BAYESIAN INFERENCE IN LINE TRANSECTS WITH DOUBLE COUNT SAMPLING AND IMPERFECT ON-LINE DETECTION

Abstract

For the management and conservation of wild animal populations it is fundamental to know its abundance. However, if imperfect detection, a very common phenomenon in field counts, is ignored, abundance will be underestimated. We show that Bayesian hierarchical models for double observer distance sampling data are capable of simultaneously estimating abundance and detection probabilities and propose a simple model where detection probabilities are modeled as logit or probit regressions of distance-to-line and give its implementation in BUGS code. With a simulation study we verify empirically that double observer information increases the precision in abundance estimates by about 30% when compared with estimates from distance data only. We further verify that the model is capable to correctly estimate observer-specific detection probability, but underestimates abundance by 12% on average. We also apply an extension of these models to a population of loon (QUANG and BECKER, 1997; URL:http://www.jstor.org/stable/1400405.1997). Our estimate of 154 (posterior mean) was much higher than the estimated 99 individuals reported by QB although other model parameters are similar. Some new model-specific goodness-of-fit diagnostics are proposed and applied.