Results

Foraging niche transferability

Multi-colony models were not able to successfully generalise the foraging niche of modelled species across all colonies, showing generally poor global transferability (mean AUC: 0.56, range 0.4-0.82; Table 3). However, leave-group-out cross validation showed that multi-colony models could predict suitable foraging habitat at some individual colonies (Fig 3).
Exploration of inter-colony transferability revealed that colony-specific models were transferable to one, or sometimes several, other colonies, but not all colonies (Fig. 3). For many modelled species this issue manifested as clustering in colony transferability: groups of colonies that could predict to one another, but poorly to other groups of colonies (Fig. 3-J), and partially explains low global transferability of colony-specific models (Table 3). Inter-colony transferability was not explained by geographical distance or oceanographic similarity between colonies (19 of 20 Mantel tests p > 0.05; supporting information). However, with the exception of noddies, all multi-colony models had slightly better global transferability than the colony-specific model average (Table 3), indicating multi-colony models generalised information from disparate colony clusters, boosting transferability.
We found that multi-colony models frequently (species mean: 41%; Table 3) matched or outperformed colony-specific models at local prediction (multi-colony leave-group-out cross validation vs colony-specific spatial cross validation; multi-colony row values vs diagonal values in Fig. 3). This finding indicates that the global foraging niche was able to predict suitable foraging habitat at approximately 40% of colonies, with comparable or better accuracy than models of the local foraging niche, although this was strongly species dependent (0% for noddies, 66% for frigatebirds and terns; Table 3). However, all of the colony-specific models from GBR tracking were superior to their multi-colony model equivalent (brown booby from Raine Island and Swain Reefs, Fig. 3A; masked booby from Swain Reefs, Fig. 3E; wedge-tailed shearwaters from Heron Island on short trips, Fig. 3I and long trips, Fig. 3G; and noddies from Heron Island, Fig. 3C).
Despite poor overall global transferability, we observed differences between modelled species (Table 3). Booby species had similar multi-colony global transferability (AUC: 0.53-0.55) and showed clustering of inter-colony predictive performance (Fig. 3A, E, F). Brown and masked booby colony-specific models had a greater range of transferability than red-footed booby but the same overall average (AUC: 0.51). Frigatebirds and wedge-tailed shearwater short trip multi-colony models (AUC: 0.61 and 0.58, respectively) showed better global transferability than boobies, and the greatest transferability increase over the colony-specific model average. Both of these modelled species showed clustering of inter-colony predictive performance, the former demonstrating examples of good inter-species and inter-colony transferability (Fig. 3G, I). Tropicbirds and wedge-tailed shearwater long trip multi-colony models had poor global transferability, equivalent to that of boobies (AUC: 0.56 and 0.54 respectively). Sooty terns and noddies showed the poorest global transferability (AUC: 0.48 and 0.40 respectively), with the latter’s multi-colony model being outperformed by the colony-specific model average. Terns showed the best global transferability between colonies (AUC: 0.82) but were limited by sample size (Fig. 3B).
Table 3 Summary of model transferability and foraging radii. Global transferability is the predictive performance of a model averaged over all colonies (MEAN column; Fig. 3). Local prediction gives the percentage of colonies where the multi-colony model matched or outperformed self-prediction by the colony-specific model. Foraging ranges are summarised across tracked colonies (for colony breakdown see supporting information), and inclusion of known foraging areas in global foraging circles are averaged across all tracked colonies.