We found no sex-related differences in maximum foraging ranges and durations of trips made by black-winged petrels (foraging range: F₁₂₂ = 0.738, p = 0.541; trip duration: F₁₂₂ = 0.613, p = 0.462), white-necked petrels (foraging range: F₂₀₈ = 0.533, p = 0.595; trip duration: F₂₀₈ = 0.198, p = 0.843) and Kermadec petrels (foraging range: F_6.1 = -0.773, p = 0.469; trip duration: F_5.8 = -0.59, p = 0.577). Therefore, we pooled the data from both sexes within each species. When accounting for random effects of individuals making repeated trips, maximum foraging ranges differed between black-winged and white-necked petrels and black-winged and Kermadec petrels (Tukey’s post-hoc test: p < 0.001 for both relationships; Table 2 ), but not between white-necked and Kermadec petrels (Tukey’s post-hoc test: p = 0.098). Similarly, the duration of foraging trips was different between black-winged and white-necked petrels (Tukey’s post-hoc test: p = 0.0105) and between black-winged and Kermadec petrels (Tukey’s post-hoc test: p < 0.001), but not between white-necked and Kermadec petrels (Tukey’s post-hoc test: p = 0.0679). Tracking data also indicated that all three species exhibited bimodal foraging strategies, alternating between short and long foraging trips ( Figures 2 , 3 ).

Behavioral state space modeling revealed that although all three species demonstrated area-restricted search foraging behavior throughout the entire range of their breeding distributions, they tended to concentrate foraging activity nearer (< 500 km) the colony ( Figures 4 – 6 ). On average all species spent more than half of their time traveling, and a relatively small proportion of their time resting during foraging trips ( Table 3 ). We found significant between-species differences in the proportion of time spent in ARS behavior (F₂ = 10.98, p = < 0.001). Black-winged petrels spent proportionally less time in ARS than white-necked (p < 0.001, 95% C.I. = [0.02 – 0.09]) and Kermadec petrels (p < 0.001, 95% C.I. = [0.04 – 0.14]), but there was no difference between white-necked and Kermadec petrels (p = 0.2, 95% C.I. = [-0.08 – 0.01]). We found no significant difference in the proportion of time spent foraging during daylight and darkness in black-winged petrels (daylight: 30.4 ± 14.2, darkness: 28.6 ± 17.1; t-test: t ₍₂₅₈₎ = -0.986, p = 0.325) and Kermadec (daylight: 37.9 ± 17.7, darkness: 37.8 ± 20.8; t-test: t ₍₁₀₂₎ = -0.013, p = 0.989), but found that white-necked petrels spent proportionally more time foraging during daylight (daylight: 40.4 ± 15.4, darkness: 28.5 ± 21.2; t-test: t ₍₄₀₀₎ = -6.72, p < 0.001).

Foraging ranges extended over vast oceanic areas including, for black-winged ( Figure 4 ) and white-necked petrels ( Figure 5 ), almost throughout the entirety of the Tasman Sea and, to a lesser extent, at the southern limits of the Coral Sea at its 30° south limit. Approximately 55%, 35% and 42%, respectively, of black-winged, white-necked and Kermadec petrel area-restricted search foraging locations occurred in international waters (outside of EEZs). All three species foraged broadly within the warm waters of the wind-driven, southward flowing East Australian Current (EAC) and to the southern extent of the eastward-flowing Tasman Front (TF), which separates the Coral Sea to the north and the Tasman Sea to the south and flows from the east coast of Australia into the western Pacific Ocean. Black-winged petrels made use of areas bordering the Subtropical Frontal Zone in the south of the Tasman Sea and foraged over an area of almost 4 million km², white-necked petrels over approximately 5 million km² including to the east of New Zealand bordering the Subtropical Frontal Zone, but Kermadec petrels ( Figure 6 ) remained relatively close to the colony with a foraging range that spanned almost 0.5 million km². Grid cell analysis of foraging behavior revealed that gadfly petrels engaged in area-restricted search behavior throughout their entire foraging ranges. However, most grid cells occupied by foraging black-winged petrels on long trips were concentrated in the south of the Tasman Sea ( Figure 4 ). Foraging ranges of both black-winged and white-necked petrels were larger in 2018 than in 2019 ( Figures 4 , 5 ). There was a high degree of overlap in foraging ranges of all three species, but black-winged petrels appeared to concentrate their long foraging trips considerably further south than the other species ( Figures 4 – 6 ).

Evaluation metrics for habitat models constructed with spatially independent CV were substantially poorer (AUC_ROC of 0.55 – 0.77; Table 4 ) than of models fit using conventional random CV (AUC_ROC of 0.93 – 0.95; Table 4 ). Despite these apparent differences in model performance, response curves and relative importance of predictor variable contributions of conventional CV and spatially independent CV ensemble models were similar ( Supplementary Figures 4 , 5 ) indicating consistent environmental relationships, regardless of the CV method. Both methods are likely to have poor generalizability to new spatial and temporal environment data because the spatially independent CV method is specifically designed to evaluate how well model predictions transfer to new environmental contexts. Ensemble model predictive performance for black-winged petrels was approximately equivalent to random classification (AUC_ROC = 0.55; TSS = 0.19), and performance was fair for white-necked (AUC_ROC = 0.71; TSS = 0.39) and Kermadec petrels (AUC_ROC: = 0.77; TSS = 0.48) (Araujo et al., 2005; Hosmer et al., 2013). Distance from the colony ( Figure 7 ) was the most important predictor, with all three species demonstrating a high probability of foraging closer to the colony. All species showed a preference for foraging in warmer sea-surface temperatures, and for black-winged and white-necked petrels SST was the second most important environment variable for predicting the location of foraging. Black-winged and white-necked petrels preferred to forage in areas with higher wind speeds (albeit wind speed had only low (3.1%) variable importance for white-necked petrels), but Kermadec petrels preferred lower wind speeds ( Figure 7 ). Models presented some evidence that black-winged petrels expressed preference for foraging in areas with shallower thermocline depths (i.e., 9.9% variable importance; Figure 7 ), whereas white-necked petrel foraging was associated with deeper thermocline depths (i.e., 6% variable importance; Figure 7 ). Thermocline depth had weaker contribution to the Kermadec petrel models (i.e., 3.1% variable importance) with no obvious preference for foraging in waters with shallow or deep thermocline layers.

We found no evidence of any of the species showing preferences for foraging close to seamounts. Conversely, the probability of foraging was uniformly higher among all species at greater distances from seamounts; and birds did not appear to target bathymetric slopes. Kermadec petrels demonstrated some preference for foraging in areas with higher chlorophyll a concentration compared to black-winged and white-necked petrels. Additionally, for all three petrels, species distribution models could not determine any clear preferences for foraging near sub-mesocale sea-surface temperature fronts, nor transport fronts (i.e., areas with lower FSLE values that represent the outer parts of transport fronts) or areas with higher sea level anomalies. However, black-winged petrels appeared to make some general use to the north of the Subtropical Frontal Zone (Bostock et al., 2015; Figure 4 ) during long foraging trips. In 2019, only two white-necked petrels traveled to the east of New Zealand, possibly making use of productive waters to the north of the Subtropical Frontal Zone ( Figure 5 ).

MESS grids indicated that only 0.04 ± 0.08% (mean ± SD) of the foraging area during the time period when Kermadec petrels were tracked had environmental conditions outside the range of those during the time period when black-winged and white-necked petrels were tracked ( Supplementary Figure 6 ).

Our tracking data and analyses demonstrate that summer-breeding gadfly petrels from the Norfolk Island Group use waters of the East Australian Current and the Subtropical Frontal Zone. Behavioral state-space modeling demonstrated that while they do use some oceanographic features, they also travel very widely, foraging throughout most of the Tasman Sea, apparently opportunistically exploiting resources as they travel (e.g., Ventura et al., 2020). These results support the hypothesis that, while gadfly petrels do make some broad use of oceanographic features, they also use a maximal-area foraging strategy, travelling over large areas to maximize chances of encountering unpredictable resources in relatively unproductive marine ecosystems (e.g., Clay et al., 2017; Ventura et al., 2020).

Among seabirds, gadfly petrels undertake some of the most extensive foraging movements during breeding when they must travel repeatedly to and from their nests to incubate eggs or provision themselves and their altricial offspring (Clay et al., 2019; Taylor et al., 2020). The three species we studied all demonstrated bimodal foraging strategies (e.g., Weimerskirch et al., 1994) whereby birds made both short and long foraging journeys. Concentrated foraging around seabird colonies depletes local resources, which forces individuals to journey further in search of food: a phenomenon known as ‘Ashmole’s halo’ (Ashmole, 1963; Oppel et al., 2015; Weber et al., 2021). Seabirds exhibiting bimodal foraging strategies do so in response to near-colony resource depletion, whereby short foraging trips are used to provision their dependent young, whereas parents need to travel to more distant foraging grounds to meet their own energetic requirements (Weimerskirch et al., 1994; Oppel et al., 2015; Weber et al., 2021). Gadfly petrels appear exceptionally well suited to making long-distance self-provisioning journeys due to their unique morphological adaptations for fast, and energy efficient flight relative to other seabirds (Spear and Ainley, 1997; Ventura et al., 2020).

Tracking data collected over multiple years highlighted that some areas experience more intensive foraging than others. This is true for both the number of individuals foraging in a given region as well as the intensity of foraging activity of individuals within that region. All three species had high foraging activity immediately surrounding the colony, with other hotspots of foraging activity relatively diffusely spread over the remaining large foraging extent, which is typical of other central place foraging seabirds that display dual foraging strategies (Magalhães et al., 2008; Raine et al., 2021). Hotspots of black-winged petrel foraging activity occurred between approximately 40°S – 47°S. The large spatial area over which gadfly petrels foraged encompasses areas of the Exclusive Economic Zones (EEZs) of Australia and New Zealand with a large proportion of foraging ranges encompassing the high seas, a “global commons” for which no country holds sovereign jurisdiction. In this study, approximately 55%, 35% and 42%, respectively, of black-winged, white-necked and Kermadec petrel foraging locations occurred outside of EEZs, which highlights the importance of internationally coordinated efforts to conserve gadfly petrel foraging habitat (e.g., Davies et al., 2021).

We demonstrated a large degree of spatial overlap in foraging areas among all three species, suggesting that competition for resources may be mediated over finer spatial scales through differences in prey acquisition methods, presumably as a result of morphological differences (e.g., body size), instead of large-scale spatial or habitat segregation (Spear et al., 2007). Diet diversity in pelagic tropical seabirds is generally low (Diamond, 1983) with most species restricted to foraging on squid and flying fish, which results in high diet overlap (Ashmole and Ashmole, 1967; Harrison et al., 1983; Ballance et al., 1997). Indeed, observations of stomach regurgitations and of parents feeding young during fieldwork of the present study suggest that the three species provisioned chicks with unidentified squid (Decapodiformes) and flying fish (Exocoetidae) and, in black-winged petrels, unidentified zooplankton (L. Halpin, personal observations, 2018-2021). Kermadec petrels are reported to be partially kleptoparasitic (Spear and Ainley, 1993). That Kermadec petrels foraged closer to the colony and spent on average proportionally more time in area-restricted search behavior may be indicative of a kleptoparasitic foraging strategy, whereby birds do not need to travel far to find prey, but instead may harass other species close to the colony that have acquired food. This species also demonstrated preferences for foraging in areas with higher chlorophyll a concentration, suggesting that within their smaller foraging range, which encompasses more oligotrophic areas, they may need to seek patches with higher productivity to obtain sufficient prey resources. All three species intensively used the areas surrounding the colony in short foraging trips, but during long trips black-winged petrels tended to forage in cooler waters much further south than white-necked petrels. This may be indicative of black-winged petrels having increased flight proficiency relative to white-necked petrels thus enabling them to exploit resources further afield.

Model predictive performance was generally poor to fair as is often the case for habitat generalists that lack strong affinity to a particular habitat type or feature (Elith et al., 2006; Andrew and Fox, 2020). We believe that the poor predictive performance was due in part to the behavioral characteristics (i.e., more generalist habitat requirements) of gadfly petrels rather than factors such as the choice of environmental variables used to predict foraging habitat. Furthermore, the response curves and relative importance of the predictor variables to model predictions were very similar between conventional cross-validation and spatially independent cross-validation approaches, demonstrating that birds have weak or wide preferences for available habitat, regardless of the cross-validation method. While the predictive performance of the Kermadec petrel species distribution model was within range of what is generally considered acceptable, we suggest that the performance of this model should be taken with caution due to the lower representativeness of the sample size (68.5%) to foraging movements of the colony-level population. MESS grids indicated a general similarity in environmental conditions experienced by Kermadec petrels compared to the other two species. We therefore do not believe that differences between the foraging characteristics of Kermadec petrels compared to the other two species were driven by differences in environmental conditions due to the different time periods in which the species were tracked.

Our results suggest that black-winged petrels target waters with shallower thermocline depths, which may indicate a more varied diet that includes zooplankton since planktivorous species tend to prefer waters with shallower thermocline depths (e.g., Spear et al., 2001; Ballance et al., 2006). Conversely, the preference for foraging in waters with deeper thermoclines expressed by white-necked petrels may indicate that the species is more piscivorous than black-winged and Kermadec petrels. That black-winged and Kermadec petrels spent almost equal proportions of their time foraging during darkness compared to daylight suggests that nocturnal foraging is likely to be an important feature of their behavior. These results are consistent with other studies indicating that tropical and subtropical seabirds often rely on deep water prey that perform diel vertical migrations to the sea surface at night (e.g., Hays, 2003). Such prey includes mesopelagic squid, fish and zooplankton (Ashmole, 1971; Imber et al., 1995; Dias et al., 2012). Similar to a closely related species, the Bonin petrel (Pterodroma hypoleuca; Harrison et al., 1983), black-winged petrels may be well-adapted for night time foraging and may rely substantially on diel migrant prey, possibly including bioluminescent squid (Ommastrephidae; Harrison et al., 1983) and zooplankton. Kermadec petrels are known to feed on diel migrant squids, particularly Onychoteuthidae (Imber et al., 1995).

Previous studies have suggested that, unlike many other pelagic seabirds, gadfly petrels have less clear habitat preferences, especially when engaged in long foraging trips during the breeding season (Clay et al., 2017; Clay et al., 2019; Ventura et al., 2020). Our study provides support for this hypothesis with evidence to suggest that the foraging strategy is likely based on opportunistic prey encounters rather than targeting of specific habitat features. For example, foraging occurred throughout the species ranges, and travel accounted for a disproportionately large percentage of time allocation within foraging trips. Additional support for this hypothesis is that, across all three species, the most important model predictor of probability of foraging behavior was the distance from the colony. From a species distribution modeling perspective, this suggests that, like other centrally placed foragers, breeding gadfly petrels’ foraging ranges are mainly constrained by their need to return to nests to provision chicks.

Possible evolutionary drivers of this lack of habitat specificity are that (sub)tropical waters and open ocean regions are less predictable than temperate and polar regions, and inshore waters. Indeed, it has been hypothesized that as ocean productivity decreases – which results in reduced prey abundance – seabird flight proficiency becomes more important because those with low flight costs will be able to reach more distant resources (Ballance et al., 1997). Furthermore, gadfly petrels’ high mobility and large foraging ranges allows birds to be less reliant on specific oceanographic features associated with moderate to high productivity, such as fronts or eddies, which can be ephemeral and are highly dynamic in marine systems (Weimerskirch, 2007), and instead allows birds to opportunistically target prey items or patches they encounter while in flight.

Gadfly petrels tracked in the present study showed no affinity for foraging near seamounts, similar to Raine et al. (2021). Several studies (e.g., Haney et al., 1995; Thompson, 2007; Morato et al., 2008) demonstrate that some seabirds – including gadfly petrels – occur in higher abundance near seamounts, particularly in the North Pacific and Atlantic. However, these studies have generally occurred during gadfly petrels’ non-breeding stages (i.e., in the austral winter) when birds are not constrained to routinely return to the colony. Given the low energetic costs of long-distance travel, it could be that for chick-provisioning gadfly petrels, targeting foraging at seamount locations yields lower foraging success than ranging widely across the Tasman Sea. Furthermore, little is known about seabird prey distributions in the oceanic Pacific (Ballance et al., 1997). It is thus also possible that the squid and flying fish prey targeted by gadfly petrels during chick-provisioning occur in no higher abundances at accessible seamounts within the species’ foraging ranges than in pelagic waters.

The opportunistic foraging strategies of gadfly petrels and the enormous areas over which they forage mean that it may be challenging or impractical to use traditional approaches to identifying and protecting key foraging habitats, such as static marine protected areas (Oppel et al., 2018). Conservation of gadfly petrel foraging habitat may therefore require focus on maintaining functional and healthy marine ecosystems and reducing generalized threats (e.g., commercial fisheries that cause high seabird bycatch, marine pollution) to birds in these areas, which will ensure long term food supplies for these oceanic wanderers. Lack of generalizability of habitat suitability models poses a challenge for conservation planners. If it is not possible to predict where suitable habitat is under current environmental conditions, then spatially dynamic conservation measures, which have been proposed and enacted in other marine contexts, may be impossible to implement (Hobday et al., 2011; Maxwell et al., 2015). However, our analysis of at sea distribution presents some opportunities. Our tracking dataset identifies some areas of high foraging activity by multiple individuals between years. This approach to examining areas with repeated use between years is likely more useful for determining important foraging areas for gadfly petrels and other seabirds that may lack predictable foraging preferences for any static and dynamic oceanographic features. Moreover, the dual foraging strategy that we identified also creates potential for different conservation strategies to provide protection for the different trip types. The short foraging trips cover a much smaller area, meaning that methods such as marine protected areas could be effective at this spatial scale (Oppel et al., 2018). Some existing protections that prohibit commercial fishing occur in the Norfolk Island National Park Zone to the north of the Island ( Figure 8 ), which encompasses a small portion of the areas used by gadfly petrels during short trips. Conversely, the vast extent covered during longer foraging trips by self-provisioning gadfly petrels means that policy-based methods such as fisheries regulation (e.g., quotas and more comprehensive enforcement) or management of marine plastic pollution will likely represent the best tools for conserving birds undertaking these long-distance trips (Oppel et al., 2018). While the risk of interactions with commercial fisheries is considered low for the small to medium sized gadfly petrels in our study (Waugh et al., 2012), their populations will benefit from increased efforts to prevent the degradation of marine ecosystems through pollution, over-fishing, and climate change both within and outside of EEZs. Although the conservation of the high seas is much more challenging than within EEZs, several large high seas areas of the Tasman Sea – which are frequented by gadfly petrels in the present study – have been identified as ecological or biologically significant marine areas (EBSAs) by Parties to the Convention on Biological Diversity (Bax et al., 2016). These areas have been the topic of discussions initiated by a United Nations call to establish marine protections beyond national jurisdictions. Provided that their nesting colony on Phillip Island remains free of introduced predators, conservation efforts for these gadfly petrels should focus on collaborative multinational cross-border efforts to conserve and restore marine ecosystem function in the Tasman Sea and the wider South Pacific.