Fig. 4 Probability of known foraging area inclusion within refined
foraging circles refined by multi-colony models with different
transferability (AUC). Coloured lines (± 95% CI) represent the
interaction between inclusion of known foraging areas in refined
foraging circles, when simulating refinement over all possible
percentiles of predicted foraging habitat suitability, and multi-colony
model transferability (five transferability values selected for
demonstration; regression equation provided to predict on log odds
scale). Black rings show the known foraging area inclusion observed at
tracked colonies when refined foraging circles were refined using the
percentile of habitat suitability selected by transferability-supported
refinement. The black line (± 95% CI) displays the trend through these
rings (from a generalised additive model), showing overall high known
foraging area inclusion and effective offset of greater refinement by
more accurate models.
Table 4. Species-specific logistic regression model coefficients for
predicting probability of known foraging area inclusion within refined
foraging circles. Intercepts give the probability of known foraging area
inclusion within unrefined global foraging circles. Coefficients are
presented on the log odds scale and can predict probability of known
foraging area inclusion for a desired percentile of habitat suitability
value (p ) from a multi-colony model with transferability
(t ) using the equation:
exp(β0+(β1*p )+(β2*p*t ))/(1+
exp(β0+(β1*p )+(β2*p*t ))).
For example, the probability of masked booby (MABO) known foraging areas
being included in a refined foraging circle refined with the
30th percentile (0.3) of foraging habitat suitability
and predicted from a poor transferability model (e.g. AUC=0.64) is:
exp(4.39+(-15.42*0.3)+(13.1*0.3*0.64))/(1+exp(4.39+(-15.42*0.3)+(13.1*0.3*0.64)))=
90%.