496 
National Marine 
Fisheries Service 
NOAA 
Fishery Bulletin 
fb- established in 1881 ^ 
Spencer F. Baird 
First U.S. Commissioner 
of Fisheries and founder 
of Fishery Bulletin 
A method for predicting trawlability in the Gulf 
of Alaska with the use of calibrated, split-beam, 
echosounder backscatter 
Paul G. von Sialay (contact author) 
David A, Somerton 
Email address for contact author: paul.von.szaIay@noaa.gov 
Resource Assessment and Conservation Engineering Division 
Alaska Fisheries Science Center 
National Marine Fisheries Sen/ice, NOAA 
7600 Sand Point Way NE 
Seattle, Washington 98115 
Abstract- —-We examined the feasibil¬ 
ity of distinguishing trawlable from 
untrawlable bottom using acoustic 
backscatter data from a calibrated 
single-beam echosounder to better 
define and map continental shelf ar¬ 
eas of the Gulf of Alaska (GOA) that 
are too rough and rocky to be sam¬ 
pled by the National Marine Fisher¬ 
ies Service’s biennial bottom trawl 
groundfish survey. Bottom classifi¬ 
cation algorithms were applied to 
backscatter data collected from ar¬ 
eas of known trawlability to provide 
9 metrics of bottom type from small 
sections of bottom (~50 records with¬ 
in a 15-min trawl tow). Prediction 
models, based on both generalized 
additive models (GAMs) and gener¬ 
alized linear models (GLMs), were 
developed to relate the bottom type 
metrics to the known state of traw¬ 
lability. The models were then tested 
to judge their performance on new 
data by using 33% cross validation. 
Although the best GAM had a high¬ 
er correct prediction rate (82.4%) 
than the best GLM (76.9%), under 
cross validation both models had 
nearly the same correct prediction 
rate (75.0%). This result is a suffi¬ 
ciently high prediction rate to allow 
the development of better trawlabil¬ 
ity maps by applying the model to 
data collected along acoustic track 
lines during the GOA bottom trawl 
surveys. 
Manuscript submitted 16 December 2016. 
Manuscript accepted 8 August 2017. 
Fish. Bull. 115:496-503 (2017). 
Online publication date: 16 August 2017. 
doi: 10.7755/FB. 115.4.6 
The views and opinions expressed or 
implied in this article are those of the 
author (or authors) and do not necessarily 
reflect the position of the National 
Marine Fisheries Service, NOAA. 
The Alaska Fisheries Science Cen¬ 
ter of the National Marine Fisher¬ 
ies Service has conducted a bottom 
trawl survey in the Gulf of Alaska 
(GOA) biennially since 1999 to as¬ 
sess the distribution and abundance 
of groundfish for fisheries manage¬ 
ment (von Szalay and Raring, 2016). 
The survey area, which consists of 59 
strata based on depth, benthic habi¬ 
tat, and management areas, spans 
the continental shelf and upper con¬ 
tinental slope from the Islands of 
Four Mountains eastward to Dixon 
Entrance, and from nearshore wa¬ 
ters to a depth of 1000 m (Fig. 1). Al¬ 
though the purpose for the survey is 
to randomly sample this area under 
the assumption that the entire area 
is trawlable (can be sampled with 
the Poly-Nor’eastern 4-seam survey 
trawl), in practice, this concept is an 
approximation because the GOA is a 
mosaic of habitat types. Some habi¬ 
tats are untrawlable; that is, they are 
too rocky, rugged, or steep to allow a 
fully random choice of sampling loca¬ 
tions when using the standard sur¬ 
vey trawl. Furthermore, the GOA has 
never been mapped with sufficient 
spatial resolution to permit identi¬ 
fication of bottom types that would 
preclude successful trawling. Conse¬ 
quently, the locations of untrawlable 
bottom (in the varying opinions of 
experienced GOA survey vessel cap¬ 
tains) and even the proportion of the 
area comprising such habitat is not 
known. The relative abundance of 
each species is therefore currently 
estimated by extrapolating the mean 
abundance in trawlable areas to 
the entire survey area. A potential 
problem with this approach is that 
it may result in biased estimates of 
abundance because fish density is a 
function of habitat type, which is cor¬ 
related with trawlability (Yoklavich 
et a!., 2000; Pirtle et ah, 2015). The 
bias is likely positive for flatfish and 
other fish species, which prefer rela¬ 
tively smooth and sandy bottoms in 
trawlable habitats (MeConnaughey 
and Smith, 2000; Busby et ah, 2005), 
and negative for roekfish (Sebastes 
spp.) and other fish species, which 
prefer rough, rocky bottoms in un¬ 
trawlable habitats (Richards, 1986; 
Stein et al., 1992; Clausen and Heif¬ 
etz, 2002; Jagielo et ah, 2003; Zim- 
mermann, 2003; Jones et ah, 2012). 
An alternative survey design now 
being considered by the Alaska Fish¬ 
eries Science Center for the GOA 
is one in which the abundance es¬ 
timates derived from survey trawl 
catches, based on catch weight per 
unit of swept area, would be extrapo- 
