Hobbs and Waite: Abundance of Phocoena phocoena in three Alaska regions 
255 
group size, observer, and survey year. Covariates were 
initially tested individually to identify functional forms 
or groupings that could reduce the number of parameters 
necessary to represent them. The discrete covariates 
(visibility, sea state, glare, group size, observer, and 
survey year) and the continuous covariate cloud cover 
(grouped into five categories: 0-20%, 20—40%, 40-60%, 
60-80%, 80-100%), were examined individually as cat- 
egorical factors. The coefficients from the categorical 
analysis were then charted against their hierarchical 
ranks. Where the coefficients appeared to fit a simple 
functional form of the hierarchical ranks (line, square 
root, natural logarithm, exponential) or could be grouped 
to reduce the number of parameters, the analysis was 
repeated with this alternative. The parameters for the 
function or grouping were estimated by using the regres- 
sion described above and compared to the result of 
the categorical factor by using Akaike’s information 
criterion (AIC). The function or grouping was used in 
the subsequent analysis if it improved the AIC. From 
this preliminary analysis, visibility and sea state were 
found to have a nearly linear effect and were treated 
as linear functions by using the hierarchical number 
as the value and by setting the best condition to one; 
group size was also considered to be linear. Cloud cover, 
glare, observer, and survey year were considered as cat- 
egorical data. Significant covariates were then combined 
in the GLM model and removed in a stepwise manner 
until the AIC had been minimized. The perception bias 
for each observer position and each transect segment 
was estimated from the final model and combined to 
estimate g{0) for each transect segment (see Appendix 
II for details). 
The program DISTANCE, vers. 3.5 (Thomas et al., 
1998) allowed only a global g(0) and thus did not ac- 
commodate g(0) to be estimated for each transect seg- 
ment from environmental and observer covariates. It 
was possible to circumvent this limitation by adjusting 
the length of each trackline to allow an estimate of 
density in the vicinity of each trackline because g(0) 
and length are multiplied together to estimate density. 
The estimates ofg(0) for each transect segment were 
averaged for all three years weighted by the transect 
lengths to estimate an average glO). The length of each 
transect segment was multiplied by its estimated g(0) 
divided by the average g(0) to generate an adjusted 
transect segment length which accounted for the g( 0). 
The adjusted transect segment lengths were then used 
in DISTANCE in place of the actual lengths and the av- 
erage g(0) calculated above was used as the global g(0) 
in DISTANCE. The standard error for the global g( 0) 
was estimated as the weighted average of the standard 
errors of the g( 0) estimates for the individual transect 
segments. 
Estimation of abundance 
The line-transect analysis program DISTANCE (vers. 
3.5) was used to estimate the observed density of harbor 
porpoise in each surveyed region. Two sighting prob- 
ability curves were estimated so that for transect seg- 
ments with usable belly observer effort data, sightings 
from the side and belly observers could be combined and 
duplicates removed or, when no belly observer data were 
available, sightings from the side observers only could be 
used. To identify significant effects of possible covariates 
for estimated strip width (presence or absence of a belly 
observer, survey year, individual observers, visibility 
levels, glare types, percent cloud cover, and sea state), 
each factor was considered separately as a covariate and 
the one with the lowest value for the AIC was retained. 
This process was repeated with the remaining possible 
covariates in an additive manner until further addition 
of covariates did not lower the AIC. Distances were 
pooled into 50-m bins to allow application of the esti- 
mate of perception bias. Densities were estimated for the 
individual areas with usable survey data. Unsurveyed 
areas such as the small bays and inlets were assigned 
the average densities from the surveyed areas of that 
stratum. These densities were then averaged, weighted 
by the area of each survey region, to estimate an aver- 
age observed density and abundance for each stock. 
Variances were calculated as in Buckland et al. (2001). 
The correction factor for availability bias is the inverse 
of the estimate of availability from Laake et al. (1997) 
(2.96 = (l/0.338), CV=0.18). This factor was applied as 
a multiplier to the observed abundance estimates to 
produce the abundance estimate for each stock. 
Incorporation of other survey data 
The vast area comprising Alaska waters made it impos- 
sible to survey all areas where harbor porpoises occur. 
Harbor porpoise sighting data were available from a 
concurrent NMFS beluga whale line-transect survey in 
Cook Inlet. For this survey, an Aero Commander aircraft 
with bubble windows was used; however, the windows 
were smaller than those of the Twin Otter aircraft, and 
the observers could not see directly below the plane. 
Survey methods were similar, except that the beluga 
whale survey was conducted at an altitude of 244 m and 
the primary focus was beluga whales. The search effort, 
therefore, was not concentrated as close to the trackline 
as it would have been if the survey had been designed 
to survey harbor porpoise. NMFS National Marine 
Mammal Laboratory has conducted these beluga whale 
surveys each year since 1993. We estimated abundance 
for harbor porpoise in Cook Inlet using the 1998 survey 
data, a strip width estimated from all beluga surveys 
(1993 to 1999), and the correction for availability bias 
from Laake et al. (1997). Perception bias could not be 
estimated for this survey. This abundance estimate was 
added to the abundance estimate from the GOA survey 
to produce a combined estimate for that stock. 
Minimum abundance estimate 
A minimum abundance estimate, W min , defined in Wade 
and Angliss (1997) as the lower 20 th percentile of the 
lognormal error distribution, is used in management 
