Stienessen et al.: Comparison of model types for prediction of seafloor trawlability in the Gulf of Alaska 191 
@ Trawlable 
» Untrawlable 
VRM (2, m) VRM (m, cm) 
Figure 3 
Mean values of seafloor characteristics derived from multibeam sonar data and used in models to 
classify habitat utilized by rockfishes (Sebastes spp.) as trawlable (black circles) or untrawlable 
(gray circles). Data were collected during fine-scale multibeam surveys conducted in 2011, 2013, 
and 2015 in the Gulf of Alaska. Plots show the effect of trawlability on 3 seafloor characteristics, 
(A) oblique incidence backscatter strength (S;, oblique), (B) vector ruggedness measure (VRM), 
and (C) bathymetric position index (BPI), over all years combined and (D) show the effect of units 
on VRM in 2011. In this study, VRM was calculated by using meters for coordinates in the Univer- 
sal Transverse Mercator system and depth in centimeters (right half of the plot in panel D), but 
Pirtle et al. (2015) used degrees for coordinates in a geographic coordinate system and depth in 
meters (left half of the plot). Error bars indicate 95% confidence intervals. 
Table 1 
Comparison of results from generalized linear models used to predict seafloor trawlability of 
habitat utilized by rockfishes (Sebastes spp.) at locations of camera stations surveyed during 
2011, 2013, and 2015 in the Gulf of Alaska. Predictor variables are seafloor characteristics 
derived from multibeam sonar data: oblique incidence backscatter strength (S,, oblique), vector 
ruggedness measure (VRM), and bathymetric position index (BPI). Values of Akaike informa- 
tion criterion (AIC), deviance explained (D”), area under the receiver operating curve for the 
training data set (training AUC), and area under the receiver operating curve for the test data 
set containing out-of-sample data (test AUC) are provided for each model. 
Model Training Test 
Year range Model AIC D? significance AUC AUC 
2011-2015 S;, oblique 151.5 0.17 2.34 x 10° 0.77 0.75 
2011-2015 S;, oblique + VRM + BPI 153.4 0.19 2.78 x 10°77 0.81 0.73 
models with 2 different variables (e.g., models with VRM and GAM having higher test AUC scores (0.75 and 0.70, 
or BPI), the randomness of the RF becomes more respectively) than the simplified BRT model and RF 
apparent. (0.65 and 0.63, respectively) (Tables 1-3). However, both 
Model fits were reasonable for each best model (test the best GAM and simplified BRT accounted for more D® 
AUC values range from 0.63 to 0.75), with the best GLM (0.31; Tables 2-3) than the best GLM (0.17; Table 1). 
