Marancik et al.: Spatial and temporal distribution of grouper larvae in the Gulf of Mexico and Straits of Florida 
5 
by using functions written for MATLAB (for Mac, vers. 
R2010a; The MathWorks Inc., Natick, MA). 
Before formal analyses of larval grouper distribu- 
tions, steps were taken to control for inconsistencies in 
sampling effort over the long time-scale of SEAMAP 
sampling. Each station sampled during SEAMAP was 
assigned to a cell within a 0.5°x0.5° resolution grid 
encompassing the northern Gulf of Mexico (23-30°N 
latitude, and 81-98°W longitude; Fig. IB). If more than 
one station was sampled within a grid cell during a 
single month of any year, the mean value of each envi- 
ronmental variable was taken. This procedure provided 
a sampling regime that was consistent over time and 
facilitated comparisons between environmental and 
larval fish data. Owing to the uneven spatial sampling 
effort among seasons and the low total abundance of 
grouper in Gulf of Mexico samples, 1) larvae were stan- 
dardized to presence or absence within each grid cell 
for each month of each year sampled, 2) no size-specific 
analyses were conducted, 3) larvae collected from bongo 
and neuston samples were combined, and 4) statisti- 
cal analyses were limited to samples collected during 
spring (April-May) and fall (September-October) from 
1986 through 2005. 
Influence of environmental factors and change over 
time Interannual variability in Gulf of Mexico regional 
larval grouper habitat use was examined by using 
generalized additive models (GAMs), a regression tech- 
nique used to fit nonlinear relationships. Seasonal 
mean surface temperature, mean surface salinity, mean 
water depth, and year for subregions of the Gulf of 
Mexico were modeled to predict interannual variability 
in percent frequency of occurrence (%FO; Hastie and 
Tibshirani, 1990; Wood, 2006). The northern Gulf of 
Mexico (north of 23°N) was divided into subregions 
(labeled a-e in Fig. 2) that reflected the presence of 
grouper larvae and orientation of the coastline in rela- 
tion to bathymetry. Within each subregion, %FO was 
calculated as the number of grid cells in which any 
grouper were present divided by the number of grid 
cells sampled during spring (April-May; 1986-2005) or 
fall (September-October; 1986-2005) surveys. GAMs 
are most effective for data sets with few zeros (years 
sampled, but no grouper collected); therefore GAMs 
were generated only for subregions and seasons (i.e., 
spring or fall) during which grouper were collected in 
at least 60% of the years being analyzed. Models of data 
collected during spring surveys were limited to depths 
<900 m to reduce the number of grid cells included in 
analyses owing to the near absence of grouper larvae 
at depths >900 m. With these restrictions, only 3 of the 
5 subregions (Fig. 2, subregions b-d) contained enough 
data on which to base a model. Data from both bongo 
and neuston net samples were combined in order to 
include as many larval grouper data as possible. GAMs 
generated for bongo data provided similar, but weaker, 
results; therefore the combined data are presented. The 
full model used to explain %FO within subregion (r) and 
season (s) was the following: 
