Development of benthic monitoring approaches for salmon aquaculture sites using machine learning, hydroacoustic data and bacterial eDNA

Abstract

Intensive caged salmon production can lead to localized perturbations of the seafloor environment where organic waste (flocculent matter) accumulates and disrupts ecological processes. As the aquaculture industry expands, the development of tools to rapidly detect changes in seafloor condition is critical. Here, we examine whether applying machine learning to two types of monitoring data could improve environmental assessments at aquaculture sites in Newfoundland. First, we apply machine learning to single beam echosounder data to detect flocculent matter at aquaculture sites over larger areas than currently achieved used drop camera imaging. Then, we use machine learning to categorize sediments by levels of disturbance based on bacterial tetranucleotide frequency distributions generated from environmental DNA. While echosounder data can detect flocculent matter with moderate success in this region, bacterial tetranucleotide frequencies are highly effective classifiers of benthic disturbance; this simplified environmental DNA-based approach could be implemented within novel aquaculture benthic monitoring pipelines.