Improving precision in multiple covariate distance sampling: A case study with whales in Alaska.

摘要

Population abundance is a key parameter in wildlife research, assessment, and management. It is often used to monitor trends across time, as input parameters in population and ecosystem modeling and to investigate habitat changes. The often low and variable precision of abundance estimates can constrain management requirements. In this study, the performance of variance estimators for conventional (CDS) and multiple covariate distance sampling (MCDS) frameworks is investigated through simulation. The estimator in which the estimated abundance can vary by line in the sampling variance component, the Laake estimator, presents better coverage than the other estimators in the presence of spatial covariates. CDS and MCDS methods and the best performing variance estimator were used to estimate that 1652 (95% CI = 1142-2389) fin (Balaenoptera physalus ), 2644 (95% CI = 1899-3680) humpback (Megaptera novaeangliae ), 1233 (95% CI = 656-2315) minke (B. acutorostrata), 251 (95% CI = 97-644) transient and 991 (95% CI = 379-2585) resident killer whales (Orcinus orca) were found in coastal waters of western Alaska and the eastern and central Aleutian Islands in the summer of 2001-2003. Annual rates of increase were estimated at 4.8% (95% CI = 4.1%-5.4%) for fin and 6.6% (95% CI = 4.7%-8.4%) for humpback whales from time series of population size estimates from 1987 to 2003. The relatively small number of killer whale sightings observed in this study (n=39) is insufficient to obtain a precise estimate of detection probability (P) and therefore population size. In, the numbers provided above, the variance of P corresponded to 36-38% and 16-26% of the total variance for the estimates of transient and resident killer whale abundances, respectively. A new approach was adopted to reduce variance of P. Sighting data from surveys conducted in various areas in the North and Eastern Tropical Pacific Ocean were combined and a new estimate of detection probability were obtained within the MCDS framework. Variances of the new Ps were reduced to 6.5-12% and 4.5-16% for transients and resident ecotypes, respectively. Precision of killer whale abundance estimates were reduced by as much as 15%. This new method can be extended to other wildlife species but presents some caveats that must be carefully considered before its application.