Enhancing automated analysis of marine soundscapes using ecoacoustic indices and machine learning

Historically, ecological monitoring of marine habitats has primarily relied on labour-intensive, non-automated survey methods. The field passive acoustic (PAM) demonstrated the potential this practice to automate surveying in habitats. This been through use ‘ecoacoustic indices’ quantify attributes from natural soundscapes. However, investigations using individual indices have had mixed success. Using PAM recordings collected at one world’s largest coral reef restoration programmes, we instead apply a machine-learning approach across suite ecoacoustic improve predictive power ecosystem health. Healthy and degraded sites were identified live cover surveys, with 90–95% 0–20% respectively. A library one-minute extracted each. Twelve calculated for each recording, up three different frequency bandwidths (low: 0.05–0.8 kHz, medium: 2–7 kHz broad: 0.05–20 kHz). these 33 index-frequency combinations differed significantly between healthy best performing single index could only correctly classify 47% recordings, requiring extensive sampling site be useful. We therefore trained regularised discriminant analysis algorithm discriminate an optimised combination indices. multi-index discriminated two habitat classes improved accuracy compared any isolation. pooled classification rate 1000 cross-validated iterations model 91.7% 0.8, mean SE) success classifying recordings. was subsequently used actively restored sites, established >24 months prior values 79.1% (±3.9) 66.5% (±3.8). Of 37/38 33/39 received as also newly <12 25.6% (±2.6), which 27/33 classified degraded. investigation highlights value combining demonstrates monitor recovery over time, reducing reliance labour-intensive in-water surveys by experts. As access recorders continues rapidly advance, effective automated will needed keep pace expanding datasets.