Fig. 2
From: Hierarchical habitat selection by a predatory fish in a patchy seascape
Workflow of the field and analytical methods used in this study. Seatrout were acoustically tagged and released in Rankin Basin. SAV surveys were conducted at each receiver each year, consisting of 20 quadrat surveys. SAV survey data as well as data from FHAP and aerial imagery were used to map habitat variables. COAs were calculated from acoustic detections in 60-min time bins then grouped by individual, daily period, and season to randomly distribute pseudo-absence points in a 1:1 ratio with the COAs. Habitat data were extracted from the habitat maps at each COA and pseudo-absence point then used to train a random forest (RF) model using a 70% training—30% testing data split. Predictor importance was determined using the decrease in accuracy method, while marginal effects were investigated with partial dependency plots.
