Exposure of Marine Turtle Nesting Grounds to Named Storms Along the Continental USA

: Named storms can cause substantial impacts on the habitat and reproductive output of threatened species, such as marine turtles. To determine the impacts of named storms on marine turtles and inform management, it is necessary to determine the exposure of marine turtle nesting grounds to recent storm activities. To address this, remote sensing information of named storm tracks coupled with nesting ground data were used to investigate the temporal and spatial overlap between nesting grounds for four species of marine turtles in the continental United States of America. All species of marine turtles were exposed to named storms, with variation in exposure driven by the spatial distribution of each population’s nesting ground, the temporal overlap between the storms and reproductive events, and nest placement on the beach. Loggerhead turtles were the most exposed species to named storms, with the northern management unit having signiﬁcantly higher exposure levels than all other loggerhead management units. Kemp’s ridley turtles, in contrast, were found to be the least exposed species to named storms. This study establishes a valuable current baseline against which to measure and compare future impacts that result as climate change progresses and storms become more frequent and intense. Importantly, cumulative and synergetic e ﬀ ects from other climatic processes and anthropogenic stressors should be considered in future analysis. S.C. visualization, M.M.P.B.F. N.W.; supervision, M.M.P.B.F.; administration, M.M.P.B.F.


Introduction
Understanding the spatial and temporal exposure of biodiversity to specific threats is crucial for targeted management and their conservation. This is particularly true for species threatened with

Named Storms
The National Hurricane Center, through the Tropical Analysis and Forecast Branch (TAFB), provides spatial files for each named storm formed in the North Atlantic region and their strength (intensity), which are derived from data and imagery from multiple satellites and estimated using the Dvorak technique [51]. Some of the most recent satellites used consist of the NOAA polar-orbiting satellites including the Advanced Microwave Sounding Unit (AMSU), the NASA Global Precipitation Mission (GPM), the European Space Agency's Advanced Scatterometer (ASCAT), and Defense Meteorological Satellite Program (DMSP) satellites, among others). Spatial data of each named storm (n = 159 tropical storms and hurricanes; hereafter "named storms" for simplicity) in the North Atlantic region between 2008 and 2018 were downloaded from the National Hurricane Center database ("Best Track" available at https://www.nhc.noaa.gov/gis/; downloaded November 2018). We obtained wind swath shapefiles for 151 named storms, which display the extent of impact of the different wind force levels of a named storm (34 kt or tropical storm wind force, 50 kt, and 64 kt or hurricane-force winds) [52]. We selected only the tropical storm wind force swaths (34 kt), to explore the minimum impact of each named storm across its entire geographical extent. For a small subset of named storms (8 out of 159) for which wind swaths were not available, we used the named storm tracks from the International Best Track Archive for Climate Stewardship (IBTraCS) dataset [53]; available at https://www.ncdc.noaa.gov/ibtracs/index.php). Because IBTraCS datasets only include the central track of a named storm, we buffered each track by 40 km to have a representation of the minimum extent of impact of that particular named storm [13]. We summed the occurrence of named storms at each grid cell (15 km grid resolution) to create a rasterized map of the total frequency of named storms. A 15 km grid resolution was used as it was considered the best spatial resolution to cover the length of most nesting grounds and provide fine enough resolution to display differences in the frequency of storms. All spatial analyses were computed in ArcGIS 10.6.

Temporal Overlap
The annual (total number of named storms per year) and monthly (average number of named storms per month across the study period) distribution of named storms that crossed nesting beaches were plotted and assessed with a Chi-square test to depict temporal trends in named storm activity. To assess the temporal overlap between marine turtle nesting and hatching season and named storms during the Atlantic hurricane season, we visually compared the monthly distribution of named storms with the nesting and hatching seasons for each species. In addition, to explore trend changes in the monthly distribution of named storms across the study period, we explored temporal changes by splitting the temporal scale of our data in two time-periods (2008-2012 and 2013-2018) and comparing them graphically.

Spatial Overlap
The exposure of marine turtle nesting grounds to named storms was calculated by multiplying the frequency of named storms that crossed a nesting ground by the density class of that nesting beach (see Excel S1). To make the named storm frequency and marine turtle density spatial layers comparable, we normalized each dataset with a linear transformation. The resulting values ranged from 0 to 1 and were classified based on equal intervals into "no exposure" (0), "low" (≤0.25), "medium" (0.25-0.50), "high" (0.5-0.75) and "very high" (>075) exposure [54]. For loggerhead turtles, we calculated the average exposure for each of the management units. For leatherback turtles, Kemp's ridley turtles, and green turtles, we calculated the overall average exposure across the species because management units have not yet been defined (leatherback turtles) or are in the process of being defined (green turtles, see [55,56]) or because the consistent nesting range is restricted to one region (Kemp's ridley turtles).
To assess differences in the exposure of the different species and among loggerhead turtle management units to named storms, we fitted general linear models (GLMs) with negative binomial distributions. Additionally, to assess differences in the frequency of named storms among loggerhead turtle subpopulations, a GLM with Poisson distribution was fitted. For all models, we applied a post-hoc Tukey's HSD to perform pairwise comparisons. All statistical analyses were performed in R 3.5.1 [57].

Case Study: Disturbance by Named Storms in North Carolina, USA
Changes in hatching success (as defined by [58]) for loggerhead turtle, Kemp's ridley turtle, leatherback turtle, and green turtle nests laid on beaches in southern versus northern North Carolina, USA, and before and after Hurricane Florence was explored. Hurricane Florence made landfall near Wrightsville Beach in south-central North Carolina on 14 September 2018 as a Category 1 hurricane, but heavy rainfall and strong winds were observed starting on 10 September (https: //www.weather.gov/mhx/Florence2018) (Figure 1). Hurricane Florence was accompanied by a storm surge of 2-4 meters and a devastating rainfall of 51-76 cm, which caused extensive flooding and damage to coastal areas.
Hatching success of individual nests was evaluated by examining contents of each nest 72 hours after primary emergence of hatchlings or after 70 days of incubation if there was no sign of hatchling emergence. In cases where nest markers were lost due to the hurricane, efforts were made to find any remaining contents of the nests by digging in the sand near the nest's recorded GPS location. If no nest contents were found, the nest was assumed to have been lost to erosion associated with the storm and assigned 0% hatching success. Eight nests could not be assigned hatching success due to other circumstances (e.g., coyote predation around the time of expected emergence) and were removed from the analyses. Nests were grouped as pre-hurricane Florence if they had been laid on or before 19 July, which is 57 days prior to September 10, and as post-hurricane Florence if they were laid after 19 July. The value of 57 days was the average incubation period for nests that produced emergent hatchlings during the two weeks immediately prior to 10 September. For spatial comparisons, we grouped nests as those laid south or north of 35 • N, near Ocracoke Inlet, nearly 185 km straight line distance NE from where Hurricane Florence made landfall. For spatial comparisons, only nests laid after 19 July were used in the analyses. circumstances (e.g., coyote predation around the time of expected emergence) and were removed from the analyses. Nests were grouped as pre-hurricane Florence if they had been laid on or before 19 July, which is 57 days prior to September 10, and as post-hurricane Florence if they were laid after 19 July. The value of 57 days was the average incubation period for nests that produced emergent hatchlings during the two weeks immediately prior to 10 September. For spatial comparisons, we grouped nests as those laid south or north of 35 °N, near Ocracoke Inlet, nearly 185 km straight line distance NE from where Hurricane Florence made landfall. For spatial comparisons, only nests laid after 19 July were used in the analyses.

Temporal Overlap between Nesting Season and Named Storms
The total frequency of named storms that crossed marine turtle nesting grounds during the study period (between one and six named storms per year) oscillated across years without any clear pattern (Chi-squared test; X 2 = 9.8333, df = 10, p = 0.4552) (Figure 2a). However, the total monthly named storm frequency was significantly higher during the summer months (May to August) (Chi-squared test; X 2 = 15.333, df = 6, p = 0.01782) (Figure 2b).

Temporal Overlap between Nesting Season and Named Storms
The total frequency of named storms that crossed marine turtle nesting grounds during the study period (between one and six named storms per year) oscillated across years without any clear pattern (Chi-squared test; X 2 = 9.8333, df = 10, p = 0.4552) (Figure 2a). However, the total monthly named storm frequency was significantly higher during the summer months (May to August) (Chisquared test; X 2 = 15.333, df = 6, p = 0.01782) (Figure 2b). When looking at trends between two periods (2008-2012 and 2013-2018), monthly average named storm frequency has shifted from having one clear peak in August (average ± SE = 1.6 ± 0.3 storms/month) for the first period (2008-2012) to a more even distribution from June throughout September between 2013-2018 (average ± SE = 0.6 ± 0.17 storms/month) ( Figure 3). In terms of temporal overlap, most of the nesting season for loggerhead turtles occurs just before and throughout the Atlantic hurricane season, while nesting season for green turtles occurs in its entirety during the hurricane season. For leatherback turtles and Kemp's ridley turtles, the first half of the nesting season occurs outside the hurricane season ( Figure 3). When considering the hatching season, all species experience an overlap with the hurricane season, with the majority of the hatching season for loggerhead turtles and green turtles occurring throughout the peak of the hurricane season. In contrast, for leatherback turtles and Kemp's ridley turtles, only the second half of the hatching season overlaps with the peak hurricane season ( Figure 3). When looking at trends between two periods (2008-2012 and 2013-2018), monthly average named storm frequency has shifted from having one clear peak in August (average ± SE = 1.6 ± 0.3 storms/month) for the first period (2008-2012) to a more even distribution from June throughout September between 2013-2018 (average ± SE = 0.6 ± 0.17 storms/month) ( Figure 3). In terms of temporal overlap, most of the nesting season for loggerhead turtles occurs just before and throughout the Atlantic hurricane season, while nesting season for green turtles occurs in its entirety during the hurricane season. For leatherback turtles and Kemp's ridley turtles, the first half of the nesting season occurs outside the hurricane season ( Figure 3). When considering the hatching season, all species experience an overlap with the hurricane season, with the majority of the hatching season for loggerhead turtles and green turtles occurring throughout the peak of the hurricane season. In contrast, for leatherback turtles and Kemp's ridley turtles, only the second half of the hatching season overlaps with the peak hurricane season (Figure 3).

Spatial Exposure of Nesting Grounds to Named Storms
Spatial distribution of named storms from 2008-2018 varied across the study region. The northern extent of the eastern USA coast, from Georgia to North Carolina, overlapped with the highest frequency of named storms (11-15 named storms over ten years, or ~1 named storm every year; Figure 4). At the species level, there were significant differences in the average exposure of nesting grounds to named storms across the study period (2008-2018) (GLM fit with negative binomial distribution; X 2 = 30.843, df = 3, p = 9.17e-7), with loggerhead turtles being the most exposed species to named storms and Kemp's ridley turtles the least exposed (Table 1, Figure 4). A post-hoc Tukey test (Table S1) indicated that the exposure of Kemp's ridley turtles is significantly lower than the exposure of all other species. Similarly, the exposure of leatherback turtles was significantly lower than the exposure of loggerhead turtles (Table S1). There were no significant differences between the exposure of green turtles and that of loggerhead turtles or leatherback turtles (Table S1).

Spatial Exposure of Nesting Grounds to Named Storms
Spatial distribution of named storms from 2008-2018 varied across the study region. The northern extent of the eastern USA coast, from Georgia to North Carolina, overlapped with the highest frequency of named storms (11-15 named storms over ten years, or~1 named storm every year; Figure 4). At the species level, there were significant differences in the average exposure of nesting grounds to named storms across the study period (2008-2018) (GLM fit with negative binomial distribution; X 2 = 30.843, df = 3, p = 9.17e-7), with loggerhead turtles being the most exposed species to named storms and Kemp's ridley turtles the least exposed (Table 1, Figure 4). A post-hoc Tukey test (Table S1) indicated that the exposure of Kemp's ridley turtles is significantly lower than the exposure of all other species. Similarly, the exposure of leatherback turtles was significantly lower than the exposure of loggerhead turtles (Table S1). There were no significant differences between the exposure of green turtles and that of loggerhead turtles or leatherback turtles (Table S1).    Figure 4) across their nesting range. Nearly 45% of nesting grounds were exposed to medium levels of named storms while 5% of nesting grounds reached high and very high exposure levels. In addition, exposure differed significantly at management unit level (GLM fit with negative binomial distribution; X 2 = 148.16, df = 5, p = 2.210 -16 ). Based on a post-hoc Tukey test (Table S2), the exposure to named storms of the northern unit was significantly higher than the exposure to named storms of all other management units. Further, within the Gulf of Mexico, the northwestern management unit had significantly more exposure to named storms than the southwestern management unit (Excel S1). Within the northern unit, exposure reached very high levels in Cape Island and Lighthouse Island in South Carolina, both of which are very high-density nesting grounds for loggerhead turtles (Figure 4; Excel S1). High levels of exposure to named storms were also recorded in other very high-density nesting grounds across the nesting range, such as Edisto Beach State Park and Botany Bay Island in South Carolina (northern unit), and Cape St. George and St. Joseph Peninsula in Florida (northwestern unit) (Figure 4; Excel S1). Lowest exposure levels were recorded within the southern and central Florida management units, namely in the south-eastern (e.g., Lower Matecumbe Sea Oats Beach and Fat Deer Key), south-western (e.g., Marco Island and Big Hickory Island), central-eastern (e.g., Fort Pierce Inlet State Park and North Saint Lucie County Beaches), and central-western (e.g., North Captiva Island (South) and Sanibel Island (East)) management units. Named storm frequency along the northern unit averaged 10 named storms from 2008-2018 (or roughly one named storm every year), while average frequency of named storms that crossed the remaining management units ranged between 4.8 (one named storm every 2.3 years in the south-western unit) and 6.7 (one named storm every 1.5 years in the north-western unit) ( Table 1).

Green Turtles
Exposure of green turtles to named storms reached, on average, low levels across the range (0.24, range 0.07-0.42; Table 1, Figure 4). Across the nesting range, over 40% of nesting grounds recorded medium levels of exposure to named storms (Excel S1). Nearly 35% of nesting grounds along the coast of Florida displayed medium exposure to named storms, including all very high-density nesting grounds along the eastern coast (i.e., South Brevard, Coral Cave, and Tequesta beaches) and 24 out of the 38 high-density nesting grounds (Figure 4; Excel S1). Medium levels of exposure were also recorded for all medium density nesting grounds from Georgia to North Carolina. In contrast, the lowest exposure to named storms recorded was for St. Joseph Peninsula State Park, St. Vincent National Wildlife Refuge and Ten Thousand Islands National Wildlife Refuge in Florida (Figure 4; Excel S1). In terms of named storm frequency, green turtle nesting grounds in South Carolina and North Carolina received each an average of one annual named storm every year, while those in Georgia received one named storm every 1.2 years, and those in Florida and Alabama received roughly one named storm on a biannual basis.

Leatherback Turtles
Leatherback turtle nesting grounds displayed, on average, low levels of exposure to named storms (0.22, range 0.08-0.32; Table 1, Figure 4), with just over 30% of nesting grounds across the range recording medium levels of exposure (Excel S1). The highest levels of exposure to named storms were registered for Florida, where nearly 40% of nesting grounds displayed medium exposure levels -all located along the eastern coast-including all very high-density nesting grounds (i.e., Hobe Sound National Wildlife Refuge, John D. MacArthur Beach State Park, Jupiter/Juno Beach, and Palm Beach Shores) (Figure 4; Excel S1). Furthermore, exposure of high-density nesting grounds along the coast of Florida was evenly split between low (56% of nesting beaches) and medium (44%) levels of exposure (Excel S1). In contrast, the lower levels of exposure to named storms were recorded for Sanibel Island (East) and Jetty Park in Florida, and Hilton Head Island in South Carolina. Exposure to named storms remained low across all other states (Georgia, South Carolina and North Carolina). Across the nesting range, named storm frequency averaged one named storm every 1.6 years.

Kemp's Ridley Turtles
The exposure of Kemp's ridley turtles was, on average, low across their range in Texas (0.12, range 0.05-0.21; Table 1, Figure 4). The highest levels of exposure to named storms were recorded for all very high-and high-density nesting grounds (i.e., Boca Chica, South Padre Island, and North Padre Island) and one medium density nesting ground (Matagorda Island) (Figure 4; Excel S1). The lowest exposure to named storms was recorded for the Matagorda Peninsula, Galveston Island, Mustang Island, and San Jose Island. Named storm frequency along Kemp's ridley nesting grounds was on average one named storm every three years.

Case Study: Disturbance by Named Storms in North Carolina, USA
A total of 799 marine turtle nests (loggerhead turtles = 765, green turtles = 20, Kemp's ridley turtles = 12, and leatherback turtles = 2) were laid in North Carolina during the 2018 nesting season, with the first documented nest laid on 10 May and the last nest laid on 06 September. Kemp's ridley turtles nested from 18 May to 07 July and leatherback turtles nested from 19 May to 02 June, and therefore were not impacted by Hurricane Florence. One-hundred and twenty-two (16%) loggerhead nests were laid after 19 July. These nests had significantly lower hatchling success when compared to those nests laid on 19 July or earlier (p < 0.0001, Kruskal-Wallace nonparametric test, with Dunn's multiple comparison test post-hoc, Table 2). Green turtle hatchling success was not significantly different for nests laid before or after hurricane Florence (Table 2). Together, loggerhead turtle and green turtle nests laid north of 35 • had significantly higher hatching success than those laid south of 35 • (p < 0.001, Kruskal-Wallace nonparametric test), Dunn's multiple comparison test post-hoc revealed that only the hatching success of green turtles nests in the northern part of North Carolina was significantly higher (Table 3).

Discussion
All species of marine turtles nesting in the continental USA are exposed to named storms and associated impacts. Variation in exposure to named storms was driven by the spatial distribution of each population's nesting grounds, with populations nesting in the northern extent of the eastern USA from Georgia to North Carolina being the most exposed. Consequently, loggerhead turtles were found to be the most exposed species, with the northern management unit having significantly higher exposure levels than all other loggerhead management units. Kemp's ridley turtles, in contrast, were found to be the least exposed species to named storms given their restrained geographic extent, with nesting only on the western coast of the Gulf of Mexico.
The impact that each nesting ground species, and ultimately populations will experience from named storms is a function of their exposure to the storms, the temporal overlap between the storms and nesting and hatching, the importance of each of the nesting grounds in terms of numbers of incubating eggs during the storm, and nest placement on the beach [13,14]. In our study, a significantly higher number of named storms occurred during the summer months which overlaps with marine turtle nesting and hatching season in the region. However, leatherback turtles and Kemp's ridley turtles have been less exposed to storms as they nest earlier in the season (the first half of their nesting season occurs outside the hurricane season) and thus their eggs are exposed to storm for a shorter period (only the beginning of the storm season) which historically has had fewer storms. Nevertheless, our work indicates that storm frequency is becoming more evenly distributed throughout the storm season, due to the increased duration of storm effects, which could have a larger effect on all marine turtle nesting and hatching in the southeast USA due to the prolonged duration of storm effects.
This increase in exposure can have negative impacts on the reproductive output of marine turtles at each nesting ground. As indicated by our case study and other studies, named storms can dramatically decrease hatchling success and overall productivity [18,20,59]. Impacts on the overall productivity of nesting populations will be driven by the importance of the nesting ground (in terms of the percentage of turtles from a population that nests on that site) that the named storm hits. Indeed, Fuentes et al. [13], found that the impacts of named storms to the northern Great Barrier Reef green turtle population can be 100 times higher if it hits the main nesting sites for this population rather than a trivial nesting ground. In our study, high levels of exposure to named storms were recorded in very high-density nesting grounds used by the northern management unit (Cape Island and Lighthouse Island) and high-density nesting grounds (Cape St. George and St. Joseph Peninsula) used by the northwestern management unit. Aperiodic impacts from named storms might be an issue particularly for the NW management unit as it is one of the smallest loggerhead management units within the Northwest Atlantic Ocean Distinct Population Segment [60]. Thus, any impact to this management unit can have severe implications for population stability.
Nest placement on a beach will also affect the impact that named storms will have on overall reproductive output, with nests being placed closer to the high tide line being more vulnerable [61,62]. In our case study, distance (i.e., km) from where the hurricane made landfall seems to have made little impact on the overall hatching success of loggerhead nests deposited post-Hurricane Florence.
In contrast, green turtle eggs on beaches further away from the hurricane had significantly higher hatching success. This might be due to species-specific differences in nest-site selection relative to the dune line: on the same nesting grounds, green turtles tend to lay their eggs closer to the dune line than loggerhead turtles, which tend to nest closer to the high tide line [63]. This difference may protect green sea turtle nests from excessive washover.
Even though named storms can directly impact nests that are incubating at the time of landfall, some of the reproductive strategies used by marine turtles can help mitigate the impacts of these discrete events and allow faster than expected recovery from disturbances [14]. Traits include large clutches (approximately 100 eggs/clutch), laying multiple clutches within a nesting season (2-7 clutches depending on species) which are spaced throughout the season (2-3 weeks apart), and extended nesting seasons (several months) [64]. Nevertheless, despite marine turtles having these traits and the ability to withstand the loss of a single cohort of eggs [65], frequent disturbances to incubating eggs (and thus low hatching success over time) from named storms can potentially have a profound effect on the longer-term reproductive output of a population, potentially exerting substantial selective pressures [13,66,67]. This might have been the case for marine turtles nesting in the eastern Queensland coast of Australia, as they currently nest in areas with historically low incidence of tropical cyclones [13]. It was hypothesized by the authors that areas with a high incidence of tropical cyclones might have substantially affected nesting and/or hatching success over time, reducing the number of turtles that return to nest in these areas over several generations [13], which would mean that cyclone activity was an important factor in shaping the distribution of marine turtle nesting grounds along the eastern Queensland coast. Such effects still need to be assessed in the USA.
As climate change progresses, storms may become more frequent and intense [68] and sea level may continue to rise, potentially causing significant alteration of the physical structure of nesting grounds. This coupled with contemporary anthropogenic threats (e.g., coastal development, pollution), may hinder the ability of marine turtle traits to buffer impacts from named storms. This highlights the need to understand how predicted changes in named storm frequency and distribution may affect marine turtle populations; studies such as the present one will provide baseline information for comparison. Future studies should also explore how different pressures (e.g., storm surge, temperature, rainfall), associated with named storms affect hatching success to provide further insights into how named storms with different characteristics may affect marine turtle reproductive output at a site. Importantly, cumulative and synergetic effects from other climatic processes and anthropogenic stressors also need to be considered. Climate processes will not act in isolation and they may produce unexpected changes in combination with local conditions and other pressures [69,70]. For example, seawalls and other structures to protect beaches in Florida from erosion associated with named storms can impede reproductive female marine turtles from accessing nesting habitat [71]. Additionally, the number of structures placed on some beaches increased following named storms [72], suggesting that in the future nesting females may face impediments to nesting grounds more frequently, as the frequency of named storms increases.

Conclusions
Remote sensing information on named storm track proved useful in determining the exposure of marine turtle nesting grounds to named storms and identifying populations and species at most risk from impacts. Our broad-scale assessment allowed us to identify areas with higher exposure to named storms, which warrants site-specific assessments of impacts. Site-specific assessment can benefit from the use of other datasets obtained through remote sensing (e.g., LiDAR, satellite microwave radiometers, imagery) to help monitor flood conditions, land cover change, and erosion at marine turtle nesting grounds and determine impact to productivity [73]. An understanding of the exposure of marine turtle nesting grounds to named storms, as obtained here, can be used as a baseline as climate change progresses and storms intensify and become more frequent.