72 
Fishery Bulletin 1 1 1 (1) 
boulder (Cb), with the secondary substrate indicated 
by the lower-case letter. Because the video collected 
with the SDC and ROV provided a continuous display 
of substrata, the substrate-type code was changed only 
if a substrate type encompassed more than 10 consecu- 
tive seconds of video. 
For this study, the substrate observed in the under- 
water video transects was further classified as either 
untrawlable or trawlable with reference to the stan- 
dard Poly-Nor’eastern 4-seam bottom trawl used in 
biennial bottom-trawl surveys of the Gulf of Alaska 
and Aleutian Islands by the Alaska Fisheries Science 
Center (Stauffer, 2004). The Poly-Nor’eastern bottom- 
trawl footrope comprised 10-cm disks interspersed 
with bobbins 36 cm in diameter. The untrawlable ar- 
eas were defined as any substrate containing boulders 
that reached >20 cm off the bottom of the seafloor or 
any substrate with exposed bedrock that was so rough 
that the standard bottom-trawl footrope would not eas- 
ily pass over it. Therefore, the trawlable grounds were 
those areas mostly composed of small cobble, gravel, 
sand, and mud without interspersed boulders or jagged 
rocks. The untrawlable grounds were those areas that 
contained any boulder or high-relief rock substrates. 
The same experienced observer classified the substrate 
for both the ROV and SDC video transects. 
The video data thus classified were partitioned in 
a grid of 25-m squares, or cells — a length scale that 
is a rough estimate for the accuracy of the position- 
ing systems associated with both video systems. The 
primary and secondary substrate types were given a 
numeric value based on a nominal substrate size, and 
each grid cell was assigned substrate types associated 
with the median values for all data within the cell 
boundaries. Grid cells also were assigned as trawlable 
or untrawlable if all data within a cell supported such 
a classification; otherwise, the grid cell was assigned a 
“mixed” classification. The gridded video classifications 
were then compared with the seafloor parameters (e.g., 
rugosity or normal-incidence S b ) derived from data col- 
lected with the Simrad ME70, where both types of data 
existed at the same position, to provide an indication of 
how each acoustically derived seafloor parameter was 
able to discriminate between trawlable and untraw- 
lable areas. This comparison was done for each param- 
eter separately and then done for various combinations 
of parameters to find a combination of parameters that 
best discriminated between trawlable and untrawlable 
substrate. For each parameter, a f-test was used to de- 
termine whether it was able to distinguish between 
trawlable and untrawlable seafloor at the significance 
level of a=0.05 (i.e. , where erroneous rejection of the 
null hypothesis is expected 5% of the time), and val- 
ues of standard difference (the difference between the 
sample means divided by the pooled standard devia- 
tion) were computed. When combinations of parameters 
were tested, a best-fit separation (for the goal of mini- 
mizing the classification error rate) within the multidi- 
mensional parameter space was found through exami- 
nation of the entire parameter space. To maintain a 
clear link back to the underlying data distribution, the 
separation between trawlable and untrawlable was as- 
sumed to be a line, plane, or hyperplane (a generaliza- 
tion of a plane into more than 2 dimensions), depend- 
ing on the dimension of the parameter space. 
Results 
The data showed a wide range of values and, presum- 
ably, associated substrate types. The shallowest (<100- 
m) portion of Snakehead Bank contained the highest 
oblique-incidence S b (approximately -12 dB). This re- 
gion contained similar values for the normal-incidence 
S b , and small S 6 -slope (<0.75 dB/°). Taken together, 
these data indicate a cobble seafloor on the top of the 
bank. On the northeastern side of the bank at depths 
-200 m, the oblique-incidence S h reached its lowest 
value of approximately -30 dB with a normal-incidence 
S h of -15 dB and S 6 -slope of -1.1 dB/° — values consis- 
tent with a substrate composed of very fine silt. 
The region with the highest normal-incidence S b 
(-10 to -7 dB) occurred between 154°W and 153. 9°W 
and near 56.07°N in the northwest region of the bank. 
The S 6 -slope was also high in this region, reaching up 
to 1.5 dB/°, and the oblique-incidence S b was between 
-18 dB and -15 dB. These results for the seafloor pa- 
rameters are confounding, given that the S 6 -slope was 
large enough to indicate a fine sand or silt, but the 
normal-incidence and oblique-incidence S b both indi- 
cated a coarser sediment or a higher-than-anticipated 
volume scatter contribution due to heterogeneities or 
gas (Jones et ah, 2012) within the sediment. 
The SI shows a complicated pattern that did not 
appear to be well correlated with any certain sub- 
strate type, although there were large (hundreds of 
meters) contiguous regions that exhibited high SI val- 
ues (i.e., the data did not appear to be simply random 
noise). The rugosity levels show the bank to be rela- 
tively smooth along the top, except at a sharp transi- 
tion along its northeastern edge between the 100- and 
150-m contours. The rugosity analysis also indicates 
the appearance of what may be large (wavelength 
-150 m) sand waves in the extreme southeastern por- 
tion of the study area and smaller pockmarks in the 
southwestern portion of the study area. 
The results of a comparison of the seafloor param- 
eters derived from the backscatter data that was col- 
lected with the Simrad ME70 and the substrate types 
derived from the data collected with the SDC and ROV 
are shown in Figure 3. These data show that, although 
substrate types Bb, Cb, and Gb are difficult to distin- 
guish with backscatter parameters, these 3 types are 
clearly separate from substrate type Ss. The oblique- 
incidence S b values for substrate type Ss appeared to 
be bimodal, with the majority of the values residing be- 
tween -17 and -15 dB and a substantial number of val- 
ues between -29 and -26 dB. According to the notional 
