Wa. Walsh et P. Kleiber, Generalized additive model and regression tree analyses of blue shark (Prionace glauca) catch rates by the Hawaii-based commercial longline fishery, FISH RES, 53(2), 2001, pp. 115-131
Generalized additive model (GAM) and regression tree analyses were conducte
d with blue shark, Prionace glauca, catch rates (catch per set) as reported
by National Marine Fisheries Service observers serving aboard Hawaii-based
commercial longline vessels from March 1994 through December 1997 (N=2010
longline sets). The objective was to use GAM and regression tree methodolog
y to relate catch rates to a tractable suite of readily measured or compute
d variables. Because the predictor variables are also either provided in or
easily computed from the logbooks that commercial vessels submit upon land
ing fish for sale, it is likely that a model or models fitted to accurate o
bserver data could then be applied on a fleet-wide basis to serve as a stan
dard of comparison for the logbooks. The GAM included nine spatio-temporal,
environmental, and operational variables and explained 72.1% of the devian
ce of blue shark catch rates. Latitude exerted the strongest effects of any
individual variable; longitude was the most influential variable when adju
sted for the effects of all other factors. Relatively cold sea surface temp
eratures were associated with high catch rates. The initial regression tree
included 68 terminal nodes and I I predictors. It was refined to a final t
ree with 42 terminal nodes, which reduced the root mean deviance by 65.3%.
The tree was partitioned first on latitude 26.6 degreesN, and then branched
out to reach terminal nodes after 2-8 additional partitionings. Sets south
of this latitude were characterized by lower catch rates and partitionings
on a greater number and variety of predictors. Northerly sets were charact
erized by higher and more variable blue shark catch rates. Predictions from
the two analyses were highly correlated (r=0.903. P much less than0.001).
Moreover, use of these methods in combination aided greatly in the interpre
tation of results. We conclude that GAM and regression tree analyses can be
usefully employed in the assessment of blue shark catch rates in this fish
ery. We suggest that either or both of these models could serve as comparis
on standards for commercial logbooks. Published by Elsevier Science B.V.