Estimation of detection probability in aerial surveys of Antarctic pack-ice seals

• Published in: Journal of agricultural, biological, and environmental statistics biological, and environmental statistics, 2007-03, Vol.12 (1), p.41-54
• Main Authors: Southwell, C, Borchers, D, Paxton, C.G.M, Burt, L, de la Mare, W
• Format: Article
• Language: English
• Subjects: abundance; Agronomy. Soil science and plant productions; Aircraft; algorithms; Biological and medical sciences; Biometrics; Biometrics, statistics, experimental designs, modeling, agricultural computer applications; Conditional probabilities; covariates; detection; Environmental surveys; equations; estimation; Flight paths; Fundamental and applied biological sciences. Psychology; Generalities. Biometrics, experimentation. Remote sensing; generalized additive models; Geometric lines; heterogeneity; Ice; line transect surveys; linear models; Mark release recapture; mark-recapture studies; Observation aircraft; Perpendicular lines; population dynamics; population size; seals; statistical analysis; surveys; wildlife
• ISSN: 1085-7117; 1537-2693
• DOI: 10.1198/108571107x162920

We use line transect detection functions together with generalized linear and additive models to estimate detection probability when detection on the line ("g(0)") may not be certain. The methods provide a flexible way of modeling detection probability for independent observer surveys, and for investigating the effects of explanatory variables. Analysis of data from an aerial survey of pack-ice seals produced g(0) estimates substantially below 1 for some observers (it varied from 0.80 to 0.98), demonstrated a fairly complex dependence of detection probability on covariates, and showed negative correlation between observers' search width and their g(0). In addition to illustrating the utility of generalized additive models for capturing the effect of covariates on detection probability, the analysis suggests that detection functions may be sufficiently variable that use of g(0) correction factors obtained from other surveys would be inadvisable. We recommend that estimation of g(0) be considered for all aerial surveys; if g(0) is found to be very close to 1, estimation from subsequent surveys under the assumption that it is 1 may be reasonable, but without any estimation of g(0), the assumption that it is 1 is a matter of faith.