A total of 24 oceanic manta rays were tagged at three study sites between 2012 and 2022. Tags included 14 pop-up satellite archival tags (two MK10-PAT and 12 miniPAT), and 10 towed satellite tags (SPLASH10F - 321E) made by Wildlife Computers (Washington, USA). The first site was Raja Ampat in eastern Indonesia. Between September 2012 and May 2022, two MK10-PAT, nine MiniPAT, and one SPLASH10F - 321E tags were deployed using a modified pole spear while on SCUBA over a range of seasons; a subset of these tags are from Stewart et al. (2016a) and Beale et al. (2019). The second site was located near Tumbes off the coast of northern Peru. Between May and July 2018, three MiniPAT tags were deployed (detailed in Andrzejaczek et al., 2021). The third site was in the shallow coastal waters of the northeast shelf of Northland near Whangaroa in northern New Zealand. Between March 2019 and February 2022, nine SPLASH10F - 321E tags were deployed with a modified pole spear while freediving off Whangaroa Harbour, New Zealand.

Of the 24 tags, five MiniPAT tags were recovered from Indonesia, two MiniPAT tags were recovered from Peru, and one SPLASH tag was recovered from New Zealand. High sampling frequency (< 15-s) data were downloaded from the eight recovered tags, providing continuous time series of depth and temperature. The remaining 16 non-recovered tags transmitted summary data via satellite, which were aggregated into histograms and reported as mean values over intervals ranging from 6 to 24h.

Time series of depth and temperature data were downloaded from recovered tags. The first 24h of data were discarded from analysis to remove anomalous data resulting from tagging; manta rays were not restrained as part of the tagging procedures and are expected to return to normal behaviour rapidly. Depth sensor drift was present in one tag. To correct for depth sensor drift whereby the tag reported depths deeper than the animal, the software Igor Pro (WaveMetrics, 2024) was used to calculate an average zero-offset in a moving 72h window so that depth was aligned to zero at the surface. Further analyses detailed below were then completed in the R Statistical Environment (R Core Team, 2024). Non-diving periods were defined as time spent within 50 m of the surface. Pre- and post-dive intervals were therefore the durations spent not diving immediately before and after a dive. To investigate differences in non-diving periods between extreme and shallow dives, Mann-Whitney U-tests were conducted to assess significant differences in pre- and post-dive intervals. Vertical velocity was calculated from the difference in consecutive depths divided by the sampling frequency of each tag (3, 5, or 15 s). Individual dives were extracted from the depth time series using the R package ‘diveMove’ (Luque, 2007). A dive was defined as time spent below the threshold of 50 m (Thorrold et al., 2014). Mean vertical velocity of descent of dives was calculated as (maximum depth – 50)/time taken to descend from 50 m to the maximum depth; similarly, the ascent vertical velocity was calculated as (maximum depth – 50)/time taken to ascend from maximum depth to 50 m. A Mann-Whitney U test was conducted to assess differences in vertical velocities of descents and ascents between extreme and shallow dives.

Maximum likelihood tracks were estimated using Wildlife Computers’ software package ‘Global Position Estimator 3’ (GPE3). GPE3 applies a hidden Markov model (HMM) to estimate the maximum likelihood daily location of a tag (Wildlife Computers, 2022). The model considers maximum depth, light level (twilight), and sea surface temperature (SST) data along with known deployment and release locations to calculate the most likely daily location (Skomal et al., 2017). A travel speed of between 1.25 and 2.0 m s⁻¹ was input into the HMMs. This speed was calculated using a combination of several factors: mean speed between high-scoring fastloc GPS locations from SPLASH tag movements from multiple manta rays; drone video analysis of swimming oceanic manta rays in New Zealand; reported speeds of Caribbean manta ray Mobula cf. birostris (Fong et al., 2022); best fit with known locations of tag movements; and the resulting track score within GPE3 (Curnick et al., 2020). Tracks were assumed to consist of straight-line interpolated daily location estimates, providing a conservative estimate of rate of horizontal movement. Tracks did not account for land masses obstructing point-to-point locations. The straight-line distance between daily maximum likelihood locations is hereafter referred to as distance. To account for imprecise daily location estimates resulting from GPE3 error, we calculated mean distance over three days as a metric of movement, hereafter referred to as 72h distance. Three days was chosen based on the size of GPE3 error estimates, the number of twilights reported, the number and quality of locations for Splash tags, and the occurrence of gaps in transmitted data. Maximum likelihood tracks were plotted along with 50%, 75%, and 95% probability density surfaces (Andrzejaczek et al., 2021) and overlaid onto maps with the R package ‘ggOceanMaps’ (Vihtakari, 2023) using bathymetry data provided by NOAA (NOAA National Centers for Environmental Information, 2022).

To address known limitations of light-level-based location estimates (Lisovski et al., 2012), several quality assurance measures were applied to enhance track reliability. High-quality Fastloc GPS positions were used to validate GPE3 tracks where available, and GPE3 scores were optimised using known deployment and recapture locations (e.g., photo-ID matches). Travel speeds were adjusted following the approach of Curnick et al. (2020), and tracks with more than three consecutive days of missing twilight or maximum depth data were excluded from analysis. The use of 72h distance (mean three-day straight-line distance) further reduced sensitivity to daily positional uncertainty and provided a more robust measure of spatial movement.

A Spearman rank correlation test was used to explore the relationship between dissolved oxygen (DO) concentration and horizontal step depth during deep dives. Horizontal steps were defined as periods during deep or extreme dives when absolute vertical velocity was <0.1 m s⁻¹ for ≥ 1 min. DO data were obtained from the daily ‘Global Ocean Biogeochemistry Analysis and Forecast’ modelled dataset provided by Copernicus Marine Environment Monitoring Service (CMEMS) (CMEMS, 2019). For each dive, the date and most likely GPS location were matched to the nearest 0.25 × 0.25-degree grid cell in the CMEMS data, from which the vertical dissolved oxygen profile was extracted. DO data were available in 38 depth bins ranging from 0.49 to 1452 m, with greater resolution in shallow water. The DO value corresponding to each step depth was extracted from the matching vertical profile, allowing correlation analysis between step depth and local DO concentration. Validation indicates a typical uncertainty of ±5–10 µmol kg⁻¹ in the upper ocean, with greater variability below the thermocline (CMEMS, 2019). Although subject to model error, the vertical gradients are sufficiently resolved to identify relative changes in DO with depth relevant to mesopelagic diving behaviour.

Across all 24 recovered and unrecovered satellite tags, 2705 tag-days of data were recorded. Of these, 79 days included an extreme dive. To ensure the reliability of location estimates used for modelling, only daily GPE3 locations supported by high-quality twilight data (i.e., no gaps exceeding three consecutive days) and accompanying maximum depth records were included. Additionally, the datasets from Indonesia (5 of 1454 days featuring extreme deep diving) and Peru (3 of 170 days) were heavily zero-inflated and therefore excluded from modelling, as preliminary analyses indicated insufficient data to support inference. The final dataset comprised 572 days from tags deployed in New Zealand, where sufficient extreme dive occurrence and location data permitted robust modelling.

A generalised linear mixed model (GLMM) from the R package ‘lmerTest’ (Kuznetsova et al., 2020) was fitted to the occurrence of extreme dives as a function of oceanographic and location-based covariates, while accounting for the random effect due to individual variation (Howey et al., 2016). The daily GPE3 location estimates were used to calculate all position-dependent variables. A total of 13 biotic and abiotic covariates were considered. Covariates were rescaled to ensure comparability and screened for high collinearity (>0.50). The correlation structure among variables is shown in Supplementary Figure S1.

Daily distance was calculated initially; however, it was not included in the modelling. Instead 72h distance was used to allow for inaccuracy in daily location estimates. Net primary productivity and concentrations of phosphate, dissolved oxygen, and nitrate were also extracted from the CMEMS Biogeochemistry dataset but were removed due to high collinearity with other variables. Similarly, bathymetry and distance to nearest land were excluded due to high collinearity with distance to shelf edge, which is defined as the distance in km from each estimated position to the nearest location where bathymetry exceeded 200 m depths.

Mean 72h distance, distance to shelf edge, chlorophyll a concentration, moon phase, mean SST, and wind-generated waves were retained (Table 1), as they had the lowest collinearity. These were treated as fixed-effect explanatory variables. The binary response variable indicated whether an extreme dive occurred on a given day. Manta ID was included as a random effect. Stepwise model selection using the R package ‘MuMIn’ (Barton, 2022), guided by Akaike Information Criterion (AIC) and the principle of parsimony, identified the best-fitting model.

All means are reported with standard deviation (SD), and medians with interquartile range (IQR).

Animal ethics approval was granted by Murdoch University animal ethics committee permit number RW3318/21 and protocol ID 836, and by the Indonesian ethics committee of the National Research and Innovation agency (BRIN) proposal number 07062022000004.

Recovered tags and splash tags provided the highest location accuracy. Three MiniPAT tags (OM9, OM10, and OM12) were excluded from GLMM analysis due to multiple data gaps exceeding five days. Details on GPE3 scores, track speed, twilights, SSTs, and location accuracy are available in Supplementary Table S1. Mean daily distance travelled pooled across data sets was 17.1 ± 17.7 km d⁻¹ (median: 11.5 km d⁻¹, IQR: 16.8) and was lower in Indonesia (12.1 ± 10.5 km d⁻¹) than in Peru (24.0 ± 22.4 km d⁻¹) and New Zealand (32.1 ± 25.1 km d⁻¹). Oceanic manta rays tagged in Indonesia remained within the country’s exclusive economic zone (EEZ) (e.g., Figure 1A). However, those tagged in New Zealand and Peru travelled beyond national waters, entering both international waters and other nations’ EEZs (e.g., Figures 1B, C). Tags recovered from Indonesia after six-month deployments exhibited repeated latitudinal movements away from and back toward the tagging site, consistent with broad-scale north–south excursions. Although deployments in New Zealand and Peru were shorter, their movements also suggested cyclic patterns. In New Zealand, movements appeared seasonal, occurring as water temperatures began to drop, and oriented towards Fiji or Tonga. For example, OM8’s 67-d track covered >2500 km as it travelled north into tropical waters. OM8 exhibited three distinct phases during its 67-d track, representative of rays in this deployment site: an inshore phase along New Zealand’s coastal waters (mean daily straight-line distance 11.6 ± 11.1 km), a 29-d offshore phase covering 1700 km (60.3 ± 15.3 km d⁻¹), and a third phase involving lower daily straight-line distances near the Lau Islands (24.5 ± 13.4 km d⁻¹; Supplementary Table S2; Figure 1C).

Mean daily SSTs varied between tags (Tables 2, 3) and sites: Indonesia (28.0 ± 1.3°C), Peru (24.3 ± 1.0°C), and New Zealand (23.9 ± 2.6°C). OM8 experienced the greatest shift in SST, from 22.1 ± 0.23°C at the deployment site in New Zealand (−34.9°S) to 27.1 ± 0.11°C at the release location in the tropical Lau Islands (−18.5°S). This individual left the coastal waters of New Zealand heading north at the onset of the austral winter.

Analysis of the eight recovered tags across the three study sites (Table 2) revealed substantial individual variability in the diving behaviour of oceanic manta rays, with some individuals exclusively using shallow waters, whereas others dived beyond depths of 750 m. Depth and temperature were recorded at sampling frequencies of 3–15 s, providing a detailed profile of vertical movement patterns. The tagged oceanic manta rays exhibited a broad depth range, extending from the surface down to 1,246 m, in water temperatures from 29.8 to 4.6°C. Across these tags, 46,945 dives were identified from the DTS. Shallow dives (<200 m) comprised 98.36% of dives, whereas deep dives (200–500 m) and extreme dives (≥500 m) comprised 1.62% (n = 760) and 0.02% (n = 11), respectively. Four bathypelagic dives (>1000 m) were recorded, one in Indonesia and three from a ray tagged in New Zealand. The mean maximum depth across deep and extreme dives was 258.9 m (± 94.1). Individual variability in depth use was pronounced, with rays spending varying amounts of time across different depth ranges. For example, OM1 in Indonesian waters spent 51.0% of its time in shallow waters (≤50 m), 48.4% diving in the epipelagic zone between 50 and 200 m, and 0.6% in depths greater than 200 m. In contrast, OM3 in Indonesian waters spent 100% of its time in shallow waters, with a maximum depth of 174.2 m, with no recorded deep dives.

Extreme dives were recorded on 79 of 2,505 days (3.2%) across all 24 tags, with the most recorded from New Zealand rays (71 of 79). The median number of days between extreme dives from New Zealand rays was 2 (IQR 3). Visual inspection of movement tracks indicated an apparent association between the occurrence of extreme dives and subsequent increases in 72h distance (Figure 2; Supplementary Figures S26–S30). The median 72h distance for all tags was 35.3 km (IQR 38.2), with a maximum of 294 km recorded within 72h of an extreme dive.

Extreme dives in oceanic manta rays (Mobula birostris) followed a consistent sequence comprising five phases: extended pre-dive surface intervals, steep descents, horizontal steps at depth, gradually slowing ascents, and extended post-dive surface intervals. Pre-dive durations were significantly longer than for routine dives, often exceeding an hour, with dives typically initiated from depths <10 m. Descent vertical velocities reached up to 2.9 m s⁻¹, exceeding the threshold for passive gliding and suggesting actively powered descents. These descent speeds were markedly higher than those recorded during routine swimming or straight-feeding in reef manta rays (M. alfredi) (~0.75 m s⁻¹) (Fong et al., 2022) and far exceed gliding velocities (<0.2 m s⁻¹) documented in negatively buoyant elasmobranchs (Gleiss et al., 2011; Braun et al., 2014; Fontes et al., 2022). Below 200 m, descents were punctuated with horizontal steps—brief pauses in vertical movement with velocities <0.1 m s⁻¹—that often occurred near midwater dissolved oxygen peaks. Maximum depths reached 1250 m, with no prolonged bottom phase. Ascents slowed progressively, typically including additional shorter-duration steps. Post-dive surface intervals were again significantly extended, suggesting physiological recovery following extreme dives. Notably, extreme dives were followed by significantly increased horizontal displacement, with peaks in 72h distance frequently exceeding 200 km. This structure of long pre- and post-dive intervals, steep V-shaped dive profiles, interspersed steps, and absence of sustained bottom phases aligns closely with extreme dive profiles reported in shortfin mako sharks (Isurus oxyrinchus) (Sepulveda et al., 2004), elephant seals (Mirounga angustirostris) (Robinson et al., 2012), blue sharks (Prionace glauca) (Howey et al., 2016), tunas (Thunnus spp.) (Musyl et al., 2011), and scalloped hammerhead sharks (Sphyrna lewini) (Klimley, 1993), where they have often been associated with navigation, environmental sampling, or rapid transit through less favourable habitats.

The dive profiles of extreme dives differed from typical foraging dives. Oscillatory movements commonly associated with manta rays tracking prey within discrete layers such as deep scattering layers (Stewart et al., 2016b; Andrzejaczek et al., 2020) were absent, and dive depths consistently exceeded deep scattering layers depths in New Zealand (~457 m) (Batzler, 1975) and Indonesia (300 – 400 m) (Schalk, 1987), which are typically associated with mesopelagic prey. Additionally, horizontal steps, although potentially indicative of foraging in other oceanic species such as the Chilean devil ray (Mobula tarapacana) and whale shark (Rhincodon typus) (Wilson et al., 2006; Thorrold et al., 2014; Tyminski et al., 2015), did not correspond with vertical foraging behaviours of manta rays such as barrel rolling (Stevens, 2016), and individuals often continued descending after completing horizontal steps, which is difficult to reconcile with prey encounter and exploitation. Although opportunistic feeding during steps cannot be entirely ruled out, there is little direct evidence of foraging to support this assumption (Thorrold et al., 2014; Tyminski et al., 2015; Braun et al., 2022), and the observed dive profiles and vertical velocities argue against foraging as the primary function of extreme dives. The combination of high descent speeds, absence of prey-associated oscillations, and consistent increases in post-dive displacement supports the hypothesis that extreme dives serve an information-gathering or orientation function. This interpretation is further supported by similar potential behavioural links between deep dives and subsequent travel observed in scalloped hammerhead sharks (Klimley, 1993), blue sharks (Carey et al., 1990), and oceanic whitetip sharks (Carcharhinus longimanus) (Howey et al., 2016).

Certain epipelagic species appear better adapted than others to perform temporary deep-water excursions, but few spend substantial periods of time below 200 m (Braun et al., 2022). At depths exceeding 1000 m, animals experience low temperatures, high pressure, and sometimes hypoxic conditions. Ambient temperatures at these depths are well below the thermal optimum for most tropical shallow-water species (Rummer et al., 2014). As a result of thermal inertia, large-bodied marine animals experience a lag in body temperature relative to ambient conditions. By swimming at high speeds, animals minimise the time in descent and therefore presumably arrive at the maximum depth of their dive with a higher residual body temperature and/or can dive deeper before they reach a critical thermal threshold. The high vertical velocities observed during descent suggest that oceanic manta rays reduce their exposure to cold environments by minimising descent time, despite the higher energetic cost of active swimming, similar to blue sharks (e.g., Carey et al., 1990). It is also conceivable that oceanic manta rays use a form of “breath holding” similar to that observed in scalloped hammerhead sharks. Breath holding involves the suspension of convective heat transfer, which is likely achieved by reducing water flow across the gills via mouth closure or restricted gill ventilation (Royer et al., 2023). This behaviour allows scalloped hammerheads to descend to greater depths and/or stay in cold waters for longer periods, ensuring substantive cooling only occurs during the latter stages of the ascent phase once, presumably, gill ventilation is resumed (Royer et al., 2023). If oceanic manta rays use this behaviour, it would help limit heat loss, allowing them to reach greater depths while delaying the reduction of body temperature. The lower vertical velocities observed during ascent compared to descent may facilitate physiological recovery, allowing for progressive reoxygenation following anaerobic exertion, similar to behaviours observed in deep-diving tunas and sharks (Carlson and Parsons, 2001; Fitzgibbon et al., 2010).

The presence of horizontal steps during descent and ascent presents a challenge to this hypothesis. If horizontal steps represent recovery from anaerobic exercise, individuals will presumably need to allow water to pass over the gills, thereby increasing heat exchange with the surrounding water and likely accelerating body cooling. The rate at which oceanic manta rays lose body heat remains unknown, but further investigation using temperature sensors embedded in the musculature could clarify body temperature changes during descent and could reject hypotheses about the function of horizontal steps proposed earlier (sensuCarey et al., 1990). The physiological limitations of oceanic manta rays likely play a key role in shaping their deep-diving behaviours. Even among closely related species, physiological adaptations can drive marked differences in vertical movement behaviour. For example, although both oceanic manta rays and Chilean devil rays possess cranial rete, their ability to retain heat during extreme dives differs. Oceanic manta rays lack red muscle endothermy and visceral counter-current heat exchangers, which Chilean devil rays use to retain body heat during deep dives (Arostegui, 2024). This physiological limitation likely results in faster declines in body temperature in oceanic manta rays compared to Chilean devil rays, forcing them to spend less time at depth. Similar relationships between endothermic capacity and dive duration have been documented in other marine species. For instance, regional endothermy in bigeye tuna (Thunnus obesus) and swordfish (Xiphias gladius) enables prolonged deep-water foraging, whereas strictly ectothermic species like yellowfin tuna (Thunnus albacares) exhibit shorter, thermally constrained dives (Graham and Dickson, 2004). Given these differences, the shorter extreme dives observed in oceanic manta rays may necessitate alternative behavioural adaptations – such as breath-holding and prolonged pre- and post-dive warming periods—to mitigate rapid cooling and enable deep-water excursions. The significantly longer non-diving periods before and after extreme dives may represent behavioural thermoregulation, allowing manta rays to warm their bodies before and after encountering temperatures below 5 °C at depth. Such a function of these surface intervals could be validated using muscle-implanted temperature sensors, similar to studies of blue sharks (Carey et al., 1990). In addition to limiting dive duration, these physiological constraints may also contribute to the relative infrequency of extreme dives in oceanic manta rays compared to the congeneric Chilean devil ray, which routinely performs deep dives (Thorrold et al., 2014). This discrepancy may be attributed to the higher energetic and thermal costs faced by oceanic manta rays, which lack red muscle endothermy and visceral heat exchange mechanisms (Arostegui, 2024). It is possible that a single extreme dive is sufficient to fulfil specific ecological functions, such as environmental sampling or navigation, thereby reducing the necessity for repeated extreme dives over short time scales.

In addition to thermal challenges, extreme dives expose oceanic manta rays to substantial variations in dissolved oxygen (DO), including regions below the oxygen minimum zone (Dagorn et al., 2000; Roche et al., 2013; Rummer et al., 2014). Notably, the depths of the first horizontal steps during extreme dives consistently occurred within the midwater oxygen maxima located above the oxygen minimum layer and below the thermocline. The depth of the midwater O₂ maxima differs by ~300 m between Indonesian and New Zealand sampling sites data sets, which is consistent with differences in the depth of the commencement of the first dive step by manta rays in these two sites. It seems possible that the first step in the descent phase may serve as a reoxygenation break, allowing oceanic manta rays to maximise oxygen uptake before continuing their descent. In such conditions, oceanic manta rays may swim horizontally at a speed that optimises ventilation efficiency while minimising metabolic costs, similar to behaviours observed in deep-diving species like southern bluefin tuna (Thunnus maccoyii) and various ram-ventilating sharks (Carlson and Parsons, 2001; Fitzgibbon et al., 2010; Domenici et al., 2013). Alternatively (or in addition), given the velocities at which animals are swimming, it is conceivable that oceanic manta rays are using anaerobic metabolic pathways during descent and may require rest to process some of the lactate produced by this exercise. Both strategies, however, would likely expose manta rays to rapid cooling and are inconsistent with a ‘breath holding’ thermoregulatory behaviour. Further investigations into the internal temperature dynamics of oceanic manta rays and concentrations of dissolved oxygen at depth could provide useful insights into the functional significance of these steps in patterns of descent.

Oceanic manta rays exhibited distinct regional differences in movement patterns and the occurrence of extreme dives. Given the known limitations of light-based geolocation (Lisovski et al., 2012), we applied rigorous quality control to improve track reliability. Individuals tagged in New Zealand undertook long offshore movements, with all oceanic manta rays leaving the continental shelf and travelling into the tropical South Pacific, including movements towards Fiji and Tonga. These movements in New Zealand coincided with seasonal shifts in sea surface temperature, particularly during the onset of the austral winter. In Peru, only one individual moved far offshore, westward to the Galápagos Islands. Oceanic manta rays in Indonesia remained predominantly within the country’s exclusive economic zone (EEZ), moving among islands within the archipelago. The frequency of extreme dives mirrored these spatial differences. Extreme dives were most frequently recorded in New Zealand, accounting for 71 of 79 total events, with rays often initiating extreme dives within a day of leaving the continental shelf. In Peru and Indonesia, extreme dives were infrequent, corresponding with most rays remaining in shallower coastal habitats. However, consistent with the pattern in New Zealand, one individual in Peru performed an extreme dive immediately upon leaving the continental shelf. In Indonesia, limited access to offshore environments likely constrained opportunities for extreme dives. These patterns highlight both individual and regional variability in the frequency of extreme dives among oceanic manta rays, similar to that seen in the shortfin mako shark, which adjusts its vertical habitat use depending on local ocean conditions (Vaudo et al., 2016).

Our mixed-effects model identified distance from the shelf edge as a significant predictor of extreme dive occurrence, supporting the observed patterns. Extreme dives were more likely as manta rays moved offshore into deeper waters. The median interval between extreme dives was two days (IQR 3), suggesting these energetically costly events are possibly tied to broader movement decisions rather than immediate environmental conditions. High individual variation in daily maximum dive depth was also observed across and within study sites, consistent with patterns documented in Chilean devil rays, shortfin mako sharks, and yellowfin tuna (Schaefer et al., 2011; Thorrold et al., 2014; Vaudo et al., 2016). However, this variability in maximum dive depth did not correlate with bathymetry, suggesting that extreme dives are not bathymetrically constrained.

Although the precise function of extreme dives remains unclear, their rarity, timing, and characteristics collectively suggest that they are unlikely to be undertaken purely for short-term energetic benefits such as foraging, thermoregulation, predator avoidance, or energy conservation. The most compelling evidence comes from the consistent association between extreme dives and subsequent increases in horizontal movement, with rays frequently travelling over 200 km in the 72h following extreme dives. This post-dive movement supports the hypothesis that extreme dives serve an informational role, enabling manta rays to gather environmental cues that guide broader-scale movements in offshore habitats. During extreme dives, oceanic manta rays did not prolong bottom time or display foraging-associated behaviours, with these extreme dives comprising ≤3% of recorded days, suggesting minimal direct energetic gain from prey resources. Although some deep-diving taxa, including shortfin mako sharks, Chilean devil rays, bigeye tuna, leatherback turtles (Dermochelys coriacea), and billfish (Xiphiidae), are believed to exploit mesopelagic resources or exhibit sensory adaptations for deep foraging (Fritsches et al., 2000, 2005; Schaefer and Fuller, 2002; Crognale et al., 2008; Thorrold et al., 2014; Vaudo et al., 2016), the dive profiles observed here lacked the oscillatory patterns commonly associated with feeding. Additionally, the maximum depths observed exceeded known distributions of pelagic prey layers in the study regions (Batzler, 1975; Schalk, 1987). The normal distribution of deep and extreme dives around noon may suggest a connection with light levels, as seen in other taxa. For example, bigeye tuna and Chilean devil rays display increased deep-diving activity during midday, coinciding with peak light penetration (Schaefer and Fuller, 2002; Thorrold et al., 2014). Similarly, billfish exhibit visual adaptations, including colour vision and ocular endothermy, that enhance prey detection in deep, low-light environments (Fritsches et al., 2000, 2005). However, in oceanic manta rays, this midday pattern may alternatively reflect diel preferences for other behaviours, such as surface foraging at night on diel-vertically migrating prey (Hays, 2003) or energy conservation when prey availability is lower.

The substantial energetic cost associated with steep, rapid descents, combined with prolonged warming intervals near the surface before and after dives, further argues against functions related primarily to thermoregulation or energy conservation. Likewise, predator avoidance seems improbable, as extreme dives showed no indication of the fast-escape trajectories expected during a predation event (Domenici and Blake, 1997), included horizontal steps, and maximum depths remained accessible to potential predators, for example, tiger sharks (Galeocerdo cuvier) and killer whales (Orcinus orca) (Afonso and Hazin, 2015; Towers et al., 2019). In other marine megafauna, such as leatherback turtles, deep dives during inter-nesting periods are interpreted as antipredator behaviour in coastal environments (Asada et al., 2022), but the offshore and directed nature of manta ray movements following extreme dives suggests a different function.

Instead, the distinct profile of extreme dives—characterised by stepped descents, shortened bottom phases, and a markedly greater horizontal movement—suggests extreme dives may serve an informational function that aids long-distance navigation or habitat selection. Gathering information on environmental gradients, including temperature, dissolved oxygen, and potentially geomagnetic cues, could help oceanic manta rays orient and decide whether to remain in an area or disperse to new foraging grounds. Similar hypotheses have been proposed for scalloped hammerhead sharks, which use seamount-associated geomagnetic anomalies for navigation (Klimley, 1993). In sea turtles, controlled experiments have demonstrated that hatchlings can detect magnetic intensity differences and use them for regional positioning (Lohmann and Lohmann, 1996), whereas adults likely rely on bicoordinate magnetic maps—incorporating both inclination angle and field intensity—to navigate across ocean basins (Lohmann et al., 2008). Experimental evidence from bonnethead sharks (Sphyrna tiburo) further supports the ability of elasmobranchs to use geomagnetic fields as positional cues, showing that individuals can orient towards their home region when exposed to a magnetic field mimicking a distant location (Keller et al., 2021). Whether oceanic manta rays similarly sample magnetic field gradients remains unknown. Geomagnetic anomalies in the study regions have not been mapped. Additionally, as magnetic orientation in sea turtles occurs without deep diving (Lohmann et al., 2008), it is uncertain whether diving enhances magnetic sensing in oceanic manta rays. Other rays, such as yellow stingrays (Urobatis jamaicensis), are hypothesised to possess compass-like orientation abilities based on magnetic cues (Newton and Kajiura, 2020). Consequently, although geomagnetic sampling remains a plausible hypothesis, further investigation is required to determine whether extreme dives in manta rays contribute to magnetic navigation or primarily serve to sample other environmental cues.

The clear association between these dives and subsequent increases in distance travelled supports the notion that extreme dives enable the rays to sample the vertical structure of the water column, informing their navigation in relatively featureless open-ocean environments. In this context, extreme dives may represent a key, albeit infrequent, component of an adaptive movement behaviour, serving as a potentially costly but important means to survey environmental conditions to guide patterns of broad-scale movement and habitat use.