Stratification by echosounder signal to improve trawl survey precision for Pacific Ocean perch

Stratification by echosounder signal to improve trawl survey precision for Pacific Ocean perch

J.T. Fujioka, C.R. Lunsford, J. Heifetz, and D.M. Clausen

Stratification by echosounder signal to improve trawl survey precision for Pacific Ocean perchThis is part of Biology, Assessment, and Management of North Pacific Rockfishes
PDF    
To download the free PDF [2.5 MB], please enter:
-or-

Description

In 1998 echograms were recorded during individual bottom trawl hauls conducted in a study of Pacific ocean perch (Sebastes alutus) survey methods. Categorization criteria were developed on a subset of the echogram data based on signal patterns, shapes, and color. Blind tests were then conducted by individual scientists on a subset of hauls. Between-scientist agreement of high and low categories of rockfish abundance varied from 76% to 87%. Data were collected again in 1999 and categorized using the same criteria. Onboard scientists agreed with categorizations done by a shoreside scientist 65% of the time. The mean catch rate for hauls in the low echogram category was always lower than mean catch rate for the high category and statistically significant in all but one haul. Variance estimates that would result from simple double sampling using the echogram categorizations as strata definitions were predicted using the observed within-category variances at various levels of first stage (echosounder) sampling. Echosounder samples are considerably less expensive in time and cost than trawl sampling, and when 10 times as many echosounder samples were taken to stratify the trawl samples, variance improved 18-37% compared to simple random sampling, depending on the data set and the categorizer. If trawl hauls were allocated optimally, the improvement increased to 44-60% over simple random sampling. To match the variance obtained by double sampling with echosounder primary sampling, the number of random trawl hauls would have to be increased 1.8-2.5 fold depending on data set and categorizer.

Item details