Abstract
Bias in underwater visual census has always been elusive. In fact, the choice of sampling method and the behavioural traits of fish are two of the most important factors affecting bias, but they are still treated separately, which leads to arbitrarily chosen sampling methods. FishCensus, a two-dimensional agent-based model with realistic fish movement, was used to simulate problematic behavioural traits in SCUBA diving visual census methods and understand how sampling methodology affects the precision and bias of counts. Using a fixed true density of 0.3 fish/m(2) and a fixed visibility of 6 m, 10 counts were simulated for several combinations of parameters for transects (length, width, speed) and point counts (radius, rotation speed, time), generating trait-specific heatmaps for bias and precision. In general, point counts had higher bias and were less precise than transects. Fish attracted to divers led to the highest bias, while cryptic fish had the most accurate counts. For point counts, increasing survey time increased bias and variability, increasing radius reduced bias for most traits but increased bias in the case of fish that avoid divers. Rotation speed did not have a significant effect in general, but it increased bias for fish that avoid divers. Wider and longer transects and a faster swim speed are beneficial when sampling mobile species, but a narrower, shorter transect with a slow swim is beneficial for cryptic fish.