An extension of the Jolly-Seber model combining two sources of capture-recapture data

Abstract

I propose a modification of the Jolly-Seber model, the two-source Jolly-Seber (TSJS) model, to estimate population size by combining two sources of capture-recapture data of the same population where there might be an unknown overlap between two datasets. This is the case with recent surveys of whales and dolphins where researchers use individual identification records from both photo-identification and DNA profiling of skin biopsy samples. This sampling configuration results in two datasets that might contain the same individuals. This new approach enables the estimation of the overlap and the calculation of the population size using capture-recapture information arising from both sampling methods. Monte Carlo simulations are used to assess the properties of the present estimator. When all the assumptions are met, the estimator seems to be unbiased as long as the occasion-specific simultaneous sampling probability is above 0.2. Simulation analyses also indicate that the proposed method performs better than existing closed-population estimators when there is little heterogeneity among individuals in capture probabilities and when the average capture probability is high. Alternatives have been explored and a two-source version of model M0 has also been developed and compared to the TSJS estimator. Traditional closed-population estimators have been compared to the new approaches (TSJS and two-source M0 models) when the population is open and the assumption of homogeneous capture probability is violated. Both procedures are finally applied to real data on the humpback whale Megaptera novaeangliae, on the wintering grounds of New Caledonia (South Pacific), where individuals have been sampled independently by skin sampling biopsy and photo-identification or simultaneously by both methods on a same capture occasion. The proposed methods hold great promise in monitoring by providing researchers and managers with a method allowing a diversity of sampling protocols. It could be more efficient in estimating population size, in terms of both precision and bias, than models based only on one type of data. And as it is important to control variation in a sampling design, this methodology could also provide a useful way to reduce variation by increasing the sample size and, hence, to enhance the estimator precision.