8136

• UCREFRP Catalog Number: 8136
• Author: Bestgen, K. R.
• Title: Growth, survival, and starvation resistance of Colorado squawfish larvae.
• USFW Year: 1995.
• USFW - Doc Type: \
• Copyright Material: YES

..

300 C = 30.2 (0.04); diel fluctuating temperatures
(mean minimum and maximum) were 180 C = 15.7
(0.14) and 20.5 (0.20), 220 C = 19.5 (0.14) and 24.7
(0.09), and 260 C = 23.5 (0.21) and 28.3 (0.09).
Larvae were fed measured aliquots of nauplii of
brine shrimp, Artemia sp., in morning and evening
at nominal amounts of 6.25,14,32,71, and 160 nau-
plii fish-l feeding-I; total daily food ration is report-
ed throughout. Periodic counts of brine shrimp nau-
plii indicated that nominal amounts fed were not
significantly different (p = 0.15) than target
amounts. Smallest brine shrimp nauplii were able to
pass through the netting, although it was qualita-
tively estimated that escapement was < 10%. Feed-
ing rates were adjusted daily when mortality oc-
curred. Fish did not begin feeding until 4 dafter
treatments began (6 d after hatching) but were sub-
jected to experimental treatment conditions prior
to feeding so that developmental rates would be
consistent with test temperatures.
Each treatment combination of temperature (18,
22,26, or 300 C), regime (fluctuating or constant),
and food abundance (12.5,28,64,142, or 320 nauplii
fish-l day-I) was replicated three times, except that
the 220 C constant regime treatment (with five ra-
tion levels) was conducted in both 1991 and 1992 to
test for annual variation, and no 300 C fluctuating
treatment was conducted due to lack of space. Thus,
a total of 120 experimental units was used.
Survival in aquaria was monitored daily by mak-
ing counts of larvae until successive counts were the
same. Dead fish were removed, labelled, and pre-
served in 100% ethanol. Experiments were ended
when 750 degree days were accumulated: 41.7 d for
180 C treatments (fluctuating and constant regimes
had similar thermal units); 34 d for 220 C treat-
ments; 28.8 d for 260 C treatments; and 25 d for the
constant 300 C. Remaining larvae were counted,
preserved, and measured (nearest 0.1 mm TL).
Shrinkage of preserved specimens was < 3 % of
fresh TL (unpublished data) so no correction was
made. Mean growth rate (mm d-1) for each replicate
was based on the surviving fish.

199

Statistical analysis

Response surface analysis can efficiently show the
joint effects on a response of different levels of one
or more independent variables. Therefore, re-
sponse surface methods were used to analyze ex-
perimental data, where growth (G) and survival (S)
were response variables and regime, temperature,
and food abundance were the independent varia-
bles. Growth (mm d-l) data were analyzed with
PROC RSREG (SAS Institute 1988) to obtain
least-squares estimates of model coefficients; sur-
vival data (as logit S = In (Y/N-Y), where Y = num-
ber surviving out of20 (N) individuals) were model-
ed in PROC GENMOD (SAS Institute 1993) to ob-
tain maximum likelihood estimates of model coeffi-
cients and standard errors. Regression models had
the form

Y = ~o + ~IXI + ~2X2 + ~3X3 + ~l1X/ +
~22X22 + ~33X/ + ~12XIX2 + ~13XIX3 + ~23X2X3 +
~123XIX2X3 + E,

where in this instance, Y = G or logit (S); Xl = water
temperature; X2 = food abundance; X3 = fluctuat-
ing or constant regime; ~o = intercept; ~l' ~2' ~3 =
coefficients of linear terms; ~11' ~22' ~33 = coefficients
of quadratic terms; ~12' ~13' ~23' ~123 = coefficients of
cross products; E = random error, normal (G model)
or binomial (S model) distribution. Rationale for
response surface designs and analyses is detailed in
Box & Draper (1987). Recent applications to ex-
perimental data and supporting discussions are in
Clancy & King (1993) and Scholz et al. (1994).
Global growth and survival response models
were first constructed with all independent varia-
bles with final growth model selection based on
Mallows Cp and final survival model selection based
on Akaike's information criterion (AIC) adjusted
for overdispersed data (QAIC), sensu Anderson et
al. (1994). Unique optima of response variables
with respect to independent variables was achieved
by setting the partial derivatives equal to zero and
solving. Lack-of-fit (LOF) tests, residual plots, and
model F-statistics were computed to determine ad-
equacy of the growth model fit to the data. F-statis-
tics were used to determine if functions satisfied the