Abstract
Acoustic data are an invaluable source of information for characterizing the distribution and abundance of mid-trophic level organisms (MTLOs) in the ocean. These organisms play a key role in the ecosystem as prey of top predators and as predators of lower trophic level organisms, as well as in carbon export from the surface into deeper waters. This study used 38 kHz-EK60 acoustic echosounder data from six cruises spanning 2011–2017 to explore the seasonal and spatial variability in the vertical distribution of MTLOs’ from 10 to 600 m in the New Caledonian (South Pacific) Exclusive Economic Zone. A total of 16715 acoustic vertical profiles of acoustic backscattering strength were clustered into homogeneous groups. Two small shallow scattering layers (SSLs) between 0 and 100 m, and one large deep scattering layer (DSL) at around 550 m depth, characterized the mean vertical distribution of MTLOs. A machine-learning model (eXtreme Gradient tree Boosting algorithm, XGBoost) was fitted to explain the acoustic profile clusters with environmental variables as predictors. Sun inclination was the most important factor in structuring the vertical profile shapes due to the diel vertical migration signal, followed by the mean oxygen value of the top 600 m. Bathymetry, euphotic depth, 0–600 m mean temperature and SST were the next most significant variables. Isotherm depth, surface chlorophyll-a, wind, and mean salinity had a lower influence on the shape of the vertical profiles. The model was then used to construct vertical echograms at the scale of the New Caledonian EEZ, showing an accuracy up to 87% in cross validation. Across the EEZ, the shape of vertical acoustic profiles were comparable, though layer echo intensities varied spatially with a marked north-south gradient that remained relatively constant seasonally. The vertically-averaged acoustic values were characterized by a maximum to the south of the EEZ in summer, mainly driven by high oxygen values as well as shallow euphotic depth. We also estimated a migrant proportion between day DSL and night SSL of about 78%. Our methodology offers a promising approach for analyzing the control of the environment on the vertical distribution of MTLOs for other oceanic provinces, while also providing a framework to investigate the corresponding trophic interactions between MTLOs and their predators feeding at different depths and times. Moreover, our findings stress the need to consolidate knowledge on species composition to optimize acoustic data interpretation.