Weber et al: Seabed classification for trawlability determined with a multibeam echo sounder 
73 
A 
B 
20 : 
Figure 3 
The frequencies of occurrence for major and minor substrate combinations, classified from the data collected in 2009 and 
2010 with a stereo drop camera and a remotely operated vehicle as a function of different seafloor characteristics derived 
from the data collected with a Simrad ME70 multibeam echo sounder. Major (capital letter) and minor (lowercase letter) 
substrate types included Bb=boulder; C=cobble; Gg=gravel; and Ss=sand. 
values shown in Figure 1, these 2 regions would cor- 
respond to sandy gravel and very fine silt, respectively. 
The lower set of oblique-incidence S h values were found 
in the deepwater off the northern side of the bank at 
depths of 200-250 m and also on the south side of the 
bank at depths of 120-150 m. On average, the larg- 
est S fc -slope and the widest range of normal-incidence 
S b were observed on sandy substrate. The normal-in- 
cidence S b for areas classified as sandy substrate ex- 
tended to ranges higher than would be expected, a find- 
ing that could be a result of unusually high volume- 
backscatter caused by gas or heterogeneities within the 
sediment volume. The harder substrates (Bb and Cb) 
all had small S 6 -slope, as expected, and on average had 
higher SI than the sandy sediments. 
To determine how each parameter discriminated 
between trawlable or untrawlable seafloor, using clas- 
sified SDC and ROV video data as verification, the 
frequencies of occurrence for each parameter were ex- 
tracted for each substrate type (Fig. 4). T-tests indicat- 
ed that the distributions of trawlable and untrawlable 
areas of seafloor were distinguishable at the oc=0.05 
significance level (Table 1), although each parameter 
did not perform equally when discriminating between 
the 2 classifications. The 3 best individual discrimina- 
tors were the normal-incidence S b , S b - slope, and the 
oblique-incidence S b with standard differences of 0.74, 
1.12, and 1.89, respectively. Of these 3 parameters, the 
oblique-incidence S b demonstrated the clearest separa- 
tion between trawlable and untrawlable seafloor, with 
a boundary at -13.4 dB. According to modeled data 
(Fig. 1), this S b level discriminates cobble and rock 
from gravel, sand, and silt. The SI and rugosity were 
separated less well with standard differences of 0.25 
for each. 
With the oblique-incidence S b considered alone, the 
combined error rate (erroneous classifications of both 
trawlable and untrawlable seafloor) reached a mini- 
mum of 5.6% (n=303) with a boundary set at S 6 =-13.4 
dB. To determine whether this error rate could be 
lowered, additional parameters derived from the data 
collected with the Simrad ME70 were linearly com- 
bined with the oblique-incidence S b . Figure 5 shows 
the combination of the oblique- incidence S b with each 
of these other parameters, along with a line that best 
discriminated between the trawlable and untrawlable 
classifications. The largest reduction in classification 
error rate was achieved when the oblique-incidence S b 
was combined with either the normal-incidence S b or 
the SI, both of which had a marginally improved er- 
ror rate of 5.0%. When 3 parameters were combined to 
discriminate between trawlable and untrawlable sea- 
