Estimating consumption rate of Atlantic salmon smolts (Salmo salar) by striped bass (Morone saxatilis) in the Miramichi River estuary using acoustic telemetry

Abstract

Differentiating detections of a telemetered fish from those of predators which may have consumed that telemetered fish presents problems and opportunities. Previous efforts to quantitatively classify predation events have had to rely on data from unknown states of fish (i.e. unsupervised learning techniques) with the consequence that model performance cannot be refined or compared to alternate models. We circumvent this limitation by analysing acoustic telemetry track data to differentiate movement patterns of tagged Striped Bass (Morone saxitilis) from those of Atlantic Salmon (Salmo salar) smolts which were known to not have been predated by Striped Bass over a three year period in the Miramichi River estuary. A random forests classification model (i.e. supervised learning technique) was used to differentiate the movement patterns of these two species and the model was applied to Atlantic Salmon smolt movement characteristics to provide an index of Striped Bass predation-derived mortality. The optimized random forests model inferred that predation rates by Striped Bass were highly variable between years for two smolt stocks, ranging from 1.9% to 17.5%. Spatial and temporal overlap of the two species is a likely factor defining the between stock and annual variation of predation rate estimates.