Thirty-one PSATs (4 X-Tags; Microwave Telemetry, Inc. and 27 MiniPATs; Wildlife Computers Inc.) were deployed on striped marlin between 2016 and 2019 by Hawaii-based longline vessels operating in the Central North Pacific. Upon retrieval of a set, an individual that was assessed to be in good condition was tagged on deck (n = 2) or along the side of the boat via a custom-built tagging pole (n = 29). Hooks were either removed, or fishing line cut as close to the jaw as possible prior to release. Tag tethers and anchors were constructed according to materials and methods developed for bluefin tuna tagging (Lutcavage et al., 1999; Galuardi and Lutcavage, 2012), and have shown great efficacy on sailfish (Lam et al., 2016). The tagger implanted the black nylon “umbrella” dart through the musculature, into the pterygiophores at the anterior dorsal region. For fish tagged alongside the vessel, whole weight (W) was estimated by the vessel’s captain to the nearest five pounds (1 lb ~ 0.4536 kg). Eye fork length (EFL) was measured to the nearest centimeter for fish tagged on deck. W (in kg) was converted to EFL (in cm) or lower jaw fork length (LJFL, in cm) following Sun et al. (2011):

and

PSATs were programmed to record relative light level, external temperature, pressure (depth) for 12 month missions. Archived depth and temperature time series were sub-sampled to 5-minute (MiniPAT) and 15-minute (X-Tag) resolution by manufacturer routines for satellite transmission through the Argos-2 services, which limit each transmitted message to 32 bytes (W.M.O., 2018). Manufacturer routines also treated depth and temperature time series as separate data streams, so retrieved satellite messages might contain unsynchronized depth and temperature records (i.e., non-matching date-time stamps). In addition, satellite data retrieval does not guarantee delivery of all collected records, and the quantity of returned data can be highly variable (Musyl et al., 2011). Any breaks in the data recovery also create gaps in the time series. For MiniPAT, we programmed the tags to “always transmit” both depth and temperature time series, as enabling any “on-off” transmission schedule (i.e., duty-cycle) would exacerbate data mismatches in time.

All tags had a failsafe release if the tag experienced any constant depth for 3 (X-Tag) or 4 (MiniPAT) days, a hallmark of post-release mortality or tag shedding. MiniPAT firmware also included detection and transmission of a hardware failure message (“pin broke”) indicating that the tag’s nosecone pin had broken, separating the tag from its tether, causing premature release. Once a tag popped off and began transmitting, positions with Argos location classes LC 1, 2, and 3 were noted, and the first of these positions was selected as the tag’s reporting position.

Returned data were imported into and managed through Tagbase (Lam and Tsontos, 2011). Predation-associated impacts did not damage any of the released tags or its antenna, and provided data for deducing the plausible fate of the tagged individual. Predator interaction was inferred if sensor data indicated ingestion, including sustained low light levels, increased temperature indicative of a warm-bodied predator, and depth pattern changes.

Transmitted light data represented either truncated light curves around (MiniPAT) or detected time stamps of (X-Tag) sunrise and sunset. From this light information, geolocation was only performed on fish carrying tags for over 14 days (Lam et al., 2022). Protocols are detailed in Appendix 1. We defined biologged depth and temperature records as day or night, based on sunset and sunrise times recorded by X-tags and estimated for MiniPATs by the geolocation routine, TrackIt (Lam et al., 2010). From the estimated latitudes, depth and temperature records were aggregated at 10° bands to generate summary statistics.

Lunar phase and fraction illuminated were obtained with R package, suncalc (github.com/datastorm-open/suncalc). We then used linear mixed-effects models to investigate the effects of fraction of the moon illuminated on the interquartile range or maximum of daily nighttime depth, using individual fish as a fixed effect. Models were built by R package, nlme (cran.r-project.org/web/packages/nlme/index.html).

Since data were only returned via satellites (i.e., no downloaded archive from a physically recovered PSAT), gaps in the depth and temperature time series were unavoidable. To reduce the influence of missing or incongruous data on the construction of depth-temperature profiles and percentage time-at-depth histograms, we aggregated time series records at 5-day intervals for each MiniPAT. Within a given 5-day period, a regression spline was fitted through the daytime depths and temperatures using R package, splines (Perperoglou et al., 2019). This 5-day window was chosen after experimenting with various time intervals (3-14 d) to produce smoothed depth-temperature profiles. For percentage time-at-depth histograms, depth records were binned at 10-m intervals by day and night. Based on the depth distribution relative to a corresponding depth-temperature profile, we visually determined if there was a prominent use of the top 10 m, mixed layer and thermocline during a given 5-day period by the tagged marlin. Cluster analysis and other automated classification techniques were applied, but failed to isolate meaningful patterns, likely due to the irregular presence of gaps in the time series. Any visually identified depth pattern was then investigated with linear mixed-effects models. Possible effects of month, 10°-latitude band and lunar illumination on the frequency of occurrence of a depth pattern were examined using individual fish as a fixed effect. Lastly, two measures of vertical activity, in addition to mean depth, were generated hourly for each tagged fish: vertical cumulative distance (Vdist) is the sum of all unsigned depth changes between successive time points, and vertical net displacement (Vplace) is the sum of all signed depth changes between successive time points. A large Vdist reflects a fish that frequently swam up and down in the water column, while a positive Vplace means at the end of an hour, the fish was at a deeper depth than at the beginning of the hour.

To aid the interpretation of fish movement, the following geographical information system resources were accessed: bathymetric values from Smith & Sandwell Topography (0.0167° resolution, version 11.1), sea surface height (variable ‘zos’) and temperature (variable ‘thetao’) monthly mean values [product ID GLOBAL_REANALYSIS_PHY_001_030] from the Copernicus Marine Service Global Monitoring and Forecasting Centre (CMEMS; marine.copernicus.eu). Additionally, daily sea surface temperature (SST) [product ID METOFFICE-GLO-SST-L4-NRT-OBS-SST-MON-V2]. Mixed layer depth (MLD) was extracted along reconstructed tracks from NCEP Global Ocean Data Assimilation System monthly data (psl.noaa.gov/data/gridded/data.godas.html), which defines MLD as the depth where the buoyancy difference with respect to the surface level is equal to 0.03 cm s^-2.

With the understanding that PSATs do not identify or confirm spawning, we evaluated changes in vertical movement (mean depth, Vdist, Vplace) that might be associated with spawning habitats during the spawning season. Since it wasn’t possible to assign reproductive status of released fish, we referenced the maturity ogive developed by NOAA Pacific Islands Fisheries Science Center (PIFSC) observer sampling program (Humphreys and Brodziak, 2019; Humphreys and Brodziak, 2021). PIFSC determined that active spawning females (n = 37, 151-188 cm EFL) were concentrated in a T-shaped area (between 25-30°N & 150-165°W, and 20-25°N & 155-160°W; see Appendix 1) from May to July. The boundary of this active spawning area was included in our analysis of vertical movement with respect to life history requirements.

From cumulative depth distributions (Figure 1), striped marlin spent 38 and 81% of their day and night, respectively, in the top 5 m. Swimming depths were 90% encapsulated within 120 and 37 m during day and night, respectively. Their median (interquartile range, IQR) daytime and nighttime depths were 44 m (2-82) and 2 m (0-3), respectively, with significant differences in diel depth and temperature occupancy (two-sample Kolmogorov–Smirnov test, p < 0.0001). Linear mixed effects models found no significant relationship between fraction of the moon illuminated and daily nighttime depth IQR (p = 0.06) or maximum depth (p = 0.79). The majority of mixed layer depths associated with striped marlin tracks were between 46 and 83 m (50-90% of all available records). Median daytime temperature was 23.3°C (IQR = 20.8-24.5) and 24.6°C (IQR = 23.6-25.4), nighttime. Tagged marlin experienced SSTs of 16.9-30.9°C, mirroring their nighttime temperature range. However, during the day a shift to cooler temperatures occurred, in which 90% of records fell within 12-26.7°C. Pooling all data, striped marlin experienced ambient temperatures between 7.6 and 31.8°C (Table 1).

Even as fish remained near the surface throughout the day they showed marked diel shifts in depth distribution and temperature across their ranges (Figure 2; Table 3). In daytime, tagged fish ventured to the top of the thermocline, except between 40 and 50°N, where temperatures experienced were cooler, near 19°C (Figure 2). A reduction in vertical habitat at 40-50°N was reflected by a shallow mean depth (7.9 m) and the smallest depth IQR (9.5 m, Table 3). In 10-30°N where most tagged fish ventured, they spent 24-34% of their time below the mixed layer (mean: 46-69 m; Table 3), reaching 100-150 m (Figure 2). Below 10°N, they reduced their time in the top 5 m, and shifted daytime occupancy to 40-60 m, right above the base of the mixed layer, where temperatures were at least 0.5-1°C cooler than at the surface (Figure 2). Mean mixed layer depth varied between 12 and 91 m across striped marlin’s range (Figure 2), with MLD being the shallowest at latitudes >30°N and deepening southwards (Table 3). On average, in the Central North Pacific striped marlin spent 27% of their time below the mixed layer (Table 3). Pairwise comparisons of logged depth and temperature records via Wilcoxon rank sum test revealed significant differences (p< 0.001) among all 10°-latitude bands.

Maximum depth recorded among all tagged marlin was 472 m (Table 1), but daily maximum depths varied with location (Figure 3), with fish reaching depths >200 m over a large area along Molokai Fracture Zone between 120-150°W, and south of 10°N. In contrast, north of 35°N, their vertical habitat was compressed, maximum depths were mostly above 100 m (Figure 3), depth IQR was narrower than those at the tropics (Table 3), and they spent little time below 80 m (Figure 2). During the first-ever documented dispersal to the Southern Hemisphere (Lam et al., 2022), between 5 and 20°S, PG01 logged a series of daily maximum depths >200 m during January 2017 (Figure 3), while elsewhere, its mean daily maximum depth was 141 m (Table 4). At that time PG01 had the largest latitudinal range (19.6°) (Table 4), and the highest daily movement rate (90 km day^-1) among all tagged fish.

Striped marlin exhibited three common patterns in their swimming depths (Appendix 2). Pattern 1 is characterized by a fish staying near the sea surface for extended periods, which occurred more often during the night (Figures 4, 5A). Pattern 2 features a high occupancy of the thermocline during day and night, either through brief (sometimes repetitive) trips from the surface or oscillatory movements within the thermocline (Figures 4, 5B, D). Pattern 3 is marked by multiple trips to different depths within the mixed layer, from the surface or ascents from the thermocline (Figures 4, 5A, C, D).

For daytime, forays into the thermocline (Pattern 2) and occupancy near the surface (Pattern 1) were most frequent, and with lower frequency, movements within the mixed layer (Pattern 3) (Appendix 3). Interestingly, during the day striped marlin often spent little time at the surface when in fracture zones and the vicinity of seamounts (Figures 4, 5D). For example, before PG31’s light sensor failed in late December, it had transited multiple geological features (Appendix 1), including Geologist Seamounts, Northwestern Hawaiian Islands, Molokai Fracture Zone, Murray Fracture Zone, and Clarion Fracture Zone over seven months. Upon arrival at each area, PG31 reduced its time near the surface and spent the majority of daytime in the thermocline. This pattern disappeared when PG31 moved to an area lacking prominent geological features (Appendices 2, 3 ). For nighttime, striped marlin utilized the surface (Pattern 1) and mixed layer (Pattern 3) (Appendix 3). Striped marlin sometimes frequented the thermocline during night (Pattern 2) (Figures 4, 5B) during all lunar phases (Appendix 3). Linear mixed-effects models did not find any significant relationship between the identified depth patterns (during day or night) with month, latitudinal band or lunar illumination.

Together, these depth patterns and swimming behaviors were often maintained for 1 to 2- weeks before marlin switched to another behavior (Appendix 2). For example, for two weeks in mid-December, PG24 initially stayed near the sea surface, making few descents (Figure 6A) but near the Molokai Fracture Zone, it switched to depths below the thermocline (150-240 m) while heading west along the fracture (Figure 4B). Its mid-January hourly vertical net displacement (Vplace) ranged between -20 to +20 m, i.e., movements began and ended at similar depths (Figure 6B). A corresponding high vertical cumulative distance (Vdist) signaled repeated descents and ascents, i.e., a “yo-yo” or oscillatory pattern. Between early and mid-February, this fish transited the Wini Seamount and southeast of the Puna Ridge where it descended to the bottom of the mixed layer (~120 m) and to the thermocline. In late February PG24 spent daylight just below the mixed layer, around 100 m (Figures 4, 6C), and then extended to 120 m in mid-late March (Figure 4D) near the Bryan Seamount. There, a high Vdist was a result of trips from the mixed layer to ~50 m and back (Figure 6D), while over the entire track, Vplace generally delineated dawn descents and dusk ascents.

A fish’s presence on spawning grounds during documented peak spawning, together with favorable environmental conditions (e.g., 24°C SST) suggests that reproductive behavior is likely. For instance, in early June PG14 (157 cm EFL) was positioned at the edges of a cyclonic eddy on the spawning ground northeast of the MHI (Figure 7A). While there, PG14 shifted from using the mixed layer during the day to staying near the sea surface in >24°C for about one week (Figure 7B) before reverting to diel vertical behavior (Figure 7C). Eight other individuals (Appendix 1) also transited the spawning ground between May and June, but did not provide sufficient time series data for a similar analysis of their vertical behavior.

Beyond foraging, the function of exemplary, deep “dives” undertaken by marlin and tuna is not well explained, nor are all the sensory systems that might be appropriated for guiding information gathering in darker, cooler ocean depths. Curiously, PG01 reached depths >200 m during its rapid transit through the West Pacific Warm Pool (~0-20°S; Wyrtki, 1989 but not elsewhere. The coarse resolution of light-based geolocation positions of PSATs (Galuardi and Lam, 2014) had precluded a detailed exploration of PG01’s deeper forays with environmental and bathymetric features. However, descents ≥250 m were also observed in four satellite-linked striped marlin tagged off New Zealand that abruptly reversed travel directions when they reached 20-21°S and increased transit speed away from the area (Sippel et al., 2011). The authors suggested that such a reversal, an apparent avoidance, may contribute to the long-established fishery pattern of a low catch-per-unit-effort between 10-20°S (Squire and Suzuki, 1990). Similarly, PG01’s elevated movement rate and deeper descents in the same region ( Table 4 ) provided another possible example of shared movement dynamics contributing to the observed spatial structuring in catch distribution.

Periodic deep descents also occur in Atlantic bluefin tuna, which make deepest “bounce” dives on the shelf at dusk and dawn (Lutcavage et al., 2000) and descents >500 m at the Gulf Stream margins (Galuardi et al., 2010). Deep descents plausibly serve a navigational purpose: magnetic orientation was proposed in pioneering tracking studies of shark and tuna (Carey and Scharold, 1990; Klimley, 1993), and supported by studies of potential bio-physical sensory mechanisms via brain magnetite particles (Walker et al., 1997; Walker and Dennis, 2005). More recently, Horton et al. (2017) presented compelling evidence in satellite-tracked marine mammals and great white sharks that gravitational and magnetic cues detected through variations within the water column and from bathymetric features (e.g., seamounts, ridges) can be integrated with astronomical cues (e.g., lunar illumination) detected near or at the sea surface to form a navigational system. Tuna, billfish and shark navigate with surprising accuracy, and return to apparently favored locations after crossing vast ocean distances (Carey and Scharold, 1990; Galuardi et al., 2010; Keller et al., 2021). This extraordinary phenomenon points to a long-standing gap in scientific understanding, especially that of neuro-sensory mechanisms (Dudchenko and Wallace, 2018; Watts et al., 2018), memory (Mouritsen, 2018) and collective behavior (Couzin, 2018). Although speculative at present, vertical transits of the water column can represent an important, yet untapped dimension of vertebrate sensory systems, and have potential to expose navigational secrets of highly migratory fishes. These sensory capabilities may remain undetected if only horizontal movement and migration are considered (Martin, 2018).

Our successful partnership with fishermen collaborators has enabled new insights into water column utilization of striped marlin across vast stretches of the open ocean where they could likely interact with longline gear of the US and international fleets. Occupancy of different parts of the water column by time of day (e.g., Figure 1 ) can be incorporated in models of fishery catch-per-unit-effort data to derive indices of relative abundance (Maunder et al., 2006; Johnson et al., 2019), or be used to evaluate gear vulnerability by capture time (Boggs, 1992; Rodrigues et al., 2022). In designing effective management measures, authorities must strike a balance between the economic benefits of fishing harvest, species conservative priorities and climate impacts (Mckuin et al., 2021). Mandatory live release is a mitigation measure applicable to both the shallow-set and deep-set longline fishery, on the account of a high survivorship (Pine et al., 2008). In this study, only two tagged fish died within 4 days of release ( Table 1 ), similar to other PSAT-tagged striped marlin released off New Zealand by the recreational fleet (Sippel et al., 2011). Our vertical data records confirmed that striped marlin regularly frequented the top 5 m of the water column, and 90% of their swimming depths were above 120 m. Additionally, at higher latitudes (>30°N), striped marlin were mostly located above 40 m ( Figure 2 ), so removing shallow (<120 m) hooks from deep-set longline in such locations might reduce striped marlin fishing mortality. While economic impacts of hook removal for bigeye tuna caught at deeper depths appeared minimal (Boggs, 1992; Beverly et al., 2009), similar cost-benefit analysis and potential spillover effects among multi-national fleets (Chan and Pan, 2016) need to be re-examined as spatiotemporal changes in the vertical habitat of striped marlin were common. Consequently, there may be advantages to specifying time- and area-based conservation measure (Jensen et al., 2010; Hilborn et al., 2022; Pons et al., 2022), such as applying particular mitigation approaches in the main spawning area during peak spawning months. As vertical and horizontal movement preferences of striped marlin and other bycatch species are further resolved, management and conservation objectives could be achieved by deploying more targeted measures, with input of the fleets and fisheries experts in regard to the best practices (Hare, 2020).

Most of our tagged fish were close to length at 50% maturity of 153.6 cm EFL (Humphreys and Brodziak, 2021), and larger than the smallest spawning female (146 cm) in Humphreys and Brodziak (2019). Given their fast growth rate at this age range (Kanaiwa et al., 2020), it’s reasonable to project that tagged individuals would reach spawning-capable size during their tracked time-at-liberty. While it’s impossible to directly observe reproductive behaviors given current tag hardware and geolocation limitations, we postulate that prolonged departures from diel swimming, to a pattern of extended time spent near the surface on a spawning ground, during the spawning season, may signal a behavioral switch from foraging to reproduction (Figure 7). While we couldn’t explicitly detect frequency of potential spawning bouts from our data, if surface occupancy were related to spawning or courtship, events would likely repeat on striped marlin’s average spawn of every 3-10 days (Kopf et al., 2012; Chang et al., 2018), and at least 27-90 times over a single season (Kopf et al., 2012). Multiple periods of surface occupancy were observed for PG14 throughout June and July, especially when it traveled along the edges of areas with an elevated sea surface height (Appendix 4). Common knowledge among fishers and fish buyers on the auction floor is that some striped marlin appear with orange colored, high-quality, lipid-rich flesh (Figure S1) between April and June, which may reflect primed somatic condition just prior to the start of the spawning season, or a unique prey choice during this time of the year. Notably, the highest Fulton’s condition factor (K) was recorded in April in mature female striped marlin off Taiwan, prior to the peak spawning months of June and July (Chang et al., 2018). Collectively, these reproductive characteristics provide multiple lines of evidence that some of the tagged fish could be spawning-capable, or approaching spawning during their tracked duration.