Comparison of model types for prediction of seafloor trawlability in the Gulf of Alaska by using multibeam sonar data


Sarah C. Stienessen, Christopher N. Rooper, Thomas C. Webe, Darin T. Jones, Jodi L. Pirtle, and Christopher D. Wilson
Cover date: 
Published online 13 September 2021

Many rockfishes (Sebastes spp.) inhabit rugged areas of seafloor that are inaccessible to survey trawl gear. Their utilization of such habitat makes estimation of their abundance difficult. Furthermore, it is often difficult to assess whether habitat is trawlable or untrawlable and to estimate the spatial extent of both habitat types. To help determine trawlability for the continental shelf in the Gulf of Alaska, we used multibeam sonar data collected in the area during 2011, 2013, and 2015. These data were used to derive 3 characteristics of the seafloor: oblique incidence backscatter strength (Sb oblique), seafloor ruggedness, and bathymetric position index. Habitat type was categorized as trawlable or untrawlable through analysis of video from deployed drift cameras. We tested the effectiveness of the use of these seafloor characteristics in prediction of habitat trawlability with 4 types of models: generalized linear model, generalized additive model, boosted regression tree, and random forest. All 4 models perform moderately well at predicting trawlability across the shelf, and results from all of them indicate that Sb oblique is the most important characteristic in discriminating between trawlable and untrawlable habitat. These results indicate that multibeam sonar data can help determine habitat type, information that in turn can help improve habitat-specific estimates of biomass of marine fish species.