Broadband acoustic modeling for fish body orientation: A functional GAM approach with common bream case

Determining the orientation of fish is crucial when processing hydroacoustic recordings, especially for accurate estimates of fish size and biomass. Variations in body orientation can significantly bias acoustic measurements. In this study, a novel modeling approach for estimating fish body orientation from pulse-compressed broadband acoustic signals is investigated. We applied a functional generalized additive model (FGAM) to fast- and slow-tapered broadband acoustic data from tethered common bream ( Abramis brama ) and evaluated its ability to estimate dorsal body orientation (±40° from true dorsal position). We assessed model reliability and generalizability using rigorous cross-validation, in which data from each fish were excluded during training and predicted using models trained only on the remaining individuals. Additionally, we subjected the model to a further, much more stringent test by applying the bream-trained model not only to a different individual, but to another species, the roach ( Rutilus rutilus ), to assess its interspecies robustness. Results show that the model captured the general trend of dorsal orientation transition from head to tail, albeit with some bias. The performance was better for slow-tapered acoustic pulses, compared to the fast variant. While the FGAM-based approach is not perfectly predictive, it provides a transparent framework for inferring individual-level traits from acoustic data. By incorporating orientation variability into size estimation, this method contributes to a more informed interpretation of broadband signals and supports better assessment of fish in ecological research and fisheries monitoring. • Fish orientation is a key source of bias in hydroacoustic size estimates. • A functional GAM extracts dorsal orientation from broadband echo signals. • Training on bream shows best accuracy with slow-tapered acoustic pulses. • Cross-validation and testing on roach confirm robustness and transferability.