Artificial lateral line for aquatic habitat modelling: An example for Lefua echigonia

Abstract

Abstract The lateral line system allows fish to sense the surrounding hydrodynamics via changes in the pressure, acceleration and velocity fields, providing additional information about physical habitat structure. Relations between fish and habitat preference have been traditionally inferred using the parameters of water depth, time-averaged velocity, substrate and vegetation to predict their abundance or presence. However, current methods rely on time-averaged point observations, ignoring the fluid-body interactions used by fish to sense the local flow environment. Here we present the first study to explore the use of artificial lateral lines as tools for aquatic habitat assessment. Relations are explored between conventional physical habitat variables in conjunction with measurements from a pressure sensor-based artificial lateral line probe (LLP). Comparisons are performed using field data of an endangered species, Lefua echigonia, in the Yagawa River (Japan). Random forest presence/absence models were created using habitat variables and the pressure-based LLP variables. Results show that the pressure-based variables were strongly correlated with the habitat variables, indicating that the LLP is able to capture the complex multivariate information encoded in the conventional variables with a single time-averaged measurement. In addition, presence/absence models of L. echigonia based on pressure-based variables marginally outperformed models based on the habitat variables. The major findings of this work are: i) LLP-based habitat models are capable of providing similar or better results than when using conventional habitat variables with considerably less field sampling effort, and ii) the probe-based method removes the subjective bias from observations and reduces model dimensionality. The results of this study indicate that fish habitat models can be efficiently and accurately carried out using a lateral line probe instead of traditional multivariate models based on the water depth, flow velocity, substrate and vegetation.