Inferring the abundance of an island megaherbivore from semi-structured distance sampling

A core component of conservation management projects is the estimation and monitoring of abundance of target species, in space and through time. Monitoring wildlife is complicated and relies on detection, which itself depends on environmental conditions, phenotypes and behaviours of the species being monitored. Aldabra giant tortoises (Aldabrachelys gigantea) have great importance as ecosystem engineers, keystone species and ecological replacements. Accurate monitoring is vital to understanding both the size and dynamics of their native and translocated populations, and the species’ influence on the wider ecosystem. Despite a long history of organised research in their native range on Aldabra Atoll in the Seychelles, to date no rigorous, atoll-wide population surveys have attempted to estimate population size of these giant tortoises. The complexity of irregular transect surveys, through a variety of habitats, in several areas of the atoll, necessitates the use of state-space statistical models that capture heterogeneities in both abundance and detection. Here we develop a Bayesian Hierarchical Distance Sampling (HDS) model to infer habitat- and season-dependent detectability and abundance of the Aldabra giant tortoise population on Aldabra. Between 2018 and 2021 the population was stable and numbered approximately 180,000 individuals, the highest estimate so far calculated for this population. We attribute this stability and high abundance to the extended history of conservation efforts on Aldabra, including invasive alien species eradication which should be maintained and expanded. We provide a reproducible workflow for Bayesian distance sampling methods using data-augmentation and habitat-dependent detectability and recommend its use on legacy datasets and future abundance surveys for similar species. These methods enable robust inference of abundance and detectability even when data are irregular or inconsistently collected, providing a framework for integrating legacy datasets with modern conservation workflows.