246 
Fishery Bulletin 113(3) 
Table 1 
Details about the 18 line-transect surveys for harbor porpoise ( Phocoena phocoena) con- 
ducted from vessels in the inland waters of Southeast Alaska in 1991-2012 during spring, 
summer, and fall seasons. 
Year 
Season 
Survey dates 
Number of 
days surveyed 
Vessel 
1991 
spring 
20 April-3 May 
14 
NOAA Ship John N. Cobb 
summer 
15-25 July 
11 
John N. Cobb 
fall 
12-25 September 
14 
John N. Cobb 
1992 
spring 
29 April-12 May 
14 
John N. Cobb 
summer 
11-24 June 
14 
John N. Cobb 
fall 
10-23 September 
13 
John N. Cobb 
1993 
spring 
30 April-13 May 
14 
John N. Cobb 
summer 
7-20 June 
14 
John N. Cobb 
2006 
spring 
1-11 May 
11 
John N. Cobb 
summer 
7-17 July 
11 
John N. Cobb 
2007 
spring 
19-28 April 
10 
John N. Cobb 
summer 
7-17 July 
10 
John N. Cobb 
fall 
10-20 September 
10 
John N. Cobb 
2010 
summer 
19 July-1 August 
14 
FV Steller 
fall 
9-22 September 
14 
FV NW Explorer 
2011 
summer 
1-14 June 
14 
RV Medeia 
2011 
fall 
25 August-7 September 
14 
RV Medeia 
2012 
summer 
7-20 July 
14 
RV Aquila 
Total 230 
veys, perpendicular distance data are pooled to obtain 
a single detection function for the whole study period 
and area, and that pooled detection function is then 
used to compute annual or stratum-specific abundance 
estimates (e.g., Hammond et al., 2002; Barlow, 2006; 
Zerbini et al., 2006). In the present analysis, to mini- 
mize variability between survey periods that could lead 
to biased estimates of this parameter because of the 
temporal changes in survey protocols described above, 
detection probabilities ( P ) were estimated for each pe- 
riod by pooling perpendicular distances for the peri- 
ods 1991-1993, 2006-2007, and 2010-2012 separately. 
Detection probability was estimated by modeling un- 
grouped data of perpendicular distances that were 
truncated at 2 km by using both conventional dis- 
tance sampling (CDS) and multiple covariate distance 
sampling (MCDS) approaches (Buckland et al., 2001; 
Marques and Buckland, 2003). The two methods differ 
in that MCDS allows for the inclusion of environmen- 
tal covariates in the estimation of detection probability 
(Innes et al., 2002; Marques and Buckland, 2003). 
Models were proposed to investigate the effects of 
covariates on P, and the models included group size as 
continuous covariates with year and Beaufort category 
as factor covariates. The Beaufort category had 2 lev- 
els: a “low” sea state (Beaufort 0-2) and a “high” sea 
state (Beaufort >3). In addition, a “ship” covariate was 
proposed for the period 2010-2012 to assess whether 
the use of ships with different height platforms had 
an effect on detection probability. This covariate had 
4 levels, namely 1 level for each ship used during the 
summer surveys. 
For each year, covariates were tested singly or in 
additive combination. It was expected that P was posi- 
tively correlated with group size and platform height 
but negatively correlated with Beaufort sea state. If a 
proposed model was inconsistent with these expecta- 
tions, then that model was deleted from the analysis 
before model selection was performed (e.g., Zerbini et 
al., 2006). The model with the lowest Akaike’s informa- 
tion criterion (AIC) was used for statistical inference 
(Burnham and Anderson, 2002). In the estimates pro- 
vided here, the probability of detecting porpoise on the 
trackline was assumed to be unity (g(0)=l; see Discus- 
sion section). 
Estimation of group size 
Porpoise were considered to be in a group when ani- 
mals were within 10-15 body lengths of each other. 
Group size has the potential to affect estimates of 
P. If larger groups are easier to detect further away 
from the trackline, then use of average group size can 
bias estimates (Buckland et al., 2001). In our explor- 
atory analysis, regression of group size versus detec- 
tion probability (Buckland et al., 2001) indicated that 
detections were independent of group size. Therefore, 
stratum-specific simple means were used after trunca- 
