Forney et al.: Aerial surveys of Phocoena phocoena abundance trends 



371 



Results 



Survey results 



A total of 16,948km of survey effort during 1986-90 

 resulted in 431 sightings, representing a total of 796 

 harbor porpoise. The overall mean number of porpoise 

 per kilometer was 0.047. The average values of por- 

 poise per kilometer over the five years are listed in 

 Table 1 for different sighting conditions and areas. 

 Mean group size was 1.85 porpoise, with a range of 

 1-10 and median 1. 



The model 



The following model provided the best fit to the ob- 

 served logarithmic estimates of porpoise per kilometer: 



P = 



^ + «Bi + «Cj + <*Ak + £i,j,k 



(2) 



where P = log e of [(porpoise/km) + 0.001], 



\i = mean value of P, 



a Bi = effect of Beaufort sea state i on P, 



a C j = effect of cloud cover j on P, 



»Ak = effect of area k on P, 



tjj^k = normally distributed error with a mean 

 of zero. 



The model selection procedure is outlined in Table 

 2. None of the included variables lost significance and 

 subsequently had to be dropped after inclusion of other 

 variables. The yearly trend was not significant, so the 

 model is essentially reduced to an analysis of variance 

 (ANOVA) model. The results of the complete ANCOVA 

 model testing for a yearly trend in the 1986-90 har- 

 bor porpoise data are shown in Table 3. The effects of 

 area, cloud cover, and Beaufort sea state were signifi- 

 cant (P< 0.0001, P<0.0006, andP<0.0021, respective- 

 ly), the yearly trend was not (P = 0.8535). None of the 

 interaction effects were significant. The parameter 



estimates for the models with and without year are 

 displayed in Table 4. 



Analysis of power to detect trends 



No trend simulations f<5 = 0) To determine the reli- 

 ability of the simulation procedure to model trends in 

 porpoise abundance, 500 simulated data sets with no 

 yearly trend were created for five, six, eight, and ten 

 survey years, using parameter set (B) in Table 4. The 

 simulated data sets were analyzed using the full 



