Unsupervised classification increases precision in acoustic broadband target strength measurements of mesopelagic fish.

Mesopelagic fish are widespread and abundant in global oceans, contributing to nutrient cycling and potential future fisheries. Estimating their distribution and abundance is challenging due to limitations of ship-based echosounders and trawling. Submersible broadband acoustic probes are often used for precise target measurements due to increased range resolution and the possibility to use higher frequencies with expanded spectra; however, broadband measurements can introduce undesired signals into the data. Using an unsupervised outlier detection approach, we enhanced the selectivity of broadband acoustic targets of mesopelagic fish in the Labrador Sea. We observed high variability in frequency response within insonified volumes and echo-traces. Applying an outlier detection algorithm, we filtered anomalous signals, improving density estimates by reducing positive skew up to 35%. Our approach increases measurement precision and provides insight into broadband echosounder applications for mesopelagic fish assessments.