
t
¼
qa
0
3 e
ðba 3 tcÞ
class ¼ a
qb
0
3 e
ðbb 3 tcÞ
class ¼ b
Two classes with independent learning parameters
FIGURE 4.—Model fits of empirical rainbow trout catch data
for Little Pantano Lake, British Columbia, which contained
157 fish tagged on day 0 of an angling experiment in 2004.
The dashed line is model 1 (see Table 2) based on the simple
case of constant catchability adjusted for seasonal effects. The
solid gray line is model 5, where catchability is a negative
exponential function of the number of times a fish has been
caught previously. The solid black line is model 8, which
incorporates learning and two classes of fish within the
population that differ in their intrinsic catchability. Seasonal
effects (d) are incorporated into all models as depicted in
Figure 2, and jaggedness of lines represents variability in
effort.
1026
ASKEY ET A L.
Discussion
There have been many hypotheses put forth to
explain seasonal decreases in recreational fishery catch
rates. Our study shows that, indeed, several compo-
nents explain this phenomenon, including learned hook
avoidance, heterogeneity among individual fish, and
environmental factors. The culmination of these
components led to a sharp decrease in daily catches
from 16 to 4 fish (tagged individuals) within 30 d of
intensive catch-and-release angling.
Learned hook avo idance was a key component
needed to explain the large data set of individual
capture histories. Our study supports previous studies
that have reported the poten tial for learned hook
avoidance in fished populations. Similar evidence in
studies of other sport fish (Anderson and Heman 1969;
Beukema 1970; Hackney and Linkous 1978) indicates
that learned hook avoidance is not restricted to rainbow
trout. However, only some of the largemouth bass
Micropterus salmoides experimental groups exhibited
learning (Anderson and Heman 1969; Hackney and
Linkous 1978). Furthermore, Tsuboi and Morita (2004)
found no evidence of learning among whitespotted char
Salvelinus leucomaenis in a Japanese stream, and
cutthroat trout O. clarkii in Yellowstone River were
estimated to be captured 9.7 times per season (Schill et
al. 1986). This suggests that differences in learning and/
or habitats may make some species more suited for
catch-and-release management than others. Previous
work has demonstrated differences in catchability
among species and s trains owing to variation in
behavior (e.g., Brauhn and Kincaid 1982; Dwyer
1990). It seems plausible that conditioning should vary
because of species-specific behavioral characteristics as
well. The lack of learning behavior demonstrated by
whitespotted char in Japan and cutthroat trout in the
Yellowstone River could also be a result of the lotic
environment. For example, the nature in which food is
presented in a stream necessitates a rapid response by
the fish or the food will be lost downstream. Therefore,
fish cannot examine and potentially reject their prey to
the same degree possible in a lake. However, disturbed
or angled fish in New Zealand streams were found to
exhibit other conditioned behaviors, such as hiding, that
would prevent their capture (Young and Hayes 2004).
Learned hook avoidance was only part of the process
suggested to explain seasonality in catch rates. It was
only possible to fit the sharp initial decline in catch
rates if intrinsic differences in fish were also taken into
account. It seems plausible that some proportion of fish
populations should be less vulnerable to angling
because of factors such as highly selective diets. Cox
and Walters (2002) proposed that some proportion of
the population is unavailable to angling due to spatial
distribution in unfishable areas. Our experimental lake
was small and relatively shallow so that the entire lake
could be fished effectively. However, our data
supported a similar hypothesis, whereby, a proportion
of the fish are highly susceptible to angling but quickly
learn to avoid hooks. Martin (1958) originally
proposed that pop ulations may contain a more
catchable group of fish that would be quickly
harvested. Yet we have shown that even when fish
are not harvested, the same decline in CPUE will occur
because of a more catchable group that exhibits
learning. This result indicates that lakes exposed to
TABLE 2.—Extended.
Model specification Parameters Fit and selection
Model number l qa
0
qa
1
qa
2
qa
.2
p qb b
a
b
b
k LL AIC DAIC
1 0.011 0.047 0.047 0.047 0.047 1 2 749.2 1,502.3 33.0
2 0.011 0.059 0.037 0.037 0.037 1 3 745.5 1,497.1 27.8
3 0.011 0.059 0.046 0.023 0.023 1 4 742.7 1,493.5 24.2
4 0.011 0.059 0.046 0.024 0.021 1 5 742.7 1,495.5 26.2
5 0.011 0.061 1 0.375 3 743.0 1,492.0 22.7
6 0.011 0.047 0.500 0.047 4 749.2 1,506.3 37.0
7 0.010 0.101 0.747 0 0.714 4 737.6 1,483.3 14.0
8 0.009 0.299 0.322 0.030 1.125 1.125 5 729.9 1,469.3 0.0
9 0.009 0.382 0.265 0.033 1.126 0.550 6 729.0 1,470.0 0.7
CATCH RATES AND FISH LEARNING 1027
fishing for the first time are likely to exhibit a short
period of exceptional angling that is rapidly reduced to
average catch rates, regardless of bag-limit regulations.
Catch rates may continue to decline with continued
pressure; however, the decline is much slower after the
original ‘‘ fishdown’’ period.
The dramatic drop in catch rates of tagged fish seen
in our experiment may have been more extreme than
would be typical in nature. Our protocol of retaining
fish in net pens to control for hooking mortality may
also have stressed fish and changed their behavior.
Given our data, we cannot separate the relative effect of
stress in net pens from the stress incurred during
capture. It was also apparent that catch rates were
affected by recruitment of fish that had grown over the
summer. However, if effort is sustained over the
summer, it seems likely that growth recruitment will
only prevent further declines after the abrupt fishdown
early in the season. We may have exaggerated the
learning ability of fish by only using two fly patterns
for all angling. Presumably we could have continued to
entice strikes if we would have changed to new
patterns that the fish had not previously seen. However,
fish probably take some cues from characteristics
common to all lures (e.g., visible hooks or fishing line).
Lastly, a small part of the decline in catch rates was
fishery independent; catch rates were found to decline
in the lightly fished lakes. Thus, other geographical
regions that have different climatic patterns and fish
species may experience different seasonal trends.
As angling effort continues to increase in many
fishing areas, managers are challenged to produce
quality angling opportunities. Trophy fisheries are
normally managed by high minimum size limits or
catch-and-release regulations. However, as demand and
effort on such fisheries increases, even catch-and-
release fisheries can produce relatively low catch rates.
This result was previously suggested based on hooking
mortality and illegal harvest (Post et al. 2002, 2003).
However, hooking mortality and illegal harvest are
potentially lowered through management actions other
than effort control. Low catch rates due to learning
limit management strategies to: (1) finding strains of
fish that have high catchability, low hooking mortality,
and exhibit poor learning ability; or (2) effort control.
Managers are faced with the difficult notion that if they
are successful in generating license sales, they will not
be able to maintain angling quality by simply setting
restrictive bag limits, size limits, or even by imple-
menting catch-and-release regulations.
Acknowledgments
We thank Chris Dormer and Bobby Beddingfield,
who worked as field assistants and angled many hours
withoutcatchingmanyfishattheendofthe
experiment. Thanks to David Obrien and Nathan
Taylor, who helped with field work, especially the
construction of the fence in Pantano Lake. Carl Walters
provided useful comments on an earlier version of the
manuscript. Critique from the associate editor and three
anonymous reviewers further improved this manu-
script. Financial support was provided by the British
Columbia Ministry of Environment and British Co-
lumbia Freshwater Fisheries Society to E.A.P. and by a
Natural Sciences and Engineering Research Council
Discovery Grant to J.R.P.
References
Aldrich, A. D. 1939. Results of seven years’ intensive
stocking of Spavinaw Lake; an impounded reservoir.
Transactions of the American Fisheries Society 68:221–
227.
Anderson, R. O., and M. L. Heman. 1969. Angling as a factor
influencing the catchability of largemouth bass. Trans-
actions of the American Fisheries Society 98:317–320.
Askey, P. J., J. R. Post, E. A. Parkinson, E. Rivot, A. Paul,
and P. A. Biro. In press. Estimation of gill net efficiency
and selectivity across multiple sampling units: a
hierarchical Bayesian analysis using mark–recapture.
Fisheries Research.
Barnhart, R. A. 1989. Symposium review: catch-and-release
fishing, a decade of experience. North American Journal
of Fisheries Management 9:74–80.
Beukema, J. J. 1970. Angling experiments with carp
(Cyprinus carpio L.), II. Decreasing catchability through
one-trial learning Netherlands Journal of Zoology 20:81–
92.
Brauhn, J. L., and H. Kincaid. 1982. Survival, growth, and
catchability of rainbow trout of four strains. North
American Journal of Fisheries Management 2:1–10.
Burnham, K. P., and D. R. Anderson. 2002. Model selection
and multimodel inference: a practical information-
theoretic approach, 2nd edition. Springer-Verlag, New
York.
Cooke, S. J., and C. D. Suski. 2005. Do we need species-
specific guidelines for catch-and-release recreational
angling to effectively conserve diverse fishery resources?
Biodiversity and Conservation 14:1195–1209.
Cox, S. P. 2000. Angling quality, effort response, and
exploitation in recreational fisheries: field and modeling
studies on British Columbia rainbow trout (Oncorhyn-
chus mykiss) lakes. Doctoral dissertation. University of
British Columbia, Vancouver.
Cox, S. P., and C. Walters. 2002. Modeling exploitation in
recreational fisheries and implications for effort manage-
ment on British Columbia rainbow trout lakes. North
American Journal of Fisheries Management 22:21–34.
Dwyer, W. P. 1990. Catchability of three strains of cutthroat
trout. North American Journal of Fisheries Management
10:458–461.
Hackney, P. A., and T. E. Linkous. 1978. Striking behavior of
the largemouth bass and use of the binomial distribution
for its analysis. Transactions of the American Fisheries
Society 107:682–688.
1028
ASKEY ET A L.
Martin, R. G. 1958. Influence of fishing pressure on bass
fishing success. Proceedings of the Annual Conference of
the Southeastern Association of the Game and Fish
Commissioners 11(1957):76–82.
Parkinson, E. A., J. R. Post, and S. P. Cox. 2004. Linking the
dynamics of harvest effort to recruitment dynamics in a
multistock, spatially structured fishery. Canadian Journal
of Fisheries and Aquatic Sciences 61:1658–1670.
Pledger, S., K. H. Pollock, and J. L. Norris. 2003. Open
capture-recapture models with heterogeneity: I. Cor-
mack-Jolly-Seber model. Biometrics 59:786–794.
Post, J. R., E. A. Parkinson, and N. T. Johnston. 1999.
Density-dependent processes in structured fish popula-
tions: interaction strengths in whole-lake experiments.
Ecological Monographs 69:155–175.
Post, J. R., M. Sullivan, S. Cox, N. P. Lester, C. J. Walters, E.
A. Parkinson, A. J. Paul, L. Jackson, and B. J. Shuter.
2002. Canada’s recreational fisheries: the invisible
collapse? Fisheries 23(1):6–17.
Post, J. R., C. Mushens, A. Paul, and M. Sullivan. 2003.
Assessment of alternative harvest regulations for sus-
taining recreational fisheries: model development and
application to bull trout. North American Journal of
Fisheries Management 23:22–34.
Richards, S. A. 2005. Testing ecological theory using the
information-theoretic approach: examples and cautionary
results. Ecology 86:2805–2814.
Schill, D. J., J. S. Griffith, and R. E. Gresswell. 1986. Hooking
mortality of cutthroat trout in a catch-and-release
segment of the Yellowstone River, Yellowstone National
Park. North American Journal of Fisheries Management
6:226–232.
Tsuboi, J., and K. Morita. 2004. Selectivity effects on wild
white-spotted charr (Salvelinus leucomaenis) during a
catch and release fishery. Fisheries Research 69:229–
238.
Van Poorten, B. T. 2003. Angling impacts on rainbow trout
dynamics. Master’s thesis. University of Calgary,
Calgary. Alberta.
Van Poorten, B. T., and J. R. Post. 2005. Seasonal fishery
dynamics of a previously unexploited rainbow trout
population with contrasts to established fisheries. North
American Journal of Fisheries Management 25:329–345.
Young, R. G., and J. W. Hayes. 2004. Angling pressure and
trout catchability: Behavioral observations of brown trout
in two New Zealand backcountry rivers. North American
Journal of Fisheries Management 24:1203–1213.
CATCH RATES AND FISH LEARNING 1029
