The Distribution of Giant Manta Rays In The Western North Atlantic Ocean Off The Eastern United States

In 2018, the giant manta ray (Manta birostris) was listed as threatened under the U.S. Endangered Species Act. We integrated decades of sightings and survey effort data from multiple sources in a comprehensive species distribution modeling (SDM) framework to evaluate the distribution of giant manta rays off the eastern United States, including the Gulf of Mexico. Manta rays were most commonly detected at productive nearshore and shelf-edge upwelling zones at surface thermal frontal boundaries within a temperature range of approximately 15–30 °C. SDMs predicted high nearshore concentrations off Northeast Florida during April, with the distribution extending northward along the shelf-edge as temperatures warm, leading to higher occurrences north of Cape Hatteras, North Carolina from June to October, and then south of Savannah, Georgia from November to March as temperatures cool. In the Gulf of Mexico, the highest nearshore concentrations were predicted near the Mississippi River delta from April to June and again from October to November. SDM predictions will allow resource managers to more effectively protect manta rays from sheries bycatch, boat strikes, oil and gas activities, contaminants and pollutants, and other threats. Aerial line-transect surveys were conducted along the U.S. Gulf of Mexico and Atlantic coast between Texas and New Jersey by the National Marine Fisheries Service SEFSC between 2010 and 2019. These aerial surveys were primarily designed to estimate the abundance of marine mammals and sea turtles in continental shelf waters. Surveys were conducted once per season in the Gulf of Mexico during 2011– 2012 as part of the Natural Resources Damage Assessment associated with the Deepwater Horizon oil spill (SEFSC-GOMNRDA, Garrison 2017), and three surveys of the Gulf of Mexico were conducted during 2017–2018 as part of the Gulf of Mexico Marine Assessment Program for Protected Species (SEFSC-GoMMAPPS; Garrison et al. 2021) and covered the continental shelf and the inner continental slope from Texas to southwest Florida. Along the U.S. Atlantic coast, similar surveys were own covering all four seasons between 2010–2019. These surveys were conducted as part of the Atlantic Marine Assessment Program for Protected Species (SEFSC-AMAPPS; Palka et al. 2017) and covered the continental shelf and the inner continental slope from southeast Florida to New Jersey. All SEFSC surveys were conducted aboard a DeHavilland DHC-6 Twin Otter and were own at an altitude of 182 m and an airspeed of 185 km/hr. Surveys were typically own during favorable sighting conditions at Beaufort sea states ≤ 4 (surface winds < 12 knots). Two independent teams of visual observers searched for marine mammals and sea turtles from directly beneath the aircraft out to a perpendicular distance of approximately 600 m from the trackline. The aircraft included bubble windows and a belly window, allowing the area underneath the trackline to be observed. Team 1 (forward) consisted of two bubble window observers, each stationed at one side of the aircraft and one data recorder. Team 2 (aft) consisted of one observer on the right bubble and one on the belly window, in addition to a data recorder. The two-team conguration


Introduction
Manta rays are lter-feeding rays in the family Mobulidae , characterized by a terminal mouth, diamond-shaped bodies with wing-like pectoral ns, and long cephalic ns (Romanov the environmental drivers of their distribution will allow researchers to more e ciently locate these rare animals for scienti c study and potentially predict changes in their distribution under climate change scenarios. Finally, SDMs provide managers a better understanding of the factors in uencing giant manta ray movements and habitat use, facilitating more effective timing and coordination of conservation efforts.

Sources and Survey Data Processing
Few dedicated surveys for mantas exist in the EUS; however, due to their large size and distinct appearance, they are often observed and recorded during visual aerial surveys that target marine mammals and sea turtles. To better characterize the distribution of mantas in EUS waters, we assembled a comprehensive geographic information system (GIS) database of manta sightings from peer-reviewed literature, survey databases, gray literature reports, and anecdotal sources (e.g., social media, press reports, personal communications). A comprehensive internet search revealed many photos and videos of manta sightings from news outlets and citizen scientists on Instagram, Facebook, YouTube, and Twitter. Publishers of sightings content were then contacted and asked about manta size, location, and date of the sighting. Additional anecdotal information was received through NOAA's reporting email: manta.ray@noaa.gov.
Photos and informal interviews with observers were used to assess the reliability of identi cation records from surveys and grey literature (D. Adams Jones). On Southeast Fisheries Science Center (SEFSC) surveys, M. tarapacana were identi ed to species; however, there was a potential for misidenti cation of M. mobular. To avoid overestimation of manta ray abundance, sightings and effort north of 35° N were excluded from the SEFSC and NARWC surveys for species distribution modeling efforts. All purported "manta ray" sightings from both surveys were retained for external validation of the models and for detection function development, given similarities in size between mantas and other large mobulid species. Correct identi cation was con rmed north of 35° N for New York State Energy Research and Development (NYSERDA) surveys as discussed below. SEFSC and NARWC observers did not note similar concerns south of Cape Hatteras. Photo archives and discussions with Georgia Aquarium observers veri ed > 1500 mantas off northeastern Florida. A supplemental review of photographic archives collected during Normandeau Associates Bureau of Ocean Energy Management aerial surveys suggested that < 30% of SEFSC and NARWC aerial survey "manta ray" sightings from North Carolina to South Carolina might be M. mobular (J. Robinson Willmott and C. Horn, unpublished data). Similarly, a supplemental review of photographic archives from Florida Fish and Wildlife Research Institute (FWRI) North Atlantic right whale aerial surveys own in winter (December-March) from approximately Savannah, Georgia (31.93° N) to Cape Canaveral, Florida (28.67° N) revealed 3 of 85 (3.5%) sightings identi ed as "manta rays" were M. mobular, with no recorded sightings of M. tarapacana, suggesting a very low potential misidenti cation rate with other large mobulids for aerial surveys conducted in that region (J. Jakush, pers. comm.). Photographs and videos were used to verify the accuracy of species identi cation from anecdotal sources. All SEFSC surveys were conducted aboard a DeHavilland DHC-6 Twin Otter and were own at an altitude of 182 m and an airspeed of 185 km/hr. Surveys were typically own during favorable sighting conditions at Beaufort sea states ≤ 4 (surface winds < 12 knots). Two independent teams of visual observers searched for marine mammals and sea turtles from directly beneath the aircraft out to a perpendicular distance of approximately 600 m from the trackline. The aircraft included bubble windows and a belly window, allowing the area underneath the trackline to be observed. Team 1 (forward) consisted of two bubble window observers, each stationed at one side of the aircraft and one data recorder. Team 2 (aft) consisted of one observer on the right bubble and one on the belly window, in addition to a data recorder. The two-team con guration allows for the estimation of detection probability on the trackline following approaches described in Laake and Borchers (2004). The aircraft location, survey effort status, and viewing conditions (e.g., sea state, glare, visibility) were recorded every 10 seconds using a data-logging program. Observers updated viewing conditions whenever needed and generally after turns into a new trackline. Upon sighting a marine mammal, sea turtle, or other target of interest (including manta rays), the observer measured the angle from the vertical to the animal (or group) using a digital clinometer. This sighting angle, θ, was converted to the perpendicular sighting distance from the trackline (PSD) by PSD = tan(θ) × Altitude. While manta rays were not the primary focus of the surveys, observers were instructed to record all large sh sightings (including rays, sharks, tuna, etc.) and collected angles for estimating sighting distances wherever possible.

Southeast Fisheries Science Center (SEFSC) Aerial Surveys
Sightings and effort data were combined for all SEFSC aerial surveys. An initial investigation of the distribution of PSDs indicated a reduction in the number of sightings very close to the trackline for Team 2, and therefore Team 2 data were truncated at the minimum distance that manta rays were observed and sightings between 0 and 3.2 m were removed from the analysis. Sighting angles were determined for both observer teams based on side (i.e., left, right) and position (e.g., belly, bubble) of each recorded sighting. Because the survey was not speci cally designed for mantas, sighted individuals were not assigned unique identi ers. Using forward team sightings as a reference, aft team sightings were matched to forward team sightings when an equal number of animals were recorded by the aft team within 15 seconds, on the same side of the aircraft, and with an angle difference of < 15 degrees. Any sightings where the aft team could not have seen the animal due to the sighting angle recorded by the forward team were eliminated from the detection function analysis. Based on histograms and quantiles of sightings distance, the right truncation distance was set at 300 m (Supplemental File 1).
Effective search effort for mantas was determined in a mark-recapture distance sampling framework using package 'mrds' in R (Laake et al. 2020) for "On Effort" sightings by the forward and aft survey teams (Supplemental File 1). The probability of detection and effective area searched were derived using the independent-observer approach assuming point independence ( Akaike 1973). Fitting of the detection model considered all possible permutations of covariates that may in uence detection probability in the surveyed strip with MCDS and detection probability on the trackline with MRDS, including Beaufort sea state, cloud cover, glare intensity (level of visual obstruction due to sea surface glare), glare coverage (proportion of viewing area obstructed), and turbidity, along with interactions between distance and observer in the MRDS function (Supplemental File 1). All combinations of variables were considered for inclusion, and the best model was selected from the candidate models based on the lowest AIC. For a given trackline segment, search effort was expressed as the multiple of trackline length, estimated detection probability within the strip, and the truncation distance. Availability bias due to diving behavior was not addressed, although ongoing satellite tagging efforts are seeking to resolve this issue.

North Atlantic Right Whale Consortium (NARWC) Surveys
The North Atlantic Right Whale Consortium's sightings database serves as a repository for sightings of marine mammals, sea turtles, and large shes, as well as for corresponding survey effort where available (North Atlantic Right Whale Consortium 2018). Survey platforms and protocols for recording manta sightings vary across the many contributors that submit data to the database. Therefore, most contributed datasets were used only for external validation of models, including sightings north of Cape Hatteras, vessel-based surveys, etc.). Surveys conducted by the New England Aquarium in a Skymaster airplane during November-March/April in 1989/90, 1990/91, and 1991/92 were the only surveys that reliably recorded sighting distances for mantas. These surveys were primarily conducted off Florida and Georgia during winter (November-April). Sightings distances for these surveys were reported in nautical mile intervals corresponding to wing strut markings on the survey platform and were converted to meters.
For distance function tting, sightings were restricted to on-effort sightings at altitudes of ≤ 366 m with Beaufort sea states of ≤ 4. AIC was used to guide selection of the best-tting detection function considering possible covariates of sea state, cloud cover, and glare using function 'ds' in the R 'Distance' package (Miller et al. 2019).
This detection function (Supplemental File) was then applied to estimate effective search area for oneffort surveys by FWRI from 2002 and 2010-2017, which were also conducted in a Skymaster and consistently recorded manta sightings but did not record detection distance. FWRI surveys were also primarily conducted off Florida and Georgia during winter. Because departures from the trackline for North Atlantic right whale sightings were not explicitly coded as such for the FWRI surveys, and it was unclear whether mantas would be recorded during this activity, off-track effort was eliminated by dropping waypoints that deviated from the previous heading by > 20 degrees. Visual inspection of tracklines indicated this approach was effective at eliminating loops off the trackline. Sightings from all other surveys were retained for external validation of model ts, but not included in the distribution modeling input due to lack of clarity regarding whether mantas would have been explicitly recorded during all on-effort surveys and lack of data suitable for tting a detection function.

New York State Energy Research and Development Authority (NYSERDA) Surveys
The New York State Energy Research and Development Authority (NYSERDA) contracted with Normandeau Associates Inc. (Normandeau) and teaming partner, APEM Ltd., to use high-resolution, largeformat aerial digital imagery to collect data on birds, marine mammals, sea turtles, cartilaginous sh, and other taxa encountered offshore within the New York Offshore Planning Area, an area spanning > 43,745 km 2 . Transect surveys of abutting imagery were conducted four times a year over three years between August 2016 and May 2019, with an area of > 3,000 km 2 imaged on each survey representing > 7% coverage of the area. Each survey took between six and eight days to complete, depending on weather conditions and ight restrictions. Image resolution was 1.5 cm at the sea surface collected with downward-facing cameras from a ight altitude of 414.5 m and at a ight speed of 220 km/hr. Animal targets were extracted from imagery using a combination of detection software and manual review, with detected animals made available to taxonomists for species-level identi cation through Normandeau's ReMOTe data portal. All identi cations underwent a quality-control review of 20% of identi cations, Endangered species went through 100% review of identi cations, and 10% of all imagery considered not to contain animal targets also underwent manual review. For the purposes of this analysis, all large rays were further reviewed to con rm species identi cation.
Effort was expressed as the swept area within the camera view, and detection probability was assumed to be 100% within the swept area for animals at or near the surface. To estimate size of detected manta rays, a measurement tool was created within the data portal utilizing known pixel resolution at the sea surface within every portion of each image. Measurements of animals using known pixel resolution at the sea surface were unable to compensate for the unknown depth of the animal in the water column and consequently an element of minor (< 5 cm) but unquanti able error is associated within estimations of disc width for observed manta rays.

Distribution Modeling
Depth was assigned to transect segments from the National Oceanic and Atmospheric Administration (NOAA) National Centers for Environmental Information Coastal Relief Model (CRM), which provides 3 arc-second resolution bathymetry for most areas in the study domain. Data gaps were lled with 1 arcminute resolution bathymetry from the NOAA ETOP01 database using the R 'marmap' package (Pante and Simon-Bouhet 2013). Slope was derived from bathymetry using Spatial Analyst in ESRI ArcMap 10.7, with higher-resolution CRM-derived bathymetry and slope retained when available. Satellite observations of daily sea surface temperature (SST); daily k490 irradiance; and 8-day averaged chlorophyll-a (Chl-a), along with model-generated estimates of primary productivity; north-bound water velocity from HYCOM models; and predicted wave height from the Global Wave Model were assigned to daily transect segments from the ERDDAP server using the R 'rerddap' package (Chamberlain 2021). Frontal gradients of SST were computed following (Belkin and O'Reilly 2009) using the R 'grec' package (Lau-Medrano 2020). Daily standardized frontal gradients ('Front-Z') were computed by dividing frontal gradient raster values by the daily maximum within the raster domain. SST was evaluated because it was hypothesized that manta distributions would be limited by thermal tolerance. Primary productivity, k490 irradiance, Chla, and Front-Z are all proxies for upwelling zones with high prey abundance, with differing levels of availability and model resolution in the time series. North-bound water velocity was hypothesized to capture potential use of the north-bound Gulf Stream and nearshore counter-current ow off the U.S. east coast. Predicted wave height was hypothesized to capture manta avoidance of rough water where surface zooplankton concentrations were unlikely and surface basking would be bioenergetically costly.
Sightings, effort, and environmental parameters for daily tracklines were summarized to a 10×10-km grid for all surveys. The 10×10-km model domain encompassed all surveys from all sources. Trackline segments were assigned to grid cells, sightings and effort were summed within each grid cell, depth and environmental characteristics were averaged within each cell, and the maximum observed value for bathymetric slope on the trackline segment was retained. Generalized additive models (GAMs) were t to all possible permutations of bathymetry and environmental parameters using the 'mgcv' package in R (Wood 2011), with log-transformed effort derived from the survey-speci c detection functions as an offset. GAMs were t with a binomial distribution using a logit link function, such that the resultant models describe the probability of species presence, also termed "habitat suitability" (Brodie et al. 2018) or "habitat preference" (Hazen et al. 2017). To minimize effects of collinearity, correlated predictor variables were not included in the same model. For example, Chl-a, primary productivity, and k490 were highly correlated and were tested separately but never included in the same model. Because Front-Z was derived from SST rasters, an interaction term for SST and Front-Z was also tested. Models were t using a binomial distribution with tensor splines limited to 3 knots (Wood 2011). Preliminary model-tting showed a tendency for overprediction at the extremes (especially peak positive values) when extrapolating predictions to other months/years. Constraining to 3 knots eliminated this issue while preserving the functional relationships with environmental covariates. The best-tting model was selected by lowest AIC and compared to three competing GAM con gurations tiered off the best-tting GAM by excluding non-signi cant terms in the model summary. For each survey, the nal model was selected by comparing residual deviance explained and predictive power as evaluated through 10-fold internal cross-validation and external validation using independent sources. Cross-validation was conducted using the 'pROC' package in R (Xavier et al. 2011), generating estimates for area under the curve (AUC) and associated false positive and false negative rates for a model t to 90% of the available data compared to sighting locations from the remaining 10% of the data. External validation was similarly accomplished using pROC to evaluate AUC for the full GAM model compared to sighting records from independent sources, including off-effort sightings from the survey under evaluation. For external validation through AUC, sighting effort was unknown, and assumed equal to mean sighting effort from distance-sampling surveys. As such, external validation outcomes were less reliable for evaluating quality of model t but useful for relative comparison between models. An additional external validation metric was developed to gauge consistency between SDM predictions and the locations of independent sightings of manta rays. We divided SDM predictions for point-speci c independent sightings by domainwide daily median Z-score transformed SDM predictions across valid depths, then centered these transformed scores to zero by subtracting one. The greater the proportion of retained Z-scores above zero, the higher the consistency between SDM predictions and independent observations (Farmer et al.

2017, Heyman et al. 2019).
We t models to each survey independently following the approaches above. Due to the limited spatiotemporal coverage of distance-sampling data for the NARWC and NYSERDA data, those SDM results were not reported independently, but were used to generate ensemble models following two approaches. The rst ('weighted ensemble') was a weighted mean prediction across the best models t to the three independent surveys (i.e. SEFSC, NARWC, and NYSERDA). Model predictions for each 10-km cell were expressed as a weighted mean across the three surveys, with the complement of the standard error (i.e., 1-SE) of the model prediction from each survey used as the weighting term, to place more emphasis on the model prediction with the least uncertainty at that given location. The second ('combined surveys') was developed by tting a GAM using the iterative tting process described above to an appended data series of all three surveys combined, with survey-speci c sighting effort expressed in the same units of swept area (m 2 ).
Annual trends in predicted manta distributions were evaluated by tting previously described distribution models for SEFSC, NARWC, and combined data to monthly average environmental conditions from January 2003 to December 2019. The monthly weighted mean latitudinal centroid of predicted manta distribution was computed for each model. Seasonal auto-regressive integrated moving average (SARIMA) time-series models with annual and monthly differencing terms were t using R package "astsa" (Stoffer 2020, Shumway & Stoffer 2017). As no annual terms were included in the model, our SDM assumed stationary overall probability of occurrence across years; thus, any interannual differences in mean probability of occurrence would be driven only by differences in the dynamic terms in the model (e.g., SST, frontal gradients, Chl-a) rather than changes in population abundance.

Sightings
Over 5000 purported manta sightings were identi ed in the EUS from 1925 to 2020 (Table 1, Fig. 1). Dedicated aerial surveys for manta rays off northeastern Florida funded by the Georgia Aquarium had the most sightings, followed by the multi-contributor NARWC data which covered much of the U.S. east coast. In the Gulf of Mexico, the highest concentration of sightings was in nearshore waters off Louisiana. On the U.S. east coast, the highest concentration of sightings was in nearshore to shelf-edge waters off Florida and Georgia and nearshore and shelf-edge waters from Cape Hatteras north to New York. Anecdotal manta sightings were also recorded in waters around Puerto Rico and the U.S. Virgin Islands. In the U.S. Virgin Islands, small manta rays were sighted in shallow coastal bays such as Cane's Bay, Maho Bay, and Francis Bay. In Puerto Rico, the majority of manta sightings were reported from the area surrounding Culebra, Vieques, and Mona Islands. The bulk of manta sightings were recorded between 26 and 30° N, with the highest number of sightings from March through May. The vast majority (82%) of sightings north of 35° N were recorded from June to September (Fig. 2).

SEFSC Surveys
The combined SEFSC surveys covered U.S. waters in the Gulf of Mexico and the U.S. east coast, with comprehensive spatiotemporal coverage of most months other than March and September ( Fig. 2A-B).
The most sightings and largest groups were reported in the Gulf of Mexico near the mouth of the Mississippi River, the east coast of Florida, and off Cape Hatteras, North Carolina ( Fig. 2A). Sightings north of 35° N were reported throughout the year, with the majority in July-August (Fig. 2B). In the markrecapture distance-sampling framework, the selected MCDS model included cloud cover, and the selected MRDS function included glare and an interaction between observer and distance (Supplemental File).

NARWC Surveys
The NARWC surveys covered the U.S. east coast, with temporal coverage of all months over the span of the survey; however, only a limited subset, primarily off Florida and Georgia in the winter, contained su cient distance-sampling information to be used in the SDM (Fig. 2C-D). The most sightings and the largest groups were sighted off Florida and Georgia (Fig. 2C). Sightings were reported in all months except May and October (Fig. 2D). The vast majority of sightings north of 35° N were reported during June-September (Fig. 2D). The selected detection function was a half-normal key function of sea state (Supplemental File). Average detection probability within the 348 m swept area was 55.5% (CV = 48.6%).
The NYSERDA surveys covered the nearshore to offshore marine environments of New York, with temporal coverage during the spring/summer of 2016-2019 and fall/winter of 2016-2018 (Fig. 2E-F). Of the 21,539 rays identi ed in the surveys, 504 were initially identi ed as manta rays; however, review of digital photo archives by a trained observer determined only 7 were actually manta rays. The majority of misidenti ed rays were M. mobular or M. tarapacana. All manta sightings and > 99% of all ray (all species) sightings were in summer. Despite comprehensive coast to shelf survey coverage, manta sightings were exclusively in August on the continental shelf edge.

Distribution Modeling
SDMs generated from combined SEFSC surveys explained 4-5% of residual deviance ( Table 2). AUC for internal and external validation were comparable and indicated "acceptable" model ts (Hosmer and Lemeshow 2013; Table 2). Sighting probability was highest at SSTs from 17 to 32°C, with peak probability around 23°C. Sighting probability was highest close to shore at strong thermal fronts. The nal SEFSC survey SDM predicted fairly high probability of occurrence (> 25%) during most of the year south of Cape Hatteras, North Carolina (Fig. 3). Peak probability of occurrence in the Gulf of Mexico and south of Cape Hatteras was predicted during cooler months (November-April), with peak probability of occurrence north of Cape Hatteras during warmer months (May-October).
The 'weighted ensemble' of SDMs generated from the SEFSC, NARWC, and NYSERDA surveys had an "acceptable" predictive t ( Table 2; Hosmer and Lemeshow 2013). The weighted ensemble SDM predicted a relatively uniform probability of occurrence in nearshore environments (Fig. 4). The 'combined surveys' SDM that integrated sightings and effort from SEFSC, NARWC, and NYSERDA surveys predicted higher probabilities of observation with warm SST, moderate Front-Z, nearshore and shelf-edge depths, moderate bathymetric slopes, and increasing Chl-a concentrations (Fig.5). AUC for internal and external validation were comparable and indicated "excellent" model ts (Hosmer and Lemeshow 2013; Table2). The combined surveys SDM explained 19% of residual deviance (Table2).
The combined surveys SDM predicted highest probabilities of detection at offshore sloped habitats (e.g., seamounts) and in the nearshore environment off Louisiana at the Mississippi River delta between April to June and again in October (Fig. 6). Probability of occurrence increased at moderate frontal gradients with SSTs between 20 and 30°C in both nearshore and shelf-edge environments with moderate slopes and high concentrations of Chl-a (Figs. 5-6).
External validation of the SEFSC and combined survey models was challenged by a lack of necessary environmental data for many independent manta ray observations. Some observations were from periods before satellite data were collected; others were too close to shore to generate frontal gradients. However, where models could be successfully t, external validation suggested high predictive utility to independent observations, especially for the combined survey model (Fig. 7). The median Z-score standardized probabilities of observation were signi cantly greater than 0 for both surveys [SEFSC: t(14657) = 241.39, p < 0.0001, x̅ (95% CI) = 0.592(0.588-0.597), Combined: t(4036) = 59.86, p < 0.0001, x̅ (95% CI) = 0.873(0.844-0.902)], con rming that SDM predictions were highly consistent with independent observations of manta rays (Fig. 7).
All SDMs predicted similar spatio-temporal distribution trends (Figs. 3, 4, 6), with higher probabilities of detection on the inside edge of the Gulf Stream from January to April, and peak nearshore distribution off Size estimates for manta rays were limited, but suggest that smaller animals, likely juveniles, were more common in the warmer waters of the Gulf of Mexico, Florida, and Georgia (Figure 9), whereas larger individuals were observed farther north. Anecdotal observations, supported by photo and/or video, of small (1.7 m) individuals from shallow bays in the U.S. Caribbean were also reported. analysis of decades of manta ray sightings across several different aerial survey platforms in the EUS indicated manta rays were most commonly detected at frontal boundaries in productive nearshore and shelf-edge upwelling zones within a thermal optima of approximately 15-30°C (Fig. 5). These ndings are consistent with ecologically-driven expectations that manta rays would be more common near areas of potentially high prey densities (upwelling zones) within thermally optimal conditions. For the Gulf of Juvenile elasmobranchs appear to have wide performance curves characteristic of thermal generalists, which may allow them to survive and succeed in shallow coastal areas subject to more rapid temperature uctuation and higher maximum temperatures (Gilchrist 1995 In this study, SST was the strongest single predictor of manta distribution, with a strong thermal preference apparent between 17 and 32°C, with a peak around 23°C. Temperature preference appears to vary by region, with manta rays off the U.S. east coast commonly found in waters from 19 to 22°C and those off the Yucatan peninsula and Indonesia between 25  found Chl-a, bathymetric slope, and SST were important drivers of manta distribution in the Western Central Atlantic (WCA); however, SST was the least important driver in their model, possibly due to the lower variability in SST in the WCA relative to the U.S. east coast.
The combined surveys model indicated highest probability of occurrence at moderately-sloped nearshore and shelf-edge habitats with moderate SST fronts and high concentrations of Chl-a; all proxies for high primary production and associated manta prey availability (Fig. 5). Due to the lack of spatially comprehensive zooplankton and micronekton sampling data, we were unable to explicitly test associations between manta rays and prey availability. However, strong associations were observed across data sources between manta ray sightings and proxies for productive upwelling zones. Within thermally-optimal bounds, models predicted higher concentrations of manta rays from the coast to the shelf south of Cape  Retrospective analyses of NYSERDA and discussions with aerial observers (  (iv) the probability of the activity impacting the individual, often expressed as a dose-response curve. Our SDMs will help managers determine when an effect of a proposed action on giant manta rays is likely based on the probability of an action taking place when giant manta rays are anticipated in the area.
Even when duration of exposure and probability of adverse effects are unknown, relative risk assessments can be used to identify preferred alternatives, following Farmer et al. (2016). To more accurately determine anticipated take of giant manta rays from proposed actions, further information is needed on movements, site delity, depth utilization, and responses to anthropogenic stressors.
The most signi cant threat to the recovery of giant manta rays across their range is intentional harvest and bycatch in sheries (Miller and Klimovich 2017). Manta rays are targeted or caught as bycatch with virtually every shing gear type, including small-scale sheries characterized by the use of driftnets, gillnets, harpoons, gaffs, traps, trawls, and longlines; and large-scale sheries using driftnets, trawls and purse seines (Croll et al. 2016). Our SDMs will help managers identify areas of spatial and temporal overlap between giant manta rays and commercial sheries, which could be used to reduce bycatch rates in the EUS. Similarly, a better understanding of the spatiotemporal distribution of the species may help improve precision of bycatch estimates by controlling for relative availability of the species to the gear on any given set and allocating observer coverage to areas of higher bycatch concern. For example, preliminary analysis of 2019-2020 data estimated mean EUS shrimp trawl take of manta rays of nearly 1700 individuals/yr (Carlson 2020); however, uncertainty was very high given the short time series and limited data. Although observer coverage on shrimp trawls in the EUS is around 1%, the relative observer coverage on shrimp trawls with trawl effort spatially weighted by manta probability of occurrence was less than 0.09% (Farmer NA, unpublished data).
A major NMFS recovery priority for giant manta rays is to investigate the impact of other threats to the species (e.g., foul-hooking, vessel strikes, entanglement, climate change, pollution, tourism) through research, monitoring, modeling, and management (NOAA 2020). Pate & Marshall (2020) and our SDMs suggest that manta rays are frequently associated with nearshore habitats; as such, they are at elevated risk for exposure to a variety of contaminants and pollutants, including brevetoxins, heavy metals, There is a strong management interest in understanding the inshore extent of manta movements in bays and tidal inlets. SDM predictions suggest seasonal trends with high probability of occurrence in large bays (e.g., Tampa Bay, Chesapeake Bay); however, reported sightings in bays are extremely limited. It is unclear if this is due to reduced water clarity, rarity of use, or very low levels of survey effort. Both  (Table 1); however, the timing and frequency of these observations was variable both seasonally and interannually, with the peak only lasting a few weeks. SDMs capture these trends (Supplemental Video 1), predicting a spring and fall peak in the survey area with the spring peak varying between March and June.
Florida, and southeastern Florida, speci cally, has the highest number of registered recreational vessels and licensed recreational anglers in the U.S., and likely the world (USCG 2019). Manta rays are exposed to exceptionally high levels of vessel tra c as well as hook-and-line shing gear from boats and piers.
Manta rays are often observed foraging on tidal out ows at major inlets in the Palm Beach County area, leading to frequent overpasses by vessels moving at high speeds (Pate & Marshall 2020 use of environmentally-friendly tackle, and fostering engagement with anglers as citizen scientists. Our SDMs will allow managers to more effectively time and coordinate management strategies, including targeted outreach efforts and developing spatiotemporal 'windows' for action agencies to reduce the risk of manta interactions. A major priority for manta ray conservation is to improve understanding of population distribution, abundance, trends, and structure through research, monitoring, and modeling ; NMFS 2020). Our preliminary presence-absence modeling approach uses distance-weighted methods to control for perception bias. With further satellite tagging data on manta ray movements and dive pro les, we may be able to address availability bias for animals that are underwater, and apply similar methods to determine population abundance. Further genetic analyses are needed to resolve the taxonomic status and relative abundance of M. birostris and M. sp. cf. birostris sensu (Marshall et al. 2009). More telemetry studies are also needed to evaluate whether giant manta rays within the Gulf of Mexico and northwestern Atlantic Ocean constitute one large, mixed population, or exist as isolated subpopulations (Stewart et al. 2016). Observing rare species incurs a heavy cost in time, resources, and boat fuel. Our ndings suggest that bathymetric maps combined with real-time satellite data may be used to effectively target manta rays for scienti c study, including the attachment of satellite tags to inform movements, site delity, and dive patterns. Following    External validation of predictive utility of SEFSC and combined ("COMBO": SEFSC, NARWC, and NYSERDA) surveys species distribution models, showing predicted median Z-score standardized probabilities for independent observations for manta rays (see Table 1). Positive Z-scores (above red line) indicate consistency between independent observations and model predictions, with higher Z-scores indicative of greater predictive utility. Time-series of predicted mean weighted latitudinal centroid for best-tting manta ray species distribution models generated from SEFSC, NARWC, and combined surveys ("COMBO") from January 2003 to December 2019 with seasonal auto-regressive integrated moving average (SARIMA) mean (red) and ±1 and ±2 standard error (gray) forecasts for January 2020 to December 2024. Estimated disc width (m) for manta rays observed by National Marine Fisheries Service Reef Visual Census (RVC) divers, Florida Manta Project (FMP) snorkelers, and NYSERDA aerial digital photographic survey, by latitude.