Ecoacoustics – the study of natural and anthropogenic sounds and their relationship to the environment – offers a remote, cost-effective, and scalable method for monitoring freshwater streams. However, processing the vast datasets generated in acoustic monitoring programs remains a challenge. Although advancements in audio recording technology have enabled large-scale acoustic monitoring, efficient analysis methods have lagged, particularly in freshwater environments. Current approaches are often time-consuming, offer only high-level insights, or require specialised IT knowledge, limiting their accessibility to ecologists.
One of the primary challenges in freshwater ecoacoustics is extracting target sounds from recordings, as water flow in lotic systems frequently masks species vocalisations. To address this, we developed a new protocol that efficiently identifies sound-types in freshwater streams without manual annotation or reliance on complex supervised machine learning algorithms. Using dissimilarity indices and nested clustering, we successfully identified prominent sound-types in Warill Creek, Kalbar, Australia, overcoming issues of acoustic overlap.
Our protocol achieved external validation scores of 75% or higher, a true positive rate exceeding 90%, and accurately identified almost 90% of the sound-types in the recordings. We have since applied this method to over 40 stream recordings collected across South-East Queensland with consistent success. This streamlined approach demonstrates a novel application of clustering techniques in freshwater ecoacoustics and highlights the need for accessible tools to analyse large acoustic datasets with minimal manual effort.