514 



Fishery Bulletin 105(4) 



sampled randomly with replacement from all survey 

 years, and the sightings associated with those seg- 

 ments were used with step-wise model building to fit 

 the multiple-covariate model of f(0\Zj). For each of 1000 

 bootstrap iterations, a parametric bootstrap was used to 

 choose the values of ^(0) by drawing randomly from a 

 logit-transformed normal distribution with a mean and 

 variance selected to give the values of ^(0) and CW(g{0)) 

 used for abundance estimation. 



Probability of detection of a cetacean group 

 along the trackline 



The line-transect parameter giO) represents the prob- 

 ability of detecting a group that is located directly on 

 the transect line. This value is often assumed to be 1.0 

 in estimating abundance for dolphins that are found in 

 large groups (Gerrodette and Forcada, 2005); therefore, 

 in these analyses it was implicit that 100% of the groups 

 located on the trackline were detected. In our study, 

 for dolphins, porpoises, and large baleen whales, data 

 from the conditionally independent observer were used 

 to estimate the trackline detection probability for the 

 primary observer team, .^;(0), with the method developed 

 by Barlow (1995): 



gi(0) = 1.0- 



■f^m 



«„a-/"i(0) 



(4) 



where the subscript 1 refers to parameters for the pri- 

 mary observers, subscript 2 refers to parameters for the 

 conditionally independent observer, and n^^, = the number 

 of sightings within the truncation distance w used for 

 estimating the line-transect parameter fiO). 



Sightings by the primary team were included in esti- 

 mating rij and fjiO) only if a conditionally independent 

 observer was also on duty. This estimator (Eq. 4) is 

 positively biased (Barlow, 1995), which results in an 

 overestimation of giO) for the primary observers. Fully 

 independent observer methods (Buckland et al., 2004) 

 are generally superior to this conditionally independent 

 method (referred to as the "removal method" by Buck- 

 land et al. [2004]); however, such methods could not be 

 used because of the need to approach groups to deter- 

 mine species and estimate group sizes. The line-tran- 

 sect parameter f(0) was estimated independently for 

 the primary and independent observers with the soft- 

 ware program Distance 5.0 (available from Thomas^); 

 half-normal models were fitted with cosine adjust- 

 ments (Buckland et al., 2001), and the best-fit model 

 was selected by AIC^,. Because of sample size limita- 

 tions, species were pooled into three categories for 

 estimating g(0): 1) delphinids (excluding killer whales). 



2 Thomas, L. 2005. Research Unit for Wildlife Population 

 Assessment, University of St. Andrews, Scotland, UK. Web- 

 site: http://www.ruwpa.st-and.ac.uk/distance/ (accessed 26 

 June 2007). 



2) large whales (most baleen whales and killer whales), 

 and 3) Ball's porpoises. Killer whales were included 

 with large baleen whales because they are very con- 

 spicuous and are seen at greater distances than are 

 other delphinids (Barlow et al., 2001). Because track- 

 line detection probabilities may vary with the size of 

 the group (Barlow, 1995) or observation conditions, 

 the numbers of sightings made by primary and inde- 

 pendent observers were tested with Fisher's exact test 

 to determine if the proportion varied with group size 

 or Beaufort sea state. Delphinids and large whales 

 were stratified into large and small groups with cut- 

 points at 20 and 3 individuals, respectively. Because 

 of sample size limitations, a single detection function 

 was fitted to large and small groups of delphinids seen 

 by the independent observer. Estimates of g^(O) were 

 stratified if this test was significant for either factor. 

 Data for estimating ^(0) included transects covered on 

 the preplanned survey grid and during more oppor- 

 tunistic survey periods, such as transits from a port 

 to the starting point on the survey grid. Coefficients 

 of variation for giO) estimates from the conditionally 

 independent method were based on Equation 9 in Bar- 

 low (1995). 



The above conditionally independent observer method 

 for estimating ^(0) requires that all animals be avail- 

 able to be seen by the primary observer team. This ap- 

 proach does not work with long-diving species that may 

 be submerged for the entire time that the ship is within 

 visual range. Values of g(0) for sperm whales, dwarf 

 sperm whales, pygmy sperm whales, and all beaked 

 whales were taken from a model of their diving behav- 

 ior, detection distances, and the searching behavior of 

 observers (Barlow, 1999). 



Trackline detection probabilities for minke whales 

 posed a special problem. Insufficient sightings were 

 made to estimate giO) from the conditionally indepen- 

 dent observer method (only one conditionally indepen- 

 dent sighting was made) and insufficient information 

 exists on their diving behavior to use the modeling 

 approach. Here we assumed thatg(O) for minke whales 

 was similar to that for small groups of delphinids (but 

 see "Discussion" section). 



Results 



Surveys 



Survey effort in Beaufort sea states from to 5 covered 

 the study areas fairly uniformly (Fig. 1). Although not 

 all the planned transects were surveyed (because of 

 inclement weather and mechanical breakdowns), the 

 holes in the survey grid were small in relation to the 

 entire study area, and all areas appeared to be well rep- 

 resented. Survey effort in calm sea conditions (Beaufort 

 states 0-2) was not as uniformly distributed (Fig. 1) and 

 was particularly poor in the Oregon-Washington region. 

 Survey effort varied among years because of the avail- 

 ability of ship time. 



