Quantitative examination of data processing effects on species classification from broadband echosounders

Abstract

Measurements from broadband echosounders hold much promise for improving the ability to discriminate biological targets in the ocean. However, it remains unclear how effectively the increased bandwidth can be leveraged in situ and few clear protocols exist to guide the user in making choices in processing these voluminous data. We used a 45–170 kHz wideband echosounder and a high definition video system to simultaneously observe aggregations of fish and invertebrates in situ using a remotely operated vehicle (ROV). The dataset contains a large number of echoes collected in a minimally disturbed context, avoiding any effects of compression/decompression of air-containing target and the identity of the targets was verified at a scale and resolution comparable to the acoustic measurements. Using these data, we quantitatively explore the effects of (1) the chosen frequency resolution, (2) the method and scale of data averaging, (3) available bandwidth, (4) data normalization choices, and the (5) use of volume scattering versus resolved single targets on the ability to discriminate echoes from three species. There was sufficient information in broadband data to accurately discriminate between echoes from known species but the accuracy of classification was strongly affected by the data processing choices.