Fig. 6

Predicted versus observed number of buzzes per dive on test set for leave-one-out split (left) and chronological split (right). The line is the identity and thus, the closer the points fall to this line, the better the predictions. Point size indicates number of observations. There are more buzzes and dives in the left plots because the leave-one-out split has larger test set and smaller training set than the chronological split. Top row shows results for the main model, while bottom row shows results for the quasi-poisson model