Norcross et al. : Habitat models for juvenile pleuronectids 
517 
feeding rates, and shallow, warm waters promote 
faster growth (Malloy and Targett, 1991; van der Veer 
et al., 1994). 
Distribution of juvenile flatfishes has been linked 
to substrate type (Tanda, 1990; Kramer, 1991; Gibson 
and Robb, 1992). Juvenile flatfishes appear to avoid 
coarse sediments (Moles and Norcross, 1995) and 
choose fine-grained sediments (Rogers, 1992; Keefe 
and Able, 1994) which vary in size from mud 
(Wyanski, 1990; van der Veer et al., 1991) to sand 
( Jager et al., 1993). In laboratory tests, rock sole pre- 
fer sand and mixed sand substrates, halibut prefer a 
combination of mud and fine sand, and yellowfin sole 
prefer mud and mixed mud sediments (Moles and 
Norcross, 1995); these findings are in agreement with 
the classification and regression trees of our study. 
Choice of settlement location is affected by the abil- 
ity of a fish to bury itself in the sediment (Gibson 
and Robb, 1992) as well as by the availability of prey 
in the substrate (Burke et al., 1991). When diets of 
juveniles of the four species were examined from the 
same collections in 1991 that were used in these 
models, it was found that epibenthic crustacean taxa 
composed most of the diets (Holladay and Norcross, 
1995). Stomach contents were related to physical 
parameters of capture, including location, depth, and 
substrate. When distribution of juveniles overlapped, 
dietary overlap was sometimes reduced, in that 
one or more groups of flatfishes appeared to alter 
their feeding (Holladay and Norcross, 1995), i.e. pref- 
erence for specific prey types did not appear to be 
a primary factor governing distribution of these 
species. 
A discriminant analysis was employed in this study 
to test whether stations could be accurately classi- 
fied into groups defined by the presence or absence 
of a given flatfish species. The classification based 
on the observed parameters resulted in relatively 
high error rates for all species; between one-sixth and 
one-third of the stations were incorrectly classified. 
Although no discrimination method is able to pre- 
dict perfectly the presence or absence of populations 
that have a gradation in abundance in marginal habi- 
tats, there are several possible reasons for the ob- 
served high error rates found in this study. For rock 
sole, halibut, and yellowfin sole, error rates for pre- 
dicting presence were generally much lower than 
error rates for predicting absence. This finding may 
indicate that these species were mostly confined to 
relatively well-defined depth-substrate characteris- 
tics. The high misclassification rate for predicting 
absence of rock sole, halibut, and yellowfin sole sug- 
gests that many stations may offer suitable depth, 
temperature, and substrate conditions for these spe- 
cies but that the species are not collected there be- 
cause their physical habitat preferences may be dif- 
ferent. The situation is different for flathead sole; 
their presence is not as predictable as their absence. 
Flathead sole are generally absent from shallow ar- 
eas with little mud, whereas they are usually, but 
not always, present in deep, muddy places. 
The classification results suggest that although the 
seven environmental variables (%sand, %mud, 
%gravel, depth, temperature, salinity, and distance 
from bay mouth) used in our discriminant analysis 
do not account fully for observed flatfish distribu- 
tions, they do provide a useful first step at defining 
juvenile flatfish habitat near Kodiak. The initial lin- 
ear discriminant function models developed with the 
1991 data (Norcross et al., 1995) are still applicable 
after incorporating 1992 data. Similar linear dis- 
criminant methods have been used to examine nurs- 
ery grounds of Solea solea (Rogers, 1992). 
Regression trees of CPUE for each species gener- 
ally agree with the results of the linear discriminant 
analyses. They determine specific values of the physi- 
cal parameters as related to the abundance of juve- 
nile flatfishes and, as easily comprehensible dia- 
grams, can be used to predict species abundance 
based on habitat parameters. 
This detailed analysis, based on CPUE and incor- 
porating both 1991 and 1992 data, does not disagree 
with the original models that we were testing 
(Norcross et al., 1995) but rather refines those mod- 
els and incorporates actual abundances (CPUE) in 
the multivariate analysis. The previous models char- 
acterized nursery areas of age-0 rock sole, flathead 
sole, Pacific halibut, and age-1 yellowfin sole on the 
basis of correlations and discriminant analyses by 
using presence or absence for 1991 data. Depth and 
substrate were statistically significant variables pre- 
viously, and a measure of distance in relation to 
mouth of the bay was included qualitatively for each 
species. Depth, temperature, sediment composition, 
and distance from bay mouth were all found to be 
important predictors of the abundance of juvenile 
pleuronectids with regression trees for the combined 
1991 and 1992 data. 
Additional factors influence the presence or ab- 
sence of these flatfish species at any given site. Pos- 
sible factors that were not included in this study are 
additional measures of location (such as position 
around the island or distance from shore), abun- 
dances of prey or predators, and a substrate or habi- 
tat parameter that would account for microhabitat 
features not reflected in sediment composition. 
A location parameter may be a categorical vari- 
able that assigns each station to a well-defined geo- 
graphical area. For example, we observed large dif- 
ferences in the abundance of halibut and rock sole 
