742 



Fishery Bulletin 93(4), 1995 



34°N- 

 124°W 



Figure 1 



Flight transects and defined areas for aerial surveys for harbor porpoise, Phocoena phocoena, in (A) 

 central California (transects 1-26; areas 1 and 2) and IB) northern California (transects 41-57; area 3). 

 Transect 7 was combined with transect 8 after 1986 and is not shown. 



(700 ft) altitude. Flights were conducted only when 

 weather conditions were good (Beaufort sea states 

 0-3, mostly with clear or partly cloudy skies). Sight- 

 ing information and environmental conditions were 

 recorded and updated throughout the survey by us- 

 ing a laptop computer connected to the aircraft's LO- 

 RAN navigation system. 



Analytical methods 



The number of porpoise observed per kilometer of 

 search effort was used as a measure of relative abun- 

 dance. These data were stratified by Beaufort sea 

 state (0-1, 2, and 3), area (transects 1-14 and 15-26 

 in central California, and 41-57 in northern Califor- 

 nia; see Fig. 1), and percent cloud cover (<25%; >25%). 

 After log-transformation, a stepwise selection pro- 

 cedure was used to construct an analysis of covari- 

 ance model of the form: 



P = n + a 1 + a 2 +... + 8(y-y) + £, (1) 



where P is the log-transformed value of the number 

 of porpoise seen per kilometer + 0.001; /u is the mean 

 value of P; the a's are factors influencing apparent 

 porpoise abundance (such as sea state); S is the coef- 

 ficient for the covariate year (y); y is the mean year; 



and e is a random error term. This additive model 

 for the log-transformed data is equivalent to a mul- 

 tiplicative model for the actual data (stratification 

 variables such as sea state are expected to change 

 the fraction of animals observed, and thus have a 

 multiplicative effect). Variability caused by unequal 

 survey coverage in each combination of sea state, 

 percent cloud cover, and geographic area was in- 

 cluded in the model by weighting by the number of 

 kilometers flown. The analysis was done separately 

 for central California alone (transects 1-26) and for 

 both central and northern California (transects 1— 

 26 and 41-57). Previous simulations (Forney et al., 

 1991) indicated that power would still be low with 

 this eight-year time series; therefore, the critical 

 value for type-I error was set at a - 0.10. This was 

 expected to provide a power of approximately 60% to 

 detect a large change in abundance of ±10% per year 

 but would still have low power (approximately 25%) 

 to detect trends on the order of ±5% per year. 



Results 



A summary of survey coverage (total kilometers sur- 

 veyed, percent surveyed under good conditions) and 



