248 
THE WILSON JOURNAL OF ORNITHOLOGY • Vol 124. No. 2. June 2012 
HG. 2. Location o| transects and plots (white dots) on Grand Colombier Island. 
content by hand when burrow-scoping inspectio 
was not possible. Occupancy was defined as 
binomial variable (1 = presence of adult, egg o 
chick; 0 = empty), and occupancy probability wa 
estimated as die proportion of occupied burrow 
over the number of checked burrows. We tested fo 
an ellect ol slope and vegetation type oi 
occupancy probability using a logistic regressiot 
(logit link, binomial distribution). Candidate mod 
es were selected using an information theoretic 
approach (Burnham and Anderson 2002) Non 
breeding birds or failed breeders within burrow, 
dwelling petrel species are known to visit oi 
temporarily occupy unoccupied burrows, poten¬ 
tially resulting in overestimates of abundance 
(Heaney et al. 2002). However, non-breeding birds 
visit the colony mainly at night and rarely stay in 
the burrow during the day (Warham 1990 
Huntington et aL 1996). Ratcliffe et al. (1998^ 
estimated the probability of a nest being occupied 
diurnally by a non breeder on a given day at 0 024 
in the closdy related European Storm Petrel 
^abates pelagians). Thus, we do not exclude 
he possibility that non-breeding birds may have 
occupied some burrows during our survey, but they 
were unhke.y to constitute a serious bias in 
estimation ol the breeding population. 
Estimating Incubation Failure Preceding 
probatihr'K WC eStimated incubad °n failure 
u P ,ed bUrr ° WS at the '"Sinning of the breeding 
season. Quadrats with active burrows covering all 
habitat types on the island were randomly selected 
(n = 13; number of burrows per quadrat = 8 ± 
0.3) by 18-19 June. Burrows were individually 
identified within each quadrat, using numbered 
30-cm long wood sticks with red colored tips. 
Contents of active burrows occupied by an 
incubating adult (n — 105) were inspected at 
"HO- day intervals (18-19 Jim. 2 Jul, 10 Jul). until 
the start of the survey, allowing us to observe if 
burrows were failed or successful. A nest was 
considered tailed if no egg was detected during 
one of the inspections. Failure was considered as a 
binomial variable and failure probability was 
estimated as the proportion of nests that failed 
over the number of active burrows. 
Burrow Detection Probability.—We used a 
double-observer approach to estimate the burrow 
detection probability of each observer (Nichols et al. 
2000). A sample of plots (n = 13) was surveyed by 
pairs of observers counting burrow entrances 
independently. Plots were selected in both lent (« 
~ 8) and herbaceous habitats (n = 5). The ftN 
observer marked all the detected burrows. Marks 
were placed within the burrow's so they were 
invisible for the second observer. The second 
observer systematically recorded previously-marked 
burrows and those missed by the first observer. The 
tank of each observer alternated randomly. We ran a 
set of models incorporating different sources of 
variation in detection probability, i.e., observer 
identity and habitat type, and selected among 
candidate models using an information theoretic 
approach (Burnham and Anderson 2002). 
