280 



Fishery Bulletin 105(2) 



an auxiliary net under the trawl net to capture those 

 fish escaping beneath the footrope — a method similar 

 to that used in the studies of Engas and God0 (1989a), 

 Walsh (1992), and Munro and Somerton (2002). Bridle 

 efficiency (kj and the herding coefficient (kj were es- 

 timated from the data obtained from a herding experi- 

 ment which consisted of repeatedly conducting trawl 

 hauls where W^ was varied by varying the length of the 

 bridles as has been done in the studies of Engas and 

 God0 (1989b), Ramm and Xiao (1995), and Somerton 

 and Munro (2001). The width of the area contacted by 

 the bridle was estimated from the data obtained from 

 an experiment by using bottom contact sensors to mea- 

 sure the off-bottom distance along the lower bridle — a 

 distance that was reported for the Poly Nor'eastern 

 trawl in Somerton (2003). 



Net efficiency experiment 



The net efficiency experiment was conducted during 

 2-10 July 1996 in the Gulf of Alaska, off the southeast 

 side of Kodiak Island (58°30'N, 149°30'W) at 135-151 

 m depth with a 45-m chartered stern trawler, the FV 

 Golden Dawn. An auxiliary net, described in the Appen- 

 dix and patterned after those described in Engas and 

 God0 (1989b) and Walsh (1992), was attached under the 

 trawl net (Fig. 1). Trawling procedures followed normal 

 survey protocols that included towing only during day- 

 light hours for 15 minutes at a vessel speed of 1.5 m/sec. 

 Catches from the trawl and the auxiliary net were kept 

 separate, sorted by species, weighed, and all individuals 

 were measured for total length (TL) in centimeters. 



The auxiliary net, which was constructed of smaller 

 10.2-cm stretch mesh polyethylene netting, had a 24.8- 

 m long headrope that was lashed directly to the fishing 

 line of the trawl net (i.e., the forward edge of the net- 

 ting) excluding the wing extension sections of the trawl 

 footrope. The auxiliary net also had a 28.0-m long, 

 1.3-cm diameter chain footrope, strung with 12.7-cm 

 rubber disks (Fig. 1), that was attached to the trawl 

 footrope at the junctions of the roller gear and wing 

 extensions, thus allowing the auxiliary footrope to move 

 independently of the trawl footrope and to follow it with 

 a separation distance of approximately 1-2 m at its cen- 

 ter. The auxiliary net was designed so that the common 

 intermediate section was attached to three separate 

 codends, each having a 3.2-cm stretch mesh liner. 



Experimental tows to verify that trawl performance 

 was not altered by the attachment of the auxiliary net 

 preceded the tows used for the measurement of net ef- 

 ficiency. During all of the experimental tows, a video 

 camera, supplied with a 50 W light, was positioned in 

 front of the footrope along the centerline of the trawl so 

 that it had an oblique view and allowed an approximate 

 measurement of the distance between the center of the 

 footrope center and the sea floor. Initially, 10 tows were 

 made with the trawl without the auxiliary net attached. 

 When the same procedure was used for the trawl with 

 the auxiliary net attached, the video recording indi- 

 cated that the off-bottom distance of the footrope at its 



center was greater. By trial and error, additional weight 

 (chain) was attached to the center of the footrope until 

 the off-bottom distance appeared equal to that of the 

 trawl footrope without the auxiliary net. A total of 53.4 

 kg of chain was attached across the centermost 6.1 m 

 of the footrope. 



Estimating /r^ from the experimental data 



Net efficiency, /;„, was estimated as a function of fish 

 length, /, by fitting an analytical model to the capture 

 probability, P (i.e., proportion of fish passing between 

 the trawl wing tips that are caught), and fish length 

 data pooled over tows. Four competing models, each 

 representing a different capture process, were considered 

 (Munro and Somerton, 2001). The first three are para- 

 metric models, which, in order of increasing complexity, 

 are expressed as 



2 -parameter logistic, P = 



1 + e 



Aa+0l\ 



[l + e-'"'^") 



3-parameter logistic, P = y 



4-parameter logistic, 

 „ ( 1 Vl-^ff e-"'""'"' 



(4) 



(5) 



-(a+/3/l 



(6) 



where a, /3, y, and 6 are free parameters to be estimated. 



The maximum likelihood procedure for fitting these 

 models, detailed in Munro and Somerton (2001), was 

 based on the assumption that the entry of individual fish 

 into either the trawl net or the auxiliary net could be 

 described as a binomial statistical process. The fourth 

 model is a nonparametric model, the cubic spline, which 

 was fitted by using an S-^• (S-PLUS, Insightful Corpora- 

 tion, Seattle, WA) function that determined the effective 

 number of parameters of the spline function with cross 

 validation (Venerables and Ripley, 1994). Of the four 

 competing models, the best fitting model was selected 

 as the one producing the lowest value of the Akaike 

 Information Criterion (AIC; Burnham and Anderson, 

 1998). Ninty-five percent confidence intervals about the 

 capture probabilities as a function of fish length were 

 estimated by using the bootstrapping method (Efron and 

 Tibshirani, 1993) where entire hauls were used as the 

 units of data resampled. 



Herding experiment 



The herding experiment was conducted 10-19 May 1998 

 aboard a 30.6-m stern trawler, the FV Hickory Wind, 

 near Kodiak Island in the Gulf of Alaska at depths rang- 

 ing from 126 to 183 m. A blocked sampling design was 



