Stienessen et al.: Comparison of model types for prediction of seafloor trawlability in the Gulf of Alaska 185 
rockfish populations that cannot be successfully sampled 
with trawl nets (Williams et al., 2010; Jones et al., 2012; 
Rooper et al., 2012; Jones et al., 2021). 
During the AFSC bottom-trawl surveys, trawl tows are 
conducted at stations selected by using a stratified ran- 
dom sampling design (von Szalay and Raring, 2018). His- 
torically, 25-km? grid cells have been designated as either 
trawlable or untrawlable; however, the initial classification 
of a grid cell is often uncertain. The trawlablilty of the sea- 
floor in grid cells has been determined by a vessel skipper 
searching for a minimum of 2 h to locate trawlable ground 
with a multibeam echo sounder. Realistically, knowledge of 
seafloor trawability is not complete across each grid cell, 
and most grid cells include both trawlable and untrawlable 
bottom types. For an entire grid cell to be designated as 
trawlable, a trawlable path of only 1.5 km along the sea- 
floor is necessary (1.5 km is the trawl length of a standard 
15-min bottom-trawl tow). Trawling in an untrawlable 
area within a grid cell classified as trawlable can result in 
substantial gear damage, lost survey time, habitat damage, 
and the cell may end up being reclassified as untrawlable. 
More accurate knowledge of the actual extent of trawlable 
and untrawlable areas within the grid cells of a bottom- 
trawl survey can help to improve biomass estimates and 
survey productivity. 
Research efforts have used the combination of acoustic 
backscatter data and analysis of underwater camera video 
to improve estimates of availability of rockfish species 
in the GOA, in part by using video images to determine 
the extent of seafloor trawlability (e.g., Jones et al., 2012; 
Rooper et al., 2012; Jones et al., 2021). These studies are 
limited by the time it takes to deploy a camera and the 
area that can be covered by a single camera. Multibeam 
sonar systems collect high-resolution acoustic bathymetry 
and backscatter data. These data can be used to generate 
comprehensive images of the seafloor, describe features of 
seafloor morphology, and discriminate among substrate 
types (e.g., Jagielo et al., 2003; Goff et al., 2004; Wilson 
et al., 2007; Brown and Blondel, 2009; Weber et al., 2013). 
If multibeam sonar data can be used to successfully cat- 
egorize seafloor trawlability, large areas of seafloor can 
be classified as trawlable or untrawlable quickly and 
efficiently. 
Researchers with the AFSC and the Center for Coastal 
and Ocean Mapping at the University of New Hampshire 
began collaborating in 2008 to optimize standard oper- 
ating procedures for seafloor mapping with a Simrad 
ME70! multibeam echo sounder (Kongsberg Maritime AS, 
Kongsberg, Norway). The Simrad ME70 was designed spe- 
cifically for fishery research applications as a calibrated, 
user-configurable multibeam sonar system intended to 
collect quantitative data on acoustic targets in the water 
column (Trenkel et al., 2008; Stienessen et al., 2019). How- 
ever, the Simrad ME70 can also be used to simultaneously 
collect bathymetry and seafloor backscatter data (Cutter 
1 Mention of trade names or commercial companies is for identi- 
fication purposes only and does not imply endorsement by the 
National Marine Fisheries Service, NOAA. 
et al., 2010). Customized software can be used to extract 
bottom detections that characterize the seafloor from data 
collected with a Simrad ME70 (Weber et al., 2013). 
A case study at Snakehead Bank, off Kodiak Island in 
Alaska, used seafloor characteristics extracted from the 
multibeam data collected with a Simrad ME70 in tandem 
with analysis of video images to distinguish trawlable from 
untrawlable habitat in the GOA (Weber et al., 2013). Ina 
subsequent study, more areas in the GOA were included, 
several characteristics of benthic terrain derived from 
multibeam sonar data at various scales were considered, 
and optical technology was used to validate seafloor clas- 
sification derived from mutlibeam sonar data (Pirtle et al., 
2015). The authors of the latter study reported that the 2 
best generalized linear models (GLMs) described 54% of 
the variation between trawlable and untrawlable seafloor 
types. These quantitative models were developed by using 
data collected with a Simrad ME70, at stations where 
video data were also recorded, and combined either oblique 
incidence backscatter strength (S,, oblique) or mosaic sea- 
floor backscatter strength with vector ruggedness measure 
(VRM) and bathymetric position index (BPI) to predict sea- 
floor trawlability. 
Each of the predictors identified by Pirtle et al. (2015) 
describe a component of seafloor morphology. Backscatter 
strength is dependent on incidence angle of the acoustic 
signal with the seafloor. At normal incidence, there is not 
much difference in S;, between strong scatterers (e.g., hard 
rock and boulders) and weak ones (e.g., fine sand). How- 
ever, S;, is higher from strong scatterers than from weak 
scatterers when the incidence angle is oblique (Jackson 
and Richardson, 2007; Lamarche et al., 2011; Weber et al., 
2013). Values of S,, oblique can therefore help differenti- 
ate between substrate types. Vector ruggedness measure 
is a measure of seafloor rugosity (i.e., seafloor complexity), 
and values of VRM account for variability in both seafloor 
slope and aspect. This predictor describes the variation in 
terrain and should differentiate smooth and rugged sub- 
strates (Sappington et al., 2007; Grohmann et al., 2011). 
Finally, BPI is a measure of elevation relative to surround- 
ing locations (Guisan et al., 1999). Values of BPI can be used 
to highlight topographic features, such as seafloor valleys 
or knolls, which are shallower or deeper than neighboring 
areas. Prior to our study, it was unclear whether the model 
that includes these 3 predictors, or seafloor characteris- 
tics, could be used to effectively classify seafloor trawlabil- 
ity outside of the study area utilized by Pirtle et al. (2015) 
or whether application of this model to other areas in the 
GOA would result in high misclassification. 
The primary objective of our study was to use a com- 
bination of seafloor characteristics derived from multi- 
beam sonar data and from analysis of underwater video 
images to determine whether the GLM that combines S,, 
oblique values with VRM and BPI values can continue 
to correctly classify seafloor trawlability over a larger 
extent of the GOA than the area classified by Pirtle et al. 
(2015). Secondarily, we wanted to determine whether 
nonlinear models are better at predicting seafloor traw- 
lability; therefore, we tested the effectiveness of 3 other 
