8163

• UCREFRP Catalog Number: 8163
• Author: Osmundson, D. B. and K. P. Burnham.
• Title: Status and Trends of the Endangered Colorado Squawfish in the Upper Colorado River.
• USFW Year: 1998.
• USFW - Doc Type: \
• Copyright Material: YES

STATUS OF ENDANGERED COLORADO SQUAWFISH

959

population abundance. Because comprehensive lit-
erature exists on CJS models, data analysis based
on these models, and analysis strategy including
model selection (e.g., Buckland 1980; Pollock
1982; Burnham et al. 1987; Pollock et al. 1990;
Burnham and Anderson 1992; Lebreton et al.
1992; Anderson et al. 1994; Burnham et al. 1995),
only minimal information is provided here.
The analysis strategy was to use the program
RELEASE (Burnham et al. 1987) to compute the
goodness of fit of the general time-specific CJS
model for the capture data. With an acceptable fit,
program SURGE (Lebreton et al. 1992) was used
to find the best model out of a set of models con-
sidered. Akaike's Information Criterion (AIC) was
used for model selection (Akaike 1973; Burnham
and Anderson 1992; Lebreton et al. 1992; Ander-
son et al. 1994; Burnham et al. 1995). The AIC is
a relative measure of how appropriate a model is
for a given set of data, with the "best" model
having the smallest AIC (AICmin). The difference
in AIC value between a given model and the best
model is defined as ~AIC. Models within about
~AIC = 2 of the best model were considered plau-
sible alternative models for the data.
The CJS model is based on two types of param-
eters: survival rate (<I>) and capture probability (p),
the probability of an individual being caught (see
Lebreton et al. 1992), which may vary by capture
occasion and time interval between occasions. A
circumflex n over a parameter indicates it is es-
timated (e.g., fJ). Models considered were the fully
time-specific CJS model and simpler models based
on a mixture of features: constant per-unit time
survival rates, year-specific survival rates, season-
specific survival rates, and capture rates that could
be year- or season-specific or constant, or that
could show a linear trend on either occasion or
year. Models with full time variation parameters
were denoted by { <1>" p,} and with no time variation
by {<I>, p}. Intermediate models included {<I>" p}
and {<I>, p,}. The notation Pyeor was used to denote
different capture rates between years with constant
capture rates for the three within-year capture oc-
casions. Survival rates differing in the intervals
between sampling (three intervals for the upper
reach) were denoted by <l>seasons. Models with sim-
ilar survival rates in short intervals (t = 3 weeks
or roughly 0.06 year) between spring samples but
different rates in the time interval over the rest of
the year (t = roughly 0.88 year in the upper reach)
were denoted <l>short; long' A model was also con-
sidered where no mortality was assumed during

the short time intervals, and the survival parameter
was denoted <l>short ~ I: long'
The CIS models were also used to ascertain ev-
idence of time trends in capture rates, which would
affect the interpretation of time trends in abun-
dance. These analyses were based on fitting CJS
models with capture probabilities (pJ having a lin-
ear trend over time. Denoted PT, these models used
logit (pJ = loge [pJ(1 - pJ] (see e.g., Burnham
et al. 1996; Franklin et al. 1996). A plausible al-
ternative model sets capture probabilities equal
within a year, but with a trend over the four years
(denoted py). By using Yi as a year code, the year-
specific time trend on capture probabilities was of
the form logit (Pi) = a + b. Yi.
Survival rates in capture-recapture models are
apparent survival probabilities, and <I> = S(F)(1 -
L), where F (fidelity) is the probability that a fish
returns to (or stays in) the river reach being sam-
pled and L is the probability of tag loss. Because
PIT tag loss or failure is minimal (Burdick and
Hamman 1993), L is likely to be close to zero.
Thus, if F is near 1, then <I> is an estimator of
physical survival rate (S), and 1 - S is the total
mortality rate.
Abundance and recruitment estimation.-Data
were too sparse to get useful occasion-specific es-
timates of population size based on open-model
methods (such as with program JOLLY; Pollock
et al. 1990). However, by grouping sampling pe-
riods into sets of three passes within a year, the
design corresponded closely enough to the robust
design (Pollock 1982) that closed-model capture-
recapture methods could be used to estimate pop-
ulation abundance each sample year. This analysis
assumes population closure over a time period of
about 6 weeks. The simplest model (the null mod-
el) of program CAPTURE (White et al. 1982) was
used to get an abundance estimate (N) by year.
This model assumes the same P applies at each
capture occasion within year, but p can vary by
year since the estimates were done separately. The
choice of this model was partly based on analyses
with SURGE that supported constant within-year
p, partly motivated by the sparseness of the data,
and supported by the model selection algorithm in
CAPTURE that suggested the use of this simplest
model for each year.
A modified open population analysis method
(program RECAP; Buckland et al. 1980) was used
as a consistency check for average population
abundance over the four years, for average annual
survival rate, and to obtain some information about
average annual recruitment rate (B) into the sam-