Understanding population status for endangered species is critical to developing and evaluating recovery plans mandated by the Endangered Species Act. For sea turtles, changes in abundance are difficult to detect because most life stages occur in the water. Currently, nest counts are the most reliable way of assessing trends. We determined the rate of growth for leatherback turtle (Dermochelys coriacea) nest numbers in Florida (USA) using a multilevel Poisson regression. We modeled nest counts from 68 beaches over 30 years and, using beach-level covariates (including latitude), we allowed for partial pooling of information between neighboring beaches. This modeling approach is ideal for nest count data because it recognizes the hierarchical structure of the data while incorporating variables related to survey effort. Nesting has increased at all 68 beaches in Florida, with trends ranging from 3.1% to 16.3% per year. Overall, across the state, the number of nests has been increasing by 10.2% per year since 1979. Despite being a small population (probably <1000 individuals), this nesting population may help achieve objectives in the federal recovery plan. This exponential growth rate mirrors trends observed for other Atlantic populations and may be driven partially by improved protection of nesting beaches. However, nesting is increasing even where beach protection has not been enhanced. Climate variability and associated marine food web dynamics, which could enhance productivity and reduce predators, may be driving this trend.
If you can't find a tool you're looking for, please click the link at the top of the page to "Go to old article view". Alternatively, view our Knowledge Base articles for additional help. Your feedback is important to us, so please let us know if you have comments or ideas for improvement.
Since the United States Endangered Species Act of 1973 (ESA 1973) legislation was enacted, some endangered species populations have stabilized or increased (Male and Bean 2005), while others have declined; many more species are candidates for listing under the Act (Pennisi 2007). The reasons for population decline and collapse vary, but formal recovery plans may be successful when they are well developed and contribute to sound management (Glick 2005, Male and Bean 2005). Overall, the ESA could be considered relatively effective: following 13 years of official protection, only 35% of listed species continue to decline (Male and Bean 2005). Assessing population trends is critical for setting objectives in recovery plans and for assessing the status of endangered species.
At present, six sea turtle species are listed as endangered or critically endangered by the International Union for Conservation of Nature (IUCN 2004). The Kemp's ridley turtle (Lepidochelys kempii), once considered the most endangered sea turtle species, experienced a resurgence in nesting numbers over the last 20 years, in large part because of nesting beach protection of females and hatchlings and because fisheries-related mortality has been reduced as a result of the implementation of Turtle Excluder Devices (TEDs) in U.S. and Mexican trawl fisheries (Heppell et al. 2005). Yet generalities about the global status of sea turtle species cannot be made because there are regional differences in population status (increasing, decreasing, stable, or unknown) and threats vary globally (Broderick et al. 2006, Godfrey and Godley 2008).
The assessment of sea turtle population status is difficult, and pinpointing the underlying causes of increases or declines in abundance is even more challenging. During many life stages, turtles may be cryptic or they may reside in developmental habitats that are poorly understood. Life history parameters such as survival, age to maturity, reproductive longevity, breeding rates, and lifetime reproductive output (fecundity) can only be roughly estimated at present. Sea turtles do not nest every year and they occasionally move among nesting beaches within a region, but the best opportunity for assessing population size or status is still on the nesting beach, where nesting females are easily accessible. Ideally, every individual nesting female would be identified and followed throughout its reproductive lifetime to estimate life history parameters, but the labor required makes this impossible. Researchers therefore rely on nest counts as a proxy for the number of nesting individuals in a population (Meylan 1982).
The first leatherback nest discovered in the continental United States was in Florida in 1947 (Carr 1952), and nesting is now quite common. The nesting season begins in late February to early March (Meylan et al. 1995), peaks in mid-to-late May, and drops off quickly during early June, with minimal nesting continuing into August.
The purpose of our study was to determine trends in leatherback nest numbers in Florida. Our goal was to use as much of the available data as possible to evaluate statewide nesting and to describe spatial patterns in nesting. We then explore ideas about the observed increase in leatherback nests in Florida and at other rookeries throughout the Caribbean, and consider the contribution Florida may make to meeting objectives in the U.S. recovery plan.
In Florida, two complementary sea turtle nesting survey programs designed and coordinated by the Florida Fish and Wildlife Conservation Commission's Fish and Wildlife Research Institute (FWC, FWRI) are conducted. Each beach jurisdiction that participates in these surveys is responsible for collecting the nesting data (under a special state permit), while FWC staff compiles, analyzes, and disseminates the data. The Statewide Nesting Beach Survey (SNBS) program began in 1979 through an agreement between FWC and the U.S. Fish and Wildlife Service. Depending on location, surveys begin as early as 1 March and continue daily through September or October annually. The abundance, distribution, and seasonality of five species of sea turtle (loggerhead Caretta caretta, green turtle Chelonia mydas, leatherback, hawksbill Eretmochelys imbricata, and Kemp's ridley) are documented by hundreds of trained volunteers on nearly 200 beaches. The Index Nesting Beach Survey (INBS) program operates within the framework of the SNBS program but has a shorter survey duration (15 May–31 August). The hallmarks of the INBS program are consistent effort and finely resolved (∼0.9-km) zone-level reporting; this survey is used mainly for analyzing nesting trends. Generally, surveys begin as soon as possible after dawn; at each beach, surveyors begin at one end and travel (by foot or vehicle) to the opposite end of the beach (usually a consistent distance each day). Actual turtles are not encountered, but each turtle track (or crawl) is assessed for whether the turtle from the night before deposited eggs (nest) or departed without nesting (false crawl). Once tracks are evaluated and counted, they are crossed out so that only new tracks are counted during subsequent surveys. For surveys with less than 7 days/week frequency, either all tracks since the last survey or only tracks from the previous night are counted. Both nests and false crawls are tallied on a daily basis and yearly totals are reported to the state. We used the Statewide Nesting Beach Survey data to analyze nesting trends for leatherbacks because that survey captures a very high proportion of leatherback nesting in Florida each year because it begins earlier than the INBS program.
We began with every record within the SNBS data set for our analyses. For each year, county, and beach, the following data were available: beach boundaries, survey distance, number of days surveyed per week, survey start and end dates, number of leatherback nests and false crawls, and dates of first and last nest observations. All records were sorted by county and year. For each county, the number of nests for each year was tabulated and the average number of nests for the most recent five-year period (2004–2008) was then calculated to represent, in general, the current nesting levels. The total number of nests for the state was calculated from individual beach counts by year (1979–2008). The level of survey effort varied widely from beach to beach. In any given year, individual beaches may not have been surveyed daily, some beaches were not surveyed every year, and beaches that were surveyed daily or yearly may not have started their surveys on the same day each year.
One of the key considerations in our choice of analysis methods for these data was the varied effort at each beach. Because one or more of these effort factors may have had a direct effect on the number of nests documented on a particular beach, it was important to be able to take survey effort into account. Classical regression methods and generalized additive models were considered because the log(nest counts) did vary linearly over time for those beaches that did have long-term data and consistent survey start and end dates. However we modeled the original beach counts from each beach for each year (1979–2008) using a multilevel model (MLM) for three important reasons. First, because we have more than one beach in the model (n = 68 beaches) over multiple years, there is a hierarchical structure to the data and MLMs are ideal for handling this type of structure. Second, MLMs allowed us to reasonably estimate trends for beaches with small sample sizes (number of years) as well as for beaches with long time series, while taking into consideration varying survey effort for all beaches (three effort predictor variables). Finally, by not excluding data from beaches with only a few years of nest counts, we could use all available information to estimate the overall statewide trend with more accuracy.
A multilevel model represents a compromise between traditional alternatives of obtaining unpooled or completely pooled parameter estimates (Gelman and Pardoe 2006) such as beach trends. The data are therefore considered partially pooled. Under complete pooling, in which a single regression model is fitted to the individual beach counts, each beach is assumed to have the same trend and intercept and all observations are considered independent. These assumptions will often be incorrect, given the hierarchical structure of the data. We would expect that beaches would differ in their suitability for nesting and that counts on the same beach in consecutive years would be more similar than a series of annual counts across different beaches. Violating these assumptions will result in the underestimation of standard errors of the trend estimates, thereby leading to trends that are too often stated as statistically significant.
At the opposite extreme, one may fit a separate regression model within each beach (no pooling). Under no pooling, trend estimation at beaches with few nest count observations will be unreliable, sometimes producing significant trends that are spurious, or, if data are meager, statistical trend detection would become more difficult with the risk of a Type II error being made. This modeling (no pooling) approach does not allow the sharing of information across beaches that are in close proximity. Because sea turtles may travel between beaches for subsequent nests, this information could be useful if shared between a beach with numerous observations and a beach with few observations.
Multilevel models recognize the hierarchical structure of the data (nest counts within beaches over multiple years) and offer partial pooling of information across beaches, allowing for individual beach trends to be obtained. In this paper, we use multilevel models for trend estimation at beaches with limited data, borrowing information about trends from nearby beaches that have robust long-term data.
Although nearly 200 beaches are surveyed yearly in Florida, only 68 beaches (66 on the east coast, two on the Panhandle) had sufficient leatherback nesting for calculating meaningful trends. Beaches with five or more nests over the entire survey period were included, as well as a few beaches that had lower level nesting but were in close proximity to more densely nested beaches.
Annual trends at the nesting beaches were modeled using a multilevel Poisson regression. This model assumes that there are two levels in the data set: the first-level units (the annual counts) are assumed to be nested within the second-level units (the beaches). Data consisted of 1385 (n) annual counts recorded at 68 (J) beaches. Table 1 contains summary details for each of the covariates for the first level in the model. For each beach, the total number of nests over the survey period (number of years), as well as the average survey days per year, the average distance surveyed daily for each year, and the average number of days surveyed per week are listed. In addition, the first and last year of surveys are indicated, although these years may or may not be consecutive (see the number of years column for the exact count).
Table 1. Trends (percentage annual change and credible interval, CI) estimated using multilevel modeling for 68 Florida (USA) beaches, south to north, used for nesting by leatherback turtles (Dermochelys coriacea), for the survey years listed.
To estimate the temporal trend at each beach, we included a first-level covariate for year x1. Annual counts yi, i = 1, … , n, were modeled using a varying-intercept, varying-slope model in which beaches were allowed to vary both in terms of their intercept α (nest count) and slope coefficient for year β1 (trend). Because sampling effort varied both among beaches and within the same beach across years, differences in survey effort among beaches and among years within a beach were accounted for by the inclusion of first-level effort covariates: length (km) of beach surveyed per day, x2; number of days per week surveyed, x3; and number of days surveyed per year, x4. The first level of the model is thus:
where subscript j[i] indicates the beach corresponding to count i. The percentage change in counts per year at beach i may be evaluated as (exp[β1j[i]] − 1) × 100.
At the second level of the model, the beach intercepts and slopes, αj and β1j, are assumed to have a bivariate normal distribution:
where μα is the average intercept, in this case the average log(beach count) when the predictors in Eq. 1 are set to zero, is the variance in the log(beach count) among beaches, μβ is the average beach slope, is the variation in slope among beaches, and ρ represents the correlation between the αj's and β1j's. By assigning the distribution in Eq. 2 to αj and β1j, estimates of individual beach intercepts and trends are smoothed toward their mean levels μα and μβ. The average percentage change in beach counts per year may be expressed as (exp[μβ] − 1) × 100.
Beach intercepts and slopes may be similar among neighboring beaches. Preliminary exploration of the data suggested a possible quadratic relationship, after adjusting for variation in sampling effort, between latitude and beach trends, and that counts were highest and most variable at beaches south of Cape Canaveral. These spatial patterns were further explored by considering additional beach-level covariates and variance parameters in the model (Eq. 2):
1) a linear or quadratic covariate, μj and , to quantify the effect of latitude on the beach trends β1j;
2) a binary indicator zj to quantify any change in the beach counts αj north (zj = 0) and south of Cape Canaveral (zj = 1);
3) parameters and ρ in Eq. 2 were replaced by ( , ρ0) for beaches north of Cape Canaveral and by ( , ρ1) for beaches to the south to quantify differences in variability among the beach intercepts (nest counts).
With the inclusion of these second-level covariates and variance parameters, the second level of the full model would be:
The second-level covariates and variance parameters were considered in a forward stepwise selection process. If there is any additional variation in the count data that is not predicted by the final model, then the data are overdispersed. We accounted for possible overdispersion by a variance component σ2 in the multilevel model in Eq. 1 :
A value for σ2 greater than zero implies that there are unknown factors influencing nest counts that were not included in the model.
We took a Bayesian estimation approach and fitted the models using Markov chain Monte Carlo (MCMC) simulations in WinBUGS 1.4.3 (Lunn et al. 2000, Spiegelhalter et al. 2003). The priors given to the unknown parameters in the model were chosen to reflect vague prior information. Specifically, a normal distribution with a mean of zero and variance of 1000 was used here as a prior for each of the regression coefficients and intercept parameters, a uniform distribution bounded between 0 and 100 (Gelman 2006) for each of the standard deviation parameters, and a uniform distribution bounded between −1 and 1 for each correlation parameter. All first- and second-level covariates were centered to reduce the correlation between the regression coefficients and to speed MCMC convergence. Results are based on samples from the MCMC chain, after discarding the first 10 000 iterations (the burn-in) to ensure that the chain had converged. Three chains were run and convergence was assessed using the Gelman-Rubin statistic modified by Brooks and Gelman (1998). We ran one chain for a further 250 000 iterations and retained 1 in every 10 samples to ensure that autocorrelation among samples was close to zero. Estimates of the model parameters (the mean and 95% credible interval (CI)) are therefore based on 25 000 samples. Models were compared using the deviance information criterion (DIC), a Bayesian alternative to Akaike's information criterion (Spiegelhalter et al. 2002). Models with DIC values differing by more than 2 were considered to be substantially different in their fit to the data (Spiegelhalter et al. 2002). The model with parameter combinations that resulted in the lowest DIC was chosen as the best model.
Leatherback nesting has been recorded in 20 of 34 coastal counties in Florida (Fig. 1); however, most nesting is concentrated along the eastern coast, south of Cape Canaveral. The counties with the greatest yearly number of nests are St. Lucie, Martin, and Palm Beach counties (Fig. 1). These three counties accounted for 83.8% of all nesting recorded in the state since 1979, even though these counties accounted for only 11.6% of the distance surveyed. Highest nest counts were reported in Palm Beach County, with 38.7% of all nesting, followed by Martin County with 32.1% and St. Lucie with 13.0%. Many of the beaches in these counties have been continuously surveyed since the early 1980s.
All leatherback nest counts in Florida for each year (1979–2008) are shown in Fig. 2. From an average of 63 nests per year in the 1980s, nest counts rose to 263 nests per year in the 1990s and then to 754 nests per year for the 2000s (with nine years of data in that average). The distance surveyed has also increased over time, but much of the additional survey effort is along beaches that are not frequently used by leatherbacks for nesting (i.e., southwest and Panhandle coastlines).
The appropriate multilevel model was selected based on all data (30 years at 68 beaches). Normality assumptions in the second level of the model were tested and found to be valid. The model with the lowest DIC value included the beach-level quadratic covariate for latitude and the overdispersion parameter. The final model was thus as follows:
When Eq. 2 is considered as the second level of the multilevel model, partial pooling allows beaches with little data to borrow information from other beaches, resulting in the estimates of trend at these beaches to be closer to the overall average than they would be if a regression model were fitted to that beach alone (no pooling). By including latitude as a beach-level covariate for the beach trends, partial pooling allows for trend estimates for beaches with little data to be adjusted to a trend that would be expected, given the latitude of the beach.
We assessed goodness of fit of the most parsimonious model for each season using Bayesian P values (Gelman et al. 1996, Morgan 2000), with deviance as the goodness-of-fit criteria. At each iteration of the Markov chain Monte Carlo (MCMC), we calculated the deviance of observed data (D) and the deviance of data simulated from the model, given the current parameter values (D*). The Bayesian P value measures the proportion of times D* when is greater than D. If the model fits the observed data well, then there should be little difference between fits of the observed and simulated data and the P value should be around 0.5. The P value for the most parsimonious model was 0.5014. We examined the posterior distributions for each parameter that we were estimating and found these to be different than our vague priors; therefore, all parameters could be reasonably estimated using this mixed-effects model.
Each variable modeled had an effect on the rate of change at each beach (Table 2). For unit-level predictors, days per week, distance, and days surveyed per year were significant. For example, at a representative beach, for each kilometer surveyed, there was a 4.9% increase in the number of nests recorded.
Table 2. Changes in leatherback nest count by unit-level predictors.
For beach-level predictors, latitude was a significant parameter and was thus included in the final model. Fig. 3 shows the annual change (percentage) in nest counts at each beach with latitude. Although all beaches showed significantly increasing trends, note that the greatest percentage changes occurred from Cape Canaveral south to Martin County (latitude = 27.1° N to 28.7° N). In the final model, the measure of the variance (overdispersion) was 0.641 with a 95% credible interval of 0.587–0.700. This overdispersion value, which is greater than zero, indicates that there may be additional factors that were not accounted for in the model. These factors are undefined but conceivably may include observer error, habitat choices by the turtles, beach renourishment projects (which may affect the amount of available habitat), hurricane effects on the beaches, and so forth.
Finally, results of the trends (percentage change in nest counts each year) at each beach surveyed from south to north are given in Table 1. The average rate of change (percentage change per year) for all beaches was significant at 10.2% (95% CI (credible interval, throughout): 8.5–11.8% over 30 years. For each sample in the MCMC chain, we computed the average of the individual beach trends (β1j; Eq. 3). The mean and interval that contained 95% of the estimates (95% CI) from the 25 000 samples was computed. It should be noted, however, that each beach trend represents the trend for the number of years surveyed at a particular beach and that the trend value is not an extrapolation of the data to the full 30 years. When the number of years surveyed was plotted against beach trends, there was no significant relationship (r2 = 0.0055), but when the number of years was plotted against the width of the credible interval for each beach, there tended to be a tighter credible interval at beaches with more years of data (r2 = 0.3804).
Our purpose was to determine the trend in the number of leatherback nests in Florida since surveys began in the late 1970s. Using a multilevel modeling approach, we found that, statewide, there has been a 10.2% increase in the number of nests per year over the past 30 years. This analysis takes into account 99.4% of all leatherback nesting recorded in Florida (10 005 of 10 065 nests) since 1979. All beaches showed a positive (increasing) trend (Table 1), ranging from a modest increase of 3.1% (Guana Tolomato Matanzas NERR) to a major increase of 16.3% (Vero Beach). It is important to report both the individual beach trends as well as the statewide trend for two reasons. First, it is advantageous for managers to have an estimate of the baseline trend at beaches that are just beginning to receive more leatherback nests. Second, because volunteers and citizen scientists are responsible for collecting the nesting data, it is important to recognize the contribution that their efforts make to estimating trends that will be included in future recovery plans and management objectives.
The number of years surveyed had little effect on the actual trend but did affect the width of the credible interval. For example, beaches with 30 years of nest counts and associated covariates (e.g., Jupiter/Juno Beach, Hutchinson Island) exhibited positive trends that were similar to the overall state trend (10.2%) with small credible intervals. The unit-level covariates (kilometers surveyed, days surveyed per week, days surveyed per year) also affected the trend to some degree, both positively and significantly (see Table 2). Over time, as beach surveyors covered more ground, they were likely to have found more nests with each kilometer surveyed. By surveying for 7 days/week, especially during the main part of the season (April–June), surveyors probably maximized nest counts as well. Although days surveyed/year was a significant predictor of the trend, relatively little change was seen by adding a day to the survey season (0.6%), probably because the survey season is already quite long, and consistently covers the majority of leatherback nesting days. However, because the units are not comparable across predictors (i.e., km, days), the percentage change values cannot be compared between predictors to determine which factor was the most important in influencing nest counts.
The increasing trend in nesting for beaches around Cape Canaveral and south to Martin County was greater than for beaches in other areas (Fig. 3). This may reflect the nesting habitat preferences of leatherbacks. Leatherbacks prefer a steeply sloping beach, with quick access to deep water, that is free of obstacles such as rocks or coral (Pritchard and Trebbau 1984). North of Cape Canaveral there is a wider continental shelf, whereas south of the cape it is quite narrow with the Gulf Stream flowing close to shore.
The multilevel modeling approach was very useful because it allowed us to use all the data that have been collected at Florida beaches over 30 years, while incorporating the variability in survey effort. This approach is the ideal method for analyzing these nest count data. First, this method allowed us to consider the hierarchical structure of the data, thereby accounting for the nonindependence of the observations (nest counts). Second, using a multilevel model allowed us to calculate a robust estimate for those beaches with little data by using information from nearby beaches (at similar latitudes) that had more complete information for longer periods of time. We were also able to estimate the statewide trend. We recommend that other studies with this type of count data combined with uneven effort variables employ a similar multilevel modeling approach for determining trends. Although we could have used classical regression or generalized additive models to analyze the data, there were limitations associated with those methods, including not being able to estimate trends at beaches with low sample sizes (years), and not being able to compute an overall trend for the state that would take into account all beaches with varying effort and variable trends.
The trend that we found in Florida seems to be mirrored throughout other Atlantic Ocean rookeries. In a leatherback stock assessment recently completed, of 11 nesting aggregations analyzed (including Florida), nine populations had increased in recent years (3–24% per year), one had remained stable, and one was decreasing slightly (Troëng et al. 2004, TEWG 2007). The conclusion of the Turtle Expert Working Group report was that leatherbacks are significantly increasing at most nesting beaches in the Atlantic (TEWG 2007).
If leatherbacks are increasing in number throughout the Atlantic, what is the underlying cause? Determining the reason for observed increases is an important step to understanding both the dynamics of endangered species populations and their recovery potential. There is evidence to suggest that both conservation measures implemented in recent decades as well as variable ocean climates may be contributing to these population changes.
The Endangered Species Act (1973) has legally protected leatherbacks for decades in the United States; this policy and combined practical conservation measures are given as the reason for observed increases at some leatherback rookeries. For example, at St. Croix, U.S. Virgin Islands, where leatherbacks have been monitored for 27 years, there has been a 13% increase in the number of females nesting there each year, and the number of hatchlings released has increased from 2000 in early years to over 49 000 annually (Dutton et al. 2005). The remarkable increase in females and hatchlings has been attributed to the protection of nesting females, resulting in relatively high annual survival rates (89.3%; Dutton et al. 2005), and also to the practice of relocating nests in danger of being washed away by high tides (Dutton et al. 2005). In Florida, increased monitoring and individual nest protection may have increased the number of hatchlings that ultimately make it into the water. However, the explanation of increased nest protection in Florida does not hold true for loggerhead turtle nests there; nest counts have declined significantly between 1989 and 2006 (Witherington et al. 2009).
However, nest protection may not be the reason for increases observed elsewhere in the Atlantic. Florida and St. Croix are small populations and are much easier to manage than a rookery with thousands of nesting females. In Trinidad and the Guianas, nests number in the tens of thousands (Girondot and Fretey 1996, Hilterman and Goverse 2005, Eckert 2006), and nesting turtles and eggs are still collected on several beaches. These rookeries are not benefiting from widespread enhanced nest protection, yet nest numbers there are also increasing. Conservation (improved nest monitoring and protection) may be one component of the leatherback increase observed in the Atlantic, but it is probably not the only reason.
Variable oceanographic conditions may also play a role by affecting the abundance of leatherback prey. Leatherbacks forage almost exclusively on gelatinous plankton (Bjorndal 1997, James and Herman 2001, Desjardin 2005, Houghton et al. 2008), mainly in northern latitudes (James and Herman 2001, James et al. 2006). As capital breeders, leatherbacks must meet their energy requirements for migration and maintenance and then devote surplus energy to building fat reserves (Hays 2000) that will allow them to be reproductively active every second or third year (Pritchard 1971, Hirth 1980, van Buskirk and Crowder 1994). The frequency of nesting cycles is therefore believed to be linked to resource availability (Stearns 1992). Recently, increases in jellyfish abundance have been documented in areas once dominated by robust fisheries (Mills 2001, Frank et al. 2005), and the removal of top predators has initiated oceanic food web changes in many areas, including the northwest Atlantic (Worm and Myers 2003). These changes may allow jellyfish to take advantage of primary production that previously was incorporated into fish biomass (Mills 2001). Therefore, the availability of jellyfish may no longer be a limitation faced by leatherbacks. As a consequence, they may be able to store fat reserves more quickly, thus returning to nest more frequently. Additionally, the removal of top predators (Myers et al. 2007) and the collapse of shark populations in the northwest Atlantic (Baum et al. 2003) may decrease at-sea mortality rates for juvenile and subadult leatherbacks, increasing the overall population even more than the effects of the release from food limitation alone.
In contrast with the Atlantic, at eastern Pacific nesting beaches in Mexico and Costa Rica, which once hosted thousands of female leatherbacks per year, extirpation may be imminent as populations have plummeted in recent decades (Eckert and Sarti 1997, Spotila et al. 2000, Spotila 2004). Wallace et al. (2006) found that leatherbacks in the eastern Pacific are likely to be severely constrained by resource limitation, taking longer intervals at sea between nesting years (3.7 ± 0.2 years; Reina et al. 2002) than their counterparts in the Atlantic (Florida = 2.2 ± 0.5 years, Stewart 2007; St. Croix = 2.5 ± 0.5 years, Dutton et al. 2005; Costa Rica = 2.8 ± 0.5 years, Troëng et al. 2004). Wallace et al. (2006) proposed that even though North Atlantic leatherbacks have higher absolute energy costs than eastern Pacific turtles, they have a more consistent foraging environment in the Atlantic basin. The productivity of the Pacific Ocean is more variable and unpredictable due to large-scale phenomena such as the El Niño Southern Oscillation (ENSO; Chavez et al. 2003). Saba et al. (2007, 2008) demonstrated that ENSO events were significantly correlated with the remigration intervals of nesting leatherbacks of the eastern Pacific rookery of Playa Grande, Costa Rica, possibly through the effects of sea surface temperature (SST) variability.
Although the underlying causes of leatherback population increases in the Atlantic remain unclear, the increasing trend in Florida leatherback nesting demonstrates that this endangered sea turtle has the capacity for population growth over relatively short time frames. This potential is important for setting and meeting objectives in the U.S. recovery plan for leatherback turtles (NMFS and USFWS 1992). One of the main objectives states that delisting may be possible if there is a demonstrated increase in the number of nests for 25 years (until 2017) at Culebra, Puerto Rico, St. Croix, USVI and along the east coast of Florida. It has already been shown that St. Croix's population is increasing (Dutton et al. 2005) and now Florida's population may assist in meeting this objective. There are significant obstacles to reaching other goals within the plan, but an increase in nesting is the first and most important target, thus representing a significant step in improving the outlook for the species. Determining what factors have definitively contributed to the population increase in Florida and elsewhere in the Atlantic would help us to better understand the elements underlying population growth potential for this long-lived species, thus allowing us to set more relevant objectives in future recovery plans.
We sincerely thank the thousands of volunteers who have collected leatherback nesting data in the State of Florida since the early 1970s. This work could not be completed every year without the generous contribution from these individuals who give their time, money, and resources to help protect Florida's nesting turtles. We also acknowledge the tremendous foresight of Alan Huff, Charles Futch, and Ross Witham, whose idea it was to start standardized nesting surveys in Florida in the first place. We thank R. Trindell and M. Koperski for managing the FWC permit holder system. We also thank the Florida Sea Turtle License Plate Fund and the U.S. Fish and Wildlife Service for funding to support nesting beach programs. K. Stewart was supported for this work by a Duke University Marine Lab Fellowship. Finally, we acknowledge L. Thorne, A. Frey, P. Dutton, and the two anonymous reviewers whose thoughtful comments substantially improved the quality of the manuscript.