Hanselman et al : Applications in adaptive cluster sampling of Gulf of Alaska rockfish 



507 



shows that the mean of ACS is stable because it 

 changes little by removing a high catch, whereas 

 the SRS mean is reduced by half. 



SR-RE results 



At every third POP random tow, a tow was made in 

 the SR-RE depth stratum. A total of 35 tows were 

 made in the SR-RE stratum. Nine random tows 

 jrielded five distinct networks with 21 network tows 

 and five edge units. The stopping rule was invoked 

 for three of the five networks. 



At the mean CPUE criterion (418 kg/km, alt. 3), 

 the adaptive estimators performed approximately 

 the same in terms of SE compared to the SRS esti- 

 mator using 71 (Table 2). With v', the SRS estimator 

 yielded a lower SE than both adaptive estimators. 

 When the criterion value increased to an arbitrarily 

 higher value (540 kg/km), the adaptive estimators 

 performed worse than SRS estimates for both /; 

 and v'. 



Time efficiency 



We recorded and compared travel time between 

 adaptive tows and simple random tows for 149 

 of the tows (Table 31. Not all the tows were used 

 because of mechanical failure or because the factory 

 capacity was reached. In the survey, 38 hours out 

 of 10 days were spent in transit between sampling tows, 

 which for a short survey was a substantial amount of the 

 available time. For POP, substantial gains in travel-time 

 efficiency were achieved with ACS. Average travel time 

 for simple random tows (0.45 h) was nearly triple that of 

 adaptive tows (0.16 h) for POP, which indicated that ACS 

 can maximize sampling tows for POP when time is limited. 

 In the SR-RE sampling, travel time for adaptive sampling 

 (0.5 h) was about the same as simple random sampling 

 (0.49 h), which was due to long linear samples that are not 

 as close together as POP tows (Fig. 1). Also, determina- 



500 1000 1500 2000 



Mean abundance 



2500 



3000 



Figure 3 



Bootstrap distributions for the 1999 adaptive sampling survey 

 (25,000 replicates). Dotted line is the sampling estimate of mean 

 abundance (kg/km) from the survey. Top graph is the distribution 

 of mean abundance estimates for simple random sampling. Bottom 

 graph is the distribution of mean abundance estimates for adaptive 

 cluster sampling (obtained with the Hansen-Hurwitz estimator). 



tion of CPUE required processing of the catch, which took 

 various amounts of time after the completion of the tow. 

 Because of this delay, we went to the opposite tow on the 

 other side of the random tow when sampling SR-RE with 

 the linear pattern, whereas there were many nearby tows 

 when sampling POP with the cross pattern. 



The travel time was added to the average tow time irom 

 gear deployment to full retrieval of 0.5 h for POP and 1.0 h 

 for SR-RE to obtain total sampling time (per sample). 

 Travel time was reduced by 31% with adaptive sampling 

 (0.66 h/sample) in relation to simple random sampling 



