Fish behaviour effects on the accuracy and precision of underwater visual census surveys. A virtual ecologist approach using an individual-based model
Underwater visual census (UVC) methods are used worldwide to monitor shallow marine and freshwater habitats and support management and conservation decisions. However, several sources of bias still undermine the ability of these methods to accurately estimate abundances of some species.
The present study introduces FishCensus, a spatially-explicit individual-based model that simulates underwater visual census of fish populations. The model features small temporal and spatial scales and uses a movement algorithm which can be shaped to reflect complex behaviours and effects of diver presence. Four different types of fish were used in the model, featuring typically problematic behavioural traits, namely schooling behaviour, cryptic habits, shyness and boldness. Corresponding control types were also modelled, lacking only the key behavioural traits. Sampling was conducted by a virtual diver using four true fish densities and employing two distinct methods: strip transects and stationary point counts.
Comparisons with control fish have shown that schooling and bold behaviours induce positive bias and reduce precision, while cryptic and shy behaviours induce negative bias and increase precision, although shy behaviour did not have a significant effect on precision in transects. By looking at deviations from true density, however, schooling, shy and bold fish densities were strongly overestimated by both methods, while cryptic fish were slightly underestimated. Schooling and bold fish had the lowest precision overall, followed by shy fish. Fish rarity decreased precision, but had no effect on bias. Stationary points had less precision than transects for all fish types, and led to much higher counts, resulting in greater overestimation of density overall.
By modelling complex behaviour, it was possible to separate the contributions of detectability and non-instantaneous sampling on bias, and gain a deeper understanding of the effect of behavioural traits on UVC estimates. The model can be used as a tool for planning and optimization of monitoring programs or to calculate conversion factors for past or ongoing surveys, assuming behavioural patterns are well replicated.