Accounting for uncertainty in duplicate identification and group size judgements in mark–recapture distance sampling

Abstract

Mark–recapture distance sampling (MRDS) surveys with two independent observers are widely used to estimate wildlife population abundance. The analysis relies on accurate identification of duplicate sightings common to both observers, and correct judgements of group size, both of which are hard to achieve for species that exhibit complex grouping patterns. In this paper, we examine the impact of these sources of uncertainty on bias and precision of abundance estimates, using a case study of 22 aerial surveys of common dolphins (Delphinus delphis) in the Hauraki Gulf, New Zealand. We develop a novel probabilistic method to identify duplicate observations, and account for various sources of uncertainty using a simulation‐intensive approach. For our case study, identifying duplicates using reasonable but arbitrary thresholds of time and angle discrepancies created a range of abundance estimates differing by up to 20%, whereas our novel threshold‐free probabilistic analysis returned an estimate roughly central to this range. Uncertainty in group size made a smaller impact of up to 5% on abundance estimates. All analysis choices returned similar values for the coefficient of variation, from 20 to 23%. Generating robust estimates of abundance, and accounting for all associated sources of uncertainty, is critical for informing conservation management. Our novel approach provides a way to eliminate arbitrary decisions associated with MRDS, and account for a wider range of uncertainties. Our method allows for the reliable application of MRDS to a wider range of terrestrial and marine species, and will be a useful tool for producing robust abundance estimates for species that exhibit complex grouping patterns.

Abundance estimates and habitat preferences of bottlenose dolphins reveal the importance of two gulfs in South Australia

“…This could potentially have biased the abundance and density estimates downwards. Lastly, group size bias is likely negligible in our study because of the small school sizes for bottlenose dolphins typical of the study area, which make it easier to count them 45,46 . In combination, these biases could have led to an underestimate in bottlenose dolphin abundance and density in the study area, but is unlikely to have resulted in overestimates in any sub-region, overall or in the different seasons.…”

“…Identification of duplicate sightings was obvious in our study because individuals formed distinct clusters in manageable densities that were easily identified by the observers. In cases where individuals of a species occur in complex grouping patterns or higher densities, other methodologies may be more suitable and time effective to reliably identify duplicate sightings and reduce biases in the abundance estimates 46 .…”

Distribution and Abundance of the Eastern Canada – West Greenland Bowhead Whale Population Based on the 2013 High Arctic Cetacean Survey

“…For example, a front right sighting would be paired with a back left sighting that occurred within the 10 second interval. Unlike the sameside dataset, this opposite-side dataset cannot possibly contain true duplicate sightings (Hamilton et al, 2018). Logistic regression was used to determine which covariates best discriminated among the 2 datasets, with the assumption that those same covariates were also useful to identify data that contained duplicate pairs.…”

Narwhal Abundance in the Eastern Canadian High Arctic in 2013

“…However, our approach is not "fully" probabilistic (i.e., it does not provide an actual probability of a pair of sightings being a duplicate). A further methodological development has been proposed by Hamilton et al (2018), who also used a comparison of opposite and same-side datasets to fit a mixture distribution fitted by maximum likelihood to estimate a probability that a same-side pair is a duplicate. These probabilities were then used in a bootstrap of the entire MRDS analysis to ensure that the uncertainty around duplicates is fully propagated into the abundance estimates.…”