Comparison of three methods of estimating the population size of an arboreal mammal in a fragmented rural landscape

Abstract

Context. Precise and accurate estimates of animal numbers are often essential for population and epidemiological models, as well as for guidance for population management and conservation. This is particularly true for threatened species in landscapes facing multiple threats. Estimates can be derived by different methods, but the question remains as to whether these estimates are comparable. Aims. We compared three methods to estimate population numbers, namely, distance sampling, mark–recapture analysis, and home-range overlap analysis, for a population of the iconic threatened species, the koala (Phascolarctos cinereus). This population occupies a heavily fragmented forest and woodland habitat on the Liverpool Plains, northwestern New South Wales, Australia, on a mosaic of agricultural and mining lands. Key results. All three methods produced similar estimates, with overlapping confidence intervals. Distance sampling required less expertise and time and had less impact on animals, but also had less precision; however, future estimates using the method could be improved by increasing both the number and expertise of the observers. Conclusions. When less intrusive methods are preferred, or fewer specialised practitioners are available, we recommend distance sampling to obtain reliable estimates of koala numbers. Although its precision is lower with a low number of sightings, it does produce estimates of numbers similar to those from the other methods. However, combining multiple methods can be useful when other material (genetic, health and demographic) is also needed, or when decisions based on estimates are for high-profile threatened species requiring greater confidence. We recommend that all estimates of population numbers, and their precision or variation, be recorded and reported so that future studies can use them as prior information, increasing the precision of future surveys through Bayesian analyses.