Generating unbiased estimates of burrowing seabird populations

Maximising survey efficiency can help reduce the trade-off between
spending limited conservation resources on identifying population changes
and responding to those changes through management. Burrow-nesting
seabirds are particularly challenging to survey because nests cannot be
counted directly. We evaluated a stratified random survey design for
generating unbiased population estimates simultaneously for four petrel
species nesting on Macquarie Island, Australia, where the survey cue,
burrow entrances, is similar for all species. We also compared the use of
design-based and model-based analyses for minimising uncertainty in
estimates. We recorded 2,845 Antarctic Prion burrows, 306 White-headed
Petrel burrows and two Blue Petrel burrows while distance-sampling along
154 km of transects. For Blue Petrels and Grey Petrels, we completed
nocturnal searches along a further 71 km and searched 249 km of tracks
during follow-up ground searches. We failed to generate unbiased
population estimates for two rare and localised species, Blue and Grey
Petrels, from our stratified random survey. Only for the most widespread
and abundant species, Antarctic Prion, did the estimate have reasonable
power to detect a rapid population change. Model-based analyses of the
stratified random survey data did not improve upon traditional
design-based analyses in terms of uncertainty in population estimates, but
they did provide useful spatial representation of current populations.
Models that used the targeted survey data did not reflect current
population sizes and distributions of the two rare and localised species.
We found that when species ecologies, distributions and abundances vary, a
multi-method approach to surveys is needed. Species with low abundance
that occur patchily across large islands are likely to be best estimated
using targeted surveys, whereas widespread and abundant species can be
accurately and precisely estimated from randomised surveys using
informative model-based analyses.