Ibis (2007), 149, 693– 700
© 2007 The Authors
Journal compilation © 2007 British Ornithologists’ Union
Blackwell Publishing Ltd
On the importance of g(0) for estimating bird population
densities with standard distance-sampling:
implications from a telemetry study
and a literature review
ERICH BÄCHLER* & FELIX LIECHTI
Swiss Ornithological Institute, CH-6204 Sempach, Switzerland
Raw count data are often used to estimate bird population densities. However, such data
do not consider detection probability. As an alternative, methods that model detection prob-
ability such as distance-sampling have been proposed. However, standard distance-sampling
provides reliable estimates for absolute density only when the underlying assumptions are
met. One of the most critical of these assumptions is that animals on a transect line or at an
observation point have to be detected with certainty (the g(0) = 1 assumption). We radiotagged
nine Orphean Warblers Sylvia hortensis and estimated their short-distance detection
probability. Birds were radio-located in 264 cases in single bushes or trees. Their visual
detection probability after a 5-min search was only 0.58 (sd = ±0.14, range = 0.38–0.80),
although the observer knew the bird’s location. Furthermore, we carried out a literature
review to assess how the g(0) = 1 assumption is handled in practice. None of the 28 standard
distance-sampling papers reviewed contained an estimation of g(0). In 57% of the papers,
the g(0) = 1 assumption was not even mentioned. Nevertheless, none of the authors declared
their estimates as being relative. Our empirical data show that the g(0) = 1 assumption
would be severely violated for a foliage-gleaning bird species at a desert stopover site outside
the breeding season. The literature review revealed that the testing of the g(0) = 1 assumption
is largely ignored in practice. We strongly suggest that more attention should be paid to the
testing of this key assumption, because results may not be reliable when it is violated. If it
is not possible to test the g(0) = 1 assumption or g(0) is less than 1, alternative methods
should be used. Another possibility is to estimate detection probability by the means of
radiotagged individuals.
Bird population trends are good indicators of
biodiversity and environment (Gregory et al. 2005).
Within this context, numerous monitoring pro-
grammes for birds have been initiated in recent
years. But despite all efforts that have been made to
improve monitoring techniques, obtaining reliable
estimates of bird abundance still remains a difficult
challenge. As a first step in the estimation procedure,
birds have to be counted. The most commonly used
counting techniques are territory mapping, line
transects, point counts and mist-netting. Count data
are often used directly as an index of population
density. Such index values do not consider detection
probability of the birds and therefore may not
represent absolute population densities (Emlen 1977,
Burnham 1981), yet they are often used as relative
values to make comparisons between datasets (e.g.
Hutto et al. 1986). For such comparisons to be
valid, mean detection probabilities of the studied
animals have to be constant over the dimensions to
be compared (e.g. Wilson & Bart 1985, Farnsworth
et al. 2002, Norvell et al. 2003). However, detection
probability of birds is affected by numerous factors
(Farnsworth et al. 2002, Rosenstock et al. 2002) and
therefore the assumption of stationary detectability
is rarely met (Norvell et al. 2003, Kéry et al. 2005).
To convert count data into estimates of absolute
density, numerous methods that model detection
probability have been developed. The most important
*Corresponding author.
Email: erich.baechler@vogelwarte.ch