178 
Fishery Bulletin 111(2) 
GAM input, SSHA, SST, and chlorophyll-a data were 
calculated on a spatial grid of l°xl° to match with the 
spatial resolution of the fisheries data. We resampled 
the remotely sensed data to resolutions of 9 km and 1° 
through the use of geographic information system tools, 
including Generic Mapping Tools (GMT, vers. 4.5.7 
[Wessel and Smith, 1998]), with the nearest-neighbor 
technique. Nearest-neighbor assignment can be applied 
with the resample function as a preprocessing step be- 
fore combination of raster data of different resolutions. 
This assignment does not change any of the values of 
cells from the input raster data sets; the cell center 
from the input raster that is closest to the cell center 
for the output processing is used. 
Nino 3.4 index The Nino 3.4 index was used as a cli- 
matic index of ENSO indicators based on SST. The in- 
dex is the average SST anomaly in the region bounded 
by 5°N to 5°S and 120-170°W. The Nino 3.4 index was 
downloaded from the NOAA Climate Prediction Cen- 
ter (http://www.cpc.ncep.noaa.gov). El Nino and La 
Nina events were identified if the 5-month running 
average of the Nino 3.4 index exceeded +0.5°C for El 
Nino or -0.5°C for La Nina for at least 5 consecutive 
months (this index is shown as the dashed line in Fig. 
2A). 
Catchability coefficient 
Catchability was defined as the proportion of available 
fish in the population that would be caught by a unit 
of fishing effort. Catchability depends on the distribu- 
tion of fishing effort in relation to the distribution of 
the target species (Ellis and Wang, 2007). The catch- 
ability coefficient is defined as the proportion of the 
total stock taken by one unit of effort (Haddon, 2011) 
and is expressed as 
C / (q E) = B, (1) 
where C = catch; 
q = the catchability coefficient; 
E = the amount of fishing effort; and 
B - stock biomass. 
The unit of effort used for calculation of longline 
catch rates is the number of hooks, as a result of 
changes in gear practices (Ward and Meyers, 2004). 
We computed the catchability coefficient for all months 
during 1997-2000. 
Empirical orthogonal function 
We applied the EOF as a statistical method to quan- 
titatively examine oceanographic parameters. EOF 
analysis is a useful technique for decomposition of a 
time series of geophysical data into temporal and spa- 
tial variability in terms of orthogonal functions or sta- 
tistical modes. EOF analysis has been used commonly 
to describe spatiotemporal ocean variability (Yoder and 
Kennely, 2003; Otero and Siegel, 2004; Radiarta and 
Saitoh, 2008; Tolan and Fisher, 2009). 
EOF analysis was applied to the raw weekly data 
set of SSHA and SST and monthly data of chiorophyll- 
a concentrations because of a lack of data in many of 
the weekly images. Here, we examined only the first 
and second dominant modes, which were statistically 
independent and significant. A more comprehensive ex- 
planation of the concept of EOF analysis has been pro- 
vided by Bjornsson and Venegas (1997). We constructed 
the EOF analysis using Matlab, vers. 7.1, software (The 
MathWorks, Inc., Natick, MA). On the basis of the re- 
sults of the EOF, we generated maps of patterns of spa- 
tial and temporal ocean variability in the EIO off Java 
with ArcGIS tools (Esri, Redlands, CA). 
Generalized additive model 
The GAM was first proposed by Hastie and Tibshirani 
(1990). The advantage of this model is that the predic- 
tor variables have nonlinear effects upon the response 
variable. We applied the binomial GAM to analyze 
positive catches, interpreted as presence (1), and null 
catches, interpreted as absence (0) (Fraile et al., 2010), 
of Bigeye Tuna and to determine the catch probability 
of Bigeye Tuna in the EIO off Java. The Bigeye Tuna 
catch data presented here are based only on the grids 
fished because, as mentioned previously, catch infor- 
mation for Bigeye Tuna was not available for unfished 
grids; therefore, we ignored the unfished strata (Wal- 
ters, 2003). Some sampling bias may have affected our 
results regarding catch variability. However, the catch 
data are nevertheless useful for examination of Big- 
eye Tuna variability in the study area, and the HR 
data based on those catch data may be indicative of 
relative changes in availability. We analyzed the pres- 
ence of Bigeye Tuna through exploration of the spatial 
trends in their catch distribution that were influenced 
by SSHA, SST, and chlorophyll-a concentrations. All 
explanatory model terms were treated as continuous 
variables and the spline smoothers were fitted initially 
to each term in the model (Zuur et al., 2009). A step- 
wise GAM was performed to determine the best-fitting 
model before application of the final GAM to the entire 
data set. Akaike’s information criteria (AIC) were used 
to determine the optimal set of explanatory variables. 
The model with the smallest AIC can be selected as the 
optimal model. GAMs were constructed in R software 
(vers. 2.14.0; R Development Core Team, 2011) with the 
gam function of the mgcv package (Wood, 2006). The 
GAMs were fitted in the form 
g(u t ) = a 0 + SjUjj) + s2(x 2i ) + s3(x 3i ) + s n 0e ni ), (2) 
where g = the link function; 
u t = the expected value of the dependent variable 
(Bigeye Tuna catch); 
a 0 = the model constant; and 
