Combining population density estimates in line transect sampling using the kernel method

Line transect methods that are pooling robust allow data from different transects or locations to be pooled for estimation of population density. This is particularly important in situations where data from individual transects are sparse and pooling is done out of necessity. In this study, we investigate a method for combining estimates from individual transects when each transect has sufficient data to support estimation with the kernel method. It is based on a minimizer of the asymptotic mean squared error of a linear combination of the individual population density estimators. The asymptotic mean squared error of the simple pooled estimator is always at least as large as the optimally combined estimator. We apply this combination to two estimates from data on a real population of mussels. Using a variety of simulations, we demonstrate the better finite-sample efficiency for combining unbalanced cases. In practice, if the detection functions were identical, it could be better to pool, but the gains are modest. On the other hand, when the detection functions are different, it can be substantially better to combine. This recommends the new linear combination.