334 
Fishery Bulletin 114(3) 
Table 2 
Thresholds for digital numbers (in pixels) for satellite 
images from the Operational Linescan System of the 
U.S. Defense Meteorological Satellite Program for the 
period 2005-2013. Thresholds were calculated from the 
histogram in Figure 2. Pixels within the class 5 range 
represent fishing vessel lights. Classes 1-4 represent 
reflected light from ocean water or light from cloud 
cover. 
Month 
Class 1 
Class 2 
Class 3 
Class 4 
Class 
August 
10 
17 
23 
30 
40 
September 
9 
16 
22 
28 
38 
October 
8 
14 
19 
27 
38 
November 
8 
13 
19 
28 
38 
December 
7 
12 
18 
27 
38 
Table 3 
Mean monthly sea-surface temperature (SST) values 
(°C) and standard deviations (SD), used to distinguish 
the light of vessels fishing for Pacific saury (Cololabis 
saira) from the lights of fishing fleets fishing for other 
fish. All lights occurring below the upper SST limit 
were categorized as locations of vessels targeting Pa- 
cific saury. 
Month 
Mean 
SD 
Upper SST limit 
August 
20.79 
2.69 
23.48 
September 
18.89 
2.47 
21.36 
October 
15.90 
2.58 
18.48 
November 
14.56 
2.88 
17.44 
December 
13.60 
2.89 
16.49 
produced and distributed by Archiving Validation and 
Interpretation of Satellite Oceanographic Data (AVISO; 
website) at a spatial resolution of 0.33°x0.33°. The sur- 
face geostrophic velocities were used to compute for EKE 
by using the following equation (Steele et al., 2010): 
EKE = 1/2 (u’2 + n’2), (2) 
where w’ and v’ = the zonal and meridional components 
of geostrophic currents, respectively. 
With the grid function of the software package Ge- 
neric Mapping Tools, vers. GMT 4.5.7 (website),we 
were able to calculate the monthly averages for each 
environmental variable from daily data sets, resampled 
to 1-km resolution and converted to Esri ASCII grid 
format(Esri, Redlands, CA) or to comma-separated val- 
ues (CSV) format, as required by the software program 
Maxent (website). 
Construction of a maximum entropy model 
To develop a model with a maximum entropy approach, 
we used the software program Maxent, vers. 3.3.3k. 
Phillips et al. (2006) provided detailed information on 
the mode of operating this software. We constructed 
models using default values for regulation parameter 
(1), maximum iteration (500), and automatic feature 
class selection. We used a cross-validation procedure 
to evaluate the performance of the models. For back- 
ground points, we generated pseudo-absences (10:1 
ratio of pseudo-absence to presence) following Barbet- 
Massin et al., (2012) on the basis of random spatial 
sampling within the study area (excluding points of 
presence of Pacific saury). We used the density.tools. 
Randoms ample command line in Maxent to generate 
the random pseudo-absences. For each monthly model, 
the data were randomly split into 2 categories: one 
category for training data (70%) and one for test data 
(30%). The test points were then used to calculate the 
area under the curve (AUG) of the receiver operating 
characteristic (ROC) (Phillips et al., 2006). 
Evaluation and validation of the model 
We used the AUG metric of the ROC curve to evaluate 
model fit (Elith et al., 2006; Phillips et al., 2006). The 
relative contribution of individual environmental vari- 
ables within the maximum entropy model was exam- 
ined by using the heuristic estimates of variable impor- 
tance based on the increase in the model gain, which is 
associated with each environmental factor and its cor- 
responding model feature. Response curves generated 
for each factor were examined to derive the favorable 
environmental ranges for potential fishing zones. 
Independent sets of monthly OLS data from 2011 
through 2013 were used to validate the models. The 
base models were used to create habitat suitability 
indices (HSIs) that assimilated similar environmental 
layers for the corresponding period from 2011 through 
2013. Spatial HSI maps were generated and over- 
lain with information on OLS data from the period 
2011-2013. 
Results 
Spatiotemporal distribution of fishing locations, and envi- 
ronmental data 
Figure 3 shows the variation in the distribution of 
fishing vessel lights from August through December 
during 2005-2013. Vessels started to appear off the Ku- 
ril Islands and east of Hokkaido in August (Fig. 3A). At 
the same time, fishing also occurred around the Sanriku 
coast of Japan and an offshore area between 150°E and 
41°N that extended northeast tol55°E and 43°N. 
During September (Fig. 3B), fishing vessels were 
distributed mostly north of 42°N, especially off east- 
