Citation
Phylogeny, biogeography and methodology: a meta-analytic perspective on heterogeneity in adult marine turtle survival rates

Material Information

Title:
Phylogeny, biogeography and methodology: a meta-analytic perspective on heterogeneity in adult marine turtle survival rates
Series Title:
Scientific Reports 8:5852
Creator:
Pfaller, Joseph
Publisher:
Springer Nature
Publication Date:
Language:
English
Physical Description:
Journal Article

Notes

Abstract:
Comparative syntheses of key demographic parameters are critical not only for identifying data gaps, but also for evaluating sources of heterogeneity among estimates. Because demographic studies frequently exhibit heterogeneity, evaluating sources of heterogeneity among estimates can inform biological patterns and conservation actions more broadly. To better understand adult survival in marine turtles and avoid drawing inaccurate conclusions from current estimates, we conducted a comprehensive meta-analysis to test how heterogeneity among estimates was partitioned among phylogenetic, biogeographic and methodological factors. Fifty-nine studies from five marine turtle species met the minimum selection criteria for inclusion in our meta-analysis. Heterogeneity among survival estimates was first partitioned between differences in ocean basin (Indo-Pacific versus Atlantic), then by differences among family/tribe within the Indo-Pacific (Chelonini versus Carettini and Dermochelidae). However, apparent differences attributed to biogeography (ocean basin effect) and phylogeny (family/tribe effect) were highly correlated with methodological differences in tag type, model type, habitat type and study duration, thereby confounding biological interpretations and complicating efforts to use many current survival estimates in population assessments. Our results highlight the importance of evaluating sources of heterogeneity when interpreting patterns among similar demographic studies and directly inform efforts to identify research priorities for marine turtles globally.
Acquisition:
Collected for University of Florida's Institutional Repository by the UFIR Self-Submittal tool. Submitted by Joseph Pfaller.

Record Information

Source Institution:
University of Florida Institutional Repository
Holding Location:
University of Florida
Rights Management:
Copyright Creator/Rights holder. Permission granted to University of Florida to digitize and display this item for non-profit research and educational purposes. Any reuse of this item in excess of fair use or other copyright exemptions requires permission of the copyright holder.

Downloads

This item is only available as the following downloads:


Full Text

PAGE 1

1 Phylogeny, biogeography and methodology: a meta-analytic perspective on heterogeneity in adult marine turtle survival ratesJoseph B. Pfaller n n wx, Milani Chaloup kay, Alan B. Bolten n n x & Karen A. Bjorndal n n xComparative syntheses of key demographic parameters are critical not only for identifying data gaps, but also for evaluating sources of heterogeneity among estimates. Because demographic studies frequently exhibit heterogeneity, evaluating sources of heterogeneity among estimates can inform biological patterns and conservation actions more broadly. To better understand adult survival in marine turtles and avoid drawing inaccurate conclusions from current estimates, we conducted a comprehensive meta-analysis to test how heterogeneity among estimates was partitioned among ’Š›Ž‘‰‡‡–‹…„‹‘‰‡‘‰”ƒ’Š‹…ƒ†‡–Š‘†‘Ž‘‰‹…ƒŽˆƒ…–‘”•t‹ˆ–›‹‡•–—†‹‡•ˆ”‘¤‡ƒ”‹‡–—”–Ž‡ species met the minimum selection criteria for inclusion in our meta-analysis. Heterogeneity among •—”‹ƒŽ‡•–‹ƒ–‡•™ƒ•¤”•–’ƒ”–‹–‹‘‡†„‡–™‡‡†‹¡‡”‡…‡•‹‘…‡ƒ„ƒ•‹f†‘ƒ…‹¤… versus –Žƒ–‹…f–Š‡„›†‹¡‡”‡…‡•ƒ‘‰ˆƒ‹Ž›–”‹„‡™‹–Š‹–Š‡f†‘ƒ…‹¤…Š‡Ž‘‹‹ versus Carettini ƒ†‡”‘…Š‡Ž‹†ƒ‡f‘™‡‡”ƒ’’ƒ”‡–†‹¡‡”‡…‡•ƒ––”‹„—–‡†–‘„‹‘‰‡‘‰”ƒ’Š›‘…‡ƒ„ƒ•‹‡¡‡…–f ƒ†’Š›Ž‘‰‡›ˆƒ‹Ž›–”‹„‡‡¡‡…–f™‡”‡Š‹‰ŠŽ›…‘””‡Žƒ–‡†™‹–Š‡–Š‘†‘Ž‘‰‹…ƒŽ†‹¡‡”‡…‡•‹–ƒ‰ type, model type, habitat type and study duration, thereby confounding biological interpretations ƒ†…‘’Ž‹…ƒ–‹‰‡¡‘”–•–‘—•‡ƒ›…—””‡–•—”‹ƒŽ‡•–‹ƒ–‡•‹’‘’—Žƒ–‹‘ƒ••‡••‡–•—””‡•—Ž–• highlight the importance of evaluating sources of heterogeneity when interpreting patterns among •‹‹Žƒ”†‡‘‰”ƒ’Š‹…•–—†‹‡•ƒ††‹”‡…–Ž›‹ˆ‘”‡¡‘”–•–‘‹†‡–‹ˆ›”‡•‡ƒ”…Š’”‹‘”‹–‹‡•ˆ‘”ƒ”‹‡–—”–Ž‡• globally. Demographic studies in ecology and conservation biology form the basis for assessing population viability and managing ecological risk1. However, estimates of key demographic parameters, such as survival and recruitment, viewed in isolation oen provide limited and/or potentially biased inferences. In this regard, comparative syntheses of similar demographic studies are critical not only for identifying data gaps, but also for evaluating sources of heterogeneity among estimates. Because demographic studies frequently exhibit heterogeneity due to system-specic nature of biological phenomena2 and study-specic dierences in methodology, identifying important sources of heterogeneity can inform biological patterns and conservation actions more broadly2 – 4. Conversely, comparative studies that fail to evaluate sources of heterogeneity risk drawing inaccurate conclusions and misleading management decisions. We conducted a systematic review and comprehensive meta-analysis5 6 of annual survival rates for adult marine turtles to generate precision-weighted, species-specic estimates and prediction intervals from existing data7 and to explicitly model sources of heterogeneity among estimates2. In theory, because all marine turtles exhibit conserved life history patterns, including slow growth and delayed sexual maturity8, adult survival rates should be high (0.90) and exhibit limited natural heterogeneity among estimates. Our goal was to quantify heterogeneity among survival estimates and test how heterogeneity is inuenced by (a) phylogeny, (b) biogeography and (c) methodology. If heterogeneity is inuenced by phylogeny or biogeography, then biological dierences in survival among species/populations may exist. In this case, survival estimates that are lower than expected would highlight important speciesor region-specic hotspots in adult mortality – areas of conservation concern. wCaretta Research Project, Savannah, GA, USA. xArchie Carr Center for Sea Turtle Research and Department of Biology, University of Florida, Gainesville, FL, USA. yEcological Modelling Services Pty Ltd, University of Queensland, St. Lucia, Queensland, Australia. Correspondence and requests for materials should be addressed to J.B.P. (email: Œ’ˆƒŽŽ‡”;—‡†— ) Received: 18 August 2017 Accepted: 28 March 2018 Published: xx xx xxxxb

PAGE 2

2 However, if heterogeneity is also strongly inuenced by methodological dierences, then statistical biases related to certain methodologies introduce articial variation and mask important biological dierences. In this case, the use of survival estimates that are lower than expected may mislead management decisions if methodological biases are not accounted for. We expect the ndings of this synthesis to highlight the importance of evaluating sources of heterogeneity when interpreting demographic estimates2 and directly inform eorts to identify research priorities for marine turtles globally9–12.MethodsLiterature review and selection criteria. We followed the PRISMA protocols for assembling a dataset suitable for meta-analytic evaluation5 6. Specically, we conducted a two-tiered literature search to compile annual survival probability estimates for adult marine turtles. A structured search was conducted in Google Scholar, Sea Turtle Document Library (seaturtle.org) and Sea Turtle Online Bibliography (Archie Carr Center for Sea Turtle Research, University of Florida) using the following Boolean search terms: survival, survivorship, mortality, and the names of the seven marine turtle genera. en, an unstructured literature search was conducted by reviewing the reference lists of all the relevant publications and reports from the structured search. References compiled from the structured and unstructured search were reviewed and only those that estimated annual survival probabilities for adult turtles were retained (i.e., multiyear or stage-based estimates and estimates for juveniles were excluded). Seventy-eight survival estimates were found in the global literature review with 59 meeting the minimum selection criteria for inclusion in the meta-analysis (Supplementary TableS1). Marine turtle taxonomy/phylogeny follows Duchene et al .13. Dermochelidae (leatherbacks) is sister to Chelonidae, which includes Chelonini (atback and green turtles) and Carettini (hawksbill, loggerhead and ridley turtles). e measure of eect size and precision of each estimate was the study-specic inverse-precision weighted annual survival probability and the associated standard error, respectively.fˆ‘”ƒ–‹‡…‘ƒ”‹ƒ–‡• e following potentially informative covariates (or moderators) were compiled for all survival estimates: specic publication ID, research group (based on common authors), type of publication (journal or report), publication year, taxonomic group (species, tribe and family), study site, geographic region with subgrouping for two oceans (Indo-Pacic and Atlantic), study duration (years), habitat type (nesting or foraging), tag type and statistical estimation type with subgrouping for modeling procedures versus enumeration calculations. Each estimate was categorized by whether tag loss, imperfect detection and temporary emigration were explicitly accounted for or not. e population of interest in each study was also characterized in terms of population size, population trend, direct harvest history and sheries bycatch impact using primarily Wallace et al .14 ,15. Details on covariate characterization can be found in the Supplementary Methods. Study duration and publication year were also used to account for various forms of publication bias16,17.Statistical modeling approach. We used a precision-weighted random-eects model approach to summarize the 59 survival estimates into species-specic subgroups before accounting for informative covariates17– 19. In this 2-level analysis, the random eect was the specic study. Each study-specic survival estimate was a response variable so-bound between 0–1. Response variables were therefore logit-transformed prior to tting the random-eects models to ensure the predicted estimates were also between 0–120 and then back-transformed for any predicted summaries. Species-specic random-eects models were tted using the metafor package for R21 and displayed in a subgroup forest plot that was augmented with the species-specic random eects estimates and the prediction intervals for those estimates7. Because many of the potentially informative covariates in our meta-analysis were highly correlated, we could not isolate individual eects or explicitly model interactions among covariates simultaneously. erefore, we rst used a recursive partitioning or conditional inference regression tree approach22, 23 to explore underlying patterns among the potential eects of 12 correlated covariates on the 59 survival estimates. e conditional inference tree model was tted and then displayed using the partykit package for R24 with a minimum split criterion of 0.95 (so P-value t t 0.05). Because the conditional inference tree approach does not account for the precision of survival estimates, we also tested a series of meta-regression models using metafor to explore other potentially informative covariates when estimate precision was explicitly considered. Models with dierent predictors were compared based on Akaike Information Criterion (AICc). We then used a mixed-eects meta-regression approach19, 25 to model annual survival rates conditioned on the main interaction eect – ‘ocean by family/tribe’ – identied within the conditional inference tree and four additional moderators identied by exploratory meta-regression modeling – tag type, model type, habitat type and study duration. We also included a 3-level hierarchical structure26 27 to test for potential non-independence between studies conducted by the same research group (based on common authors), an important assumption to account for in meta-analytic studies17, 28. In this 3-level hierarchical analysis, the random eects were study within research group. e mixed eects meta-regression models were tted using a multivariate parameterization to accommodate assessment of random eect structures29. e model-predicted study-specic inverse-precision weighted annual survival estimates and covariate effects were summarized as covariate-specific boxplots. Continuous variables, like study duration, were modeled using B-splines (splines package)30 to account for potential nonlinear functional forms29 and predicted estimates were summarized using a cubic regression spline GAM-based smoother31 implemented within the ggplots2 package for R32. e I2 statistic2 33 was used to assess the level of unexplained heterogeneity estimated in the 3-level hierarchical meta-regression model t to the 59 studies, and a simple R2 measure appropriate for the multivariate parameter ization of a meta-regression model was used as a metric of the overall model t. e Cochrane QE test was used as a formal test of residual heterogeneity34, while an omnibus F -test was used to test for signicance of the set of

PAGE 3

3 all covariates included in the meta-regression model35. Lastly, we explored potential publication bias16, 17 using a contour-enhanced funnel plot36 of the model predicted study-specic survival estimates using metafor .Data availability. e dataset supporting this article has been uploaded as part of the electronic supplementary material.Results–—†›ƒ†•’‡…‹‡••’‡…‹¤…•—”‹ƒŽ‡•–‹ƒ–‡• e 59 species-specic survival rate estimates derived from the random-eects model are summarized in Fig. 1 Here the random-eect was the specic study. It is apparent that there was considerable species-specic heterogeneity in adult survival rate estimates. is was especially apparent for green and leatherback turtles that might be attributable to geographic and methodological factors. is apparent heterogeneity was explored further using hierarchical or multilevel meta-regression modeling to account for potentially informative phylogenetic4, geographic and methodological covariates. We have also included the species-specic prediction intervals in the subgroup forest plot, which shows for instance that a new atback adult survival rate study would be expected to provide an annual estimate ca. 0.93 (95% prediction interval: 0.77–0.98). On the other hand, for leatherbacks it would be 0.85 (95% prediction interval: 0.45–0.97).fˆ‘”ƒ–‹‡…‘ƒ”‹ƒ–‡• e conditional inference tree approach identied two signicant nodes at a minimum split criterion of 0.95: (Node 1) Atlantic versus Indo-Pacic oceans and (Node 2) Chelonini versus Carettini and Dermochelidae nested within the Indo-Pacic Ocean (Fig. 2). No other covariates remained in the nal tree model at 0.95. e three terminal nodes are boxplot summaries of the model-derived survival estimates (n t t sam ple size within a terminal node). In addition to ocean and species, the best-tting models from the exploratory meta-regression approach (estimates weighted by precision) included tag type (and tag loss), model type, habitat type and study duration, suggesting the methodology as well as phylogeny and biogeography might have impor tant eects on adult marine turtle survival rates.‘ƒ”‹ƒ–‡•’‡…‹¤…‘†‡Ž’”‡†‹…–‡†•—”‹ƒŽ‡•–‹ƒ–‡• e 3-level hierarchical meta-regression model (random eects t t study within research group) was a better t than a 2-level model (random eect t t study) (log-likelihood ratio test t t 58.1, df t t 2, P t t 0.001). e 3-level model with a nonlinear functional form for study duration was a better t than the same model with a linear functional form (log-likelihood ratio test t t 19.7, df t t 2, P t t 0.001). e inclusion of the six moderators in the accepted 3-level, nonlinear regression model led to a signicant improvement in overall model t (QE t t 121.4, df t t 14, P t t 0.001). However, ca. 97% of unaccounted variance was due to residual heterogeneity (tau2 t t 0.005, 95% CI: 0.003–0.011; I2 t t 97.2), and only ca. 37% of the residual heterogeneity in the accepted model was accounted for by inclusion of the six moderators (R2 t t 0.37). is high level of residual heterogeneity was signicant (QE t t 508.6, df t t 44, P t t 0.001) and apparently typical of ecological studies2,28. Covariate-specic survival estimates were derived from the 3-level hierarchical meta-regression model t to the 59 studies conditioned on family/tribe (3 levels), ocean (2 levels), tag type (4 levels), method type (5 levels), habitat type (2 levels) and the study duration by tag type interaction term (Fig. 3 ). General patterns among model-predicted survival estimates include: (1) Chelonini Carettini ~ Dermochelidae (Fig. 3a), (2) Indo-Pacic Atlantic (Fig. 3b), (3) titanium inconel ~ PIT monel (Fig. 3c), (4) statistical modeling enumeration (Fig. 3d) and (5) foraging areas nesting beaches (Fig. 3e). While tag type was mostly ocean-specic, study duration appears to have no eect or a negative eect on survival estimates from metal tags and a positive eect on survival estimates from PIT tags, though most are of relatively short duration (Fig. 3f ).Publication bias. A contour-enhanced funnel plot of the predicted study-specic survival estimates derived from the 3-level hierarchical meta-regression model t to the 59 studies showed no evidence of any form of publication bias (Supplementary Fig.S1). ere was also no temporal eect of publication year in the conditional inference tree or the exploratory meta-regression approach. erefore, we found no evidence of publication bias in the marine turtle survival rates for which we could test for using a range of approaches.DiscussionAnnual survival estimates for adult marine turtles exhibit considerable heterogeneity. Survival estimates were generated from ve turtle species, three major oceans and ve decades of research involving numerous methodological techniques. For this reason, heterogeneity among estimates might be expected. However, because adult survival is a key demographic parameter and marine turtles exhibit conserved life-history patterns, including slow growth and delayed sexual maturity8, the extent of heterogeneity found among survival estimates com plicates accurate biological interpretations. To better understand heterogeneity in adult marine turtle survival rates, we conducted a comprehensive meta-analysis to test how heterogeneity among estimates was partitioned among phylogenetic, biogeographic and methodological factors. Results from this study represent an important step towards identifying the key factors that drive dierences among survival estimates and setting research and management priorities for marine turtles globally11,12.bƒŽ—ƒ–‹‰•‘—”…‡•‘ˆŠ‡–‡”‘‰‡‡‹–› Results from the recursive partitioning or conditional infer ence regression tree approach22, 23 indicate that heterogeneity among survival estimates was rst partitioned between differences in ocean basin (Indo-Pacific versus Atlantic), then by differences among family/tribe within the Indo-Pacic (Chelonini versus Carettini and Dermochelidae). ough oen ignored, phylogenetic non-independence can change the results of ecological meta-analyses4. However, in this case, dierences were not associated with phylogenetic similarity – Carettini is more closely related to Chelonini (both in Chelonidae) than to Dermochelidae13. erefore, survival estimates in Carettini and Dermochelidae are likely similar and

PAGE 4

4 lower than Chelonini due to other biological or methodological factors, not phylogenetic history. Estimates for Indo-Pacic Chelonini might be higher because this group includes atback turtles, which are endemic to the Indo-Pacic, exhibit dierent and potentially less vulnerable habitat-use patterns (exclusively neritic versus both Figure 1. Random-eects forest plot of the inverse-variance weighted annual survival estimates for the 59 marine turtle studies (letters aer reference year indicate dierent estimates from the same reference)37, 40– 42, 51– 88. e species-specic pooled or random-eect survival rate estimates (RE diamonds) are shown in addition to the prediction intervals (horizontal bar through each RE diamond). Plot ordered by eect size within each species and solid square t t survival rate and size of symbol reects relative weighting, horizontal bars t t 95% condence interval of each study-specic survival rate. Colored icons show potentially informative study-specic covariates.

PAGE 5

5 neritic and oceanic), and have been consistently monitored with robust methodologies (see below)37. Estimates specically from the Northwest Atlantic (as there are no estimates from the Northeast or South Atlantic) are lower than the Indo-Pacic. However, this is not consistent with any apparent dierences in natural or anthro pogenic mortality rates or any known dierences in life-history behavior among populations in dierent oceans. Predictors that might be indicative of dierences in adult mortality among turtle populations such as harvest history and sheries bycatch impact14, 15 were not found to contribute to heterogeneity among survival estimates (Fig. 2 ). erefore, our results suggest that while heterogeneity is connected to biogeography (ocean eect) and phylogeny (family/tribe eect), dierences cannot be attributed to phylogenetic similarity or apparent regional dierences in adult mortality. Instead, regionand species-specic dierences appear to be tightly linked to the use of dierent methodologies applied in dierent oceans and therefore to a dierent composition of turtle species/populations. e ocean and family/tribe predictors were highly correlated with methodological predictors, including tag type, method/model type, habitat type and study duration. Results from exploratory meta-regression modeling showed that these four factors, in addition to ocean and species, were consistently found within the best-tting models. Direct monitoring of marine turtles in the wild is inherently dicult and CMR studies remain the most reliable tool for estimating annual survival rates. However, such eorts are clearly vulnerable to statistical biases when methodologies cannot account for complexities inherent to the biology and study of marine turtles such as imperfect detection, tag loss, and temporary and permanent emigration. First, our results highlight the impor tance of tag type and its ramications on tag loss on survival estimates. We found that estimates generated from monel tags were consistently lower than estimates generated primarily from inconel, titanium and PIT tags. Estimates generated from monel tags are likely biased low and eected by study duration (Fig. 3f ) due to factors related to higher rates of tag loss and subsequent individual misidentication: monel tags have lower retention rates38, tend to corrode in seawater39 and were oen applied only singly in early studies (e.g.40, 41). Similar concerns with respect to tag loss may apply to inconel tags, although to a lesser extent than monel. Second, our results show that early attempts to estimate survival via enumeration methods that rely on time-series counts of individuals violate far more CMR assumptions than statistical modeling procedures such as CJSRE and MSORD, which are designed to estimate and account for biases including imperfect detection and temporary emigration42, 43. Although robust statistical modeling procedures are well accepted and almost universally used, the accuracy of estimates will always depend on the quality of CMR data to which the models are applied. ird, our results suggest that CMR data collected in foraging areas may be less prone to statistical biases than data collected on nesting beaches. Because marine turtles spend considerably more time in foraging areas, estimates from foraging areas may be more robust to statistical biases associated with temporary, as well as permanent, emigration. Methodological dierences clearly contribute to heterogeneity among survival estimates. Patterns among dierent methodologies and their associated biases are highly correlated with biological patterns: more low estimates come from the Atlantic Ocean, Carettini and Dermochelidae, monel tags, enumeration calculations and nesting beaches, while more high estimates come from the Indo-Pacic ocean, Chelonini, Figure 2. Conditional inference tree visualization of the eect of potentially informative covariates on the study-specic marine turtle survival rates.

PAGE 6

6 titanium tags, statistical modeling procedures and foraging areas (Figs 1 and 4 ). We proposed that if heterogeneity was inuenced by phylogeny or biogeography, then biological dierences in adult survival may exist and management action should be directed towards species/populations associated with survival estimates that are relatively low. However, this interpretation was contingent on whether or not heterogeneity was also inuenced by methodology. Because heterogeneity among survival estimates was strongly inuenced by dierent methodologies that were highly correlated with phylogenetic and biogeographic dierences, we cannot make accurate biological interpretations to condently direct management actions. Whether estimates that are lower than expected indicate elevated rates of adult mortality or simply highlight the use less robust methodologies is currently undecipherable. For this reason, eorts to apply many current survival estimates to model population viability and interpret long-term trends risk drawing inaccurate conclusions and misleading management actions. is highlights the need to critically evaluate current survival estimates and attempt to correct or account for statistical biases when possible. Additionally, future estimates should strive to apply robust methodologies Figure 3. Multi-panel display summarizing the covariate-specic model-predicted estimates derived from the 3-level hierarchical meta-regression model t to the 59 studies conditioned on the following covariates: (a) ocean (2 levels), (b) family (3 levels), (c) tag type (4 levels), (d) model type (5 levels), (e) habitat type (2 levels) and ( f) study-duration by tag-type interaction term. Closed blue dots in the boxplots shows the predicted mean, horizontal bar t t predicted median. e underlying trend in the tag-specic study-duration subplots (bottom right panel) shown by a cubic regression spline smooth with an inverse-precision weighted 95% condence polygon.

PAGE 7

7 that eliminate or minimize statistical biases, so that the detection of important regionor species-specic dier ences in adult mortality are not masked by heterogeneity associated with dierences in methodology. Because adult mortality from over-harvesting and sheries bycatch remains a signicant conservation concern for marine turtles14, 15, more work is needed to generate new, more comparable estimates that will improve the accuracy of assessments of the status and trends in marine turtle populations9,10,44 and help guide conservation measures and management in the future.f†‡–‹ˆ›‹‰†ƒ–ƒ‰ƒ’• Our systematic review of annual survival estimates for adult marine turtles revealed important regionand species-specic data gaps that should be high priorities for research to support conservation11, 12. At the region-level for all applicable species, there is a noticeable lack of estimates from the eastern North Atlantic (including the Mediterranean Sea), South Atlantic, western and northern Indian Ocean (including the Red Sea and Persian Gulf), western North Pacic and eastern South Pacic (Fig. 4 ), all areas that support globally important marine turtle populations14 and some that also host long-term research and monitoring projects capable of generating estimates of adult survival rates. At the species-level, we found estimates for all seven marine turtle species. However, all estimates for Kemp’s and olive ridley turtles (n t t 4 and 1, respectively) were excluded for not meeting the minimum selection criteria for inclusion in the meta-analysis. Because these estimates were not deemed suciently robust and all are at least two decades old, new estimates for ridleys should be a research priority, especially given their susceptibility to overexploitation from human threats such as sheries bycatch45. Globally, estimates from nesting beaches are far more frequent than estimates from foraging habitats, which are primarily from eastern Australia (Fig. 4 ). Although logistically more challenging, priority should be given to estimating survival rates in foraging habitats, especially in the well-studied northwest Atlantic where estimates from nesting beaches overwhelmingly predominate. Survival estimates from unstudied regions and species will not only improve assessments of specic populations, but will also improve our understanding of the important factors that aect survival rates and allow us to better decipher biological versus methodological sources of heterogeneity among estimates. New estimates generated from traditional CMR data should avoid methodologies that may generate articially low estimates of survival rates if possible and, if used, should acknowledge and attempt to account for statistical biases associated with those methods. Future studies may strive to implement standardized robust methodologies that eliminate or minimize potential biases and isolate the biological signal resulting from actual sources of mortality. ough not yet practically implementable in most cases, the standardized use of population-wide genetic “tagging” applied in both nesting and foraging areas holds great potential for future studies. Estimates using this approach generated across many regions and species would be directly comparable, allowing researchers and managers to identify regional hotspots in marine turtle mortality and implement more eective management strategies at a global scale.The value of meta-analysis. Meta-analysis is a powerful tool ideally suited for combining the results of demographic studies. Because demographic studies in ecology and conservation biology oen exhibit consid erable heterogeneity2, estimates of key parameters, such as adult survival rates, viewed in isolation may provide limited and/or potentially biased inferences. Our results show the value of meta-analysis in generating robust species-specic demographic estimates and identifying data gaps, but also highlight the importance of evaluating sources of heterogeneity when interpreting patterns among similar demographic studies. Because demographic studies frequently exhibit heterogeneity due to system-specic nature of biological phenomena2 and Figure 4. Global geographic distribution of the 59 survival estimates included in the meta-analysis. Icon shapes indicate dierent marine turtle species and colors indicate dierent habitats (foraging grounds versus nesting beaches). e map was generated using the Maptool function at www.seaturtle.org89.

PAGE 8

8 study-specic dierences in methodology, identifying important sources of heterogeneity can inform biological patterns and conservation actions more broadly2 – 4. Synthesized data are essential for modelling wildlife population dynamics exposed to various anthropogenic hazards46 and for testing ecological hypotheses such as the importance or not of dispersal for species exposed to habitat loss47. Because there are currently few meta-analyses of wildlife demographic rates3,48, this approach represents a productive and important area for future research. Global research on marine turtles is poised to benet from the application of meta-analysis. e accumulation of demographic and biological information from decades of research across all species and biogeographic regions has elicited global assessments and numerous review articles in recent years (e.g.14, 15, 44, 49, 50). Meta-analytic comparisons of key parameters and threats associated with the biology and conservation of marine turtles such as breeding and recruitment rates, population and individual growth rates, and sheries bycatch and plastic ingestion rates will reveal novel patterns and valuable insights, while accounting for important sources of heterogeneity that may obscure important interpretations2 – 4 25, 28. Such analyses serve to guide research and conservation priorities for marine turtles into the future.Referencest 1.t Burgman, M. A., Ferson, S. & Aaaya, H. is assessment in conservation biology. (Chapman & Hall, 1993).t 2.t Senior, A. M. et al Heterogeneity in ecological and evolutionary meta-analyses: its magnitude and implications. Ecology 97, 3293–3299, https://doi.org/10.1002/ecy.1591 (2016).t 3.t Boyce, M. S., Irwin, L. L. & Barer, Demographic meta-analysis: synthesizing vital rates for spotted owls. J. Appl. Ecol. 42, 38–49, https://doi.org/10.1111/j.1365-2664.2004.00986.x (2005).t 4.t Chamberlain, S. A. et al Does phylogeny matter? Assessing the impact of phylogenetic information in ecological meta-analysis. Ecol. Lett. 15, 627–636, https://doi.org/10.1111/j.1461-0248.2012.01776.x (2012).t 5.t Liberati, A. et al e PISMA statement for reporting systematic reviews and meta-analyses of studies that evaluate health care interventions: explanation and elaboration. PLoS Med. 6 doi: e1000100 (2009).t 6.t Vetter, D., cer, G. & Storch, I. Meta-analysis: a need for well-dened usage in ecology and conservation biology. Ecosphere 4 https://doi.org/10.1890/ES13-00062.1 (2013).t 7.t IntHout, J., Ioannidis, J. P., overs, M. M. & Goeman, J. J. Plea for routinely presenting prediction intervals in meta-analysis. BMJ Open 6, https://doi.org/10.1136/bmjopen-2015-010247 (2016).t 8.t Stearns, S. e Evolution of Life Histories (Oxford University Press, 1992).t 9.t National esearch Council. Assessment of sea-turtle status and trends: integrating demography and abundance. (National Academies Press, 2010).t 10.t Bjorndal, A. et al Better science needed for restoration in the Gulf of Mexico. Science 331, 537–538, https://doi.org/10.1126/ science.1199935 (2011).t 11.t Hamman, M. et al Global research priorities for sea turtles: informing management and conservation in the 21st century. Endang. Species es. 11, 245–269, https://doi.org/10.3354/esr00279 (2010).t 12.t ees, A. F. et al Are we woring towards global research priorities for management and conservation of sea turtles? Endang. Species es. 31, 337–382, https://doi.org/10.3354/esr00801 (2017).t 13.t Duchene, S. et al Marine turtle mitogenome phylogenetics and evolution. Mol. Phylogenet. Evol. 65, 241–250, https://doi. org/10.1016/j.ympev.2012.06.010 (2012).t 14.t Wallace, B. P. et al Global conservation priorities for marine turtles. PLoS One 6 e24510, https://doi.org/10.1371/journal. pone.0024510 (2011).t 15.t Wallace, B. P. et al Impacts of sheries bycatch on marine turtle populations worldwide: toward conservation and research priorities. Ecosphere 4, https://doi.org/10.1890/ES12-00388.1 (2013).t 16.t Murtaugh, P. A. Journal quality, effect size, and publication bias in meta-analysis. Ecology 83, 1162–1166, https://doi. org/10.1890/0012-9658(2002)083[1162:JQESAP]2.0.CO;2 (2002).t 17.t Naagawa, S. & Santos, E. S. A. Methodological issues and advances in biological meta-analysis. Evol. Ecol. 26, 1253–1274, https:// doi.org/10.1007/s10682-012-9555-5 (2012).t 18.t Gurevitch, J., Curtis, P. S. & Jones, M. H. Meta-analysis in ecology. Adv. Ecol. es. 32, 199–247, https://doi.org/10.1016/S00652504(01)32013-5 (2001).t 19.t Sutton, A. J. & Higgins, J. ecent developments in meta-analysis. Stat. Med. 27, 625–650, https://doi.org/10.1002/sim.2934 (2008).t 20.t ieschnic, & McCullough, B. D. egression analysis of variates observed on (0, 1): Percentages, proportions and fractions. Stat. Model. 3, 193–213, https://doi.org/10.1191/1471082X03st053oa (2003).t 21.t Viechtbauer, W. Conducting meta-analyses in with the metafor pacage. J. Stat. Sow. 36, 1–48, https://doi.org/10.18637/jss.v036. i03 (2010).t 22.t Strobl, C., Malley, J. & Tutz, G. An introduction to recursive partitioning: rationale, application, and characteristics of classication and regression trees, bagging, and random forests. Psychol. Methods 14, 323–348, https://doi.org/10.1037/a0016973 (2009).t 23.t Seibold, H., Zeileis, A. & Hothorn, T. Model-based recursive partitioning for subgroup analyses. Int. J. Biostat. 12, 45–63, https://doi. org/10.1515/ijb-2015-0032 (2016).t 24.t Hothorn, T. & Zeileis, A. Partyit: A modular toolit for recursive partitioning in J. Mach. Learn. es. 16, 3905–3909 (2015).t 25.t van Houwelingen, H. C., Arends, L. & Stijnen, T. Advanced methods in meta-analysis: multivariate approach and metaregression. Stat. Med. 21, 589–624, https://doi.org/10.1002/sim.1040 (2002).t 26.t onstantopoulos, S. Fixed eects and variance components estimation in three-level meta-analysis. es. Synth. Methods 2 61–76, https://doi.org/10.1002/jrsm.35 (2011).t 27.t Tuc, S. L. et al Land-use intensity and the eects of organic farming on biodiversity: a hierarchical meta-analysis. J. Appl. Ecol. 51, 746–755, https://doi.org/10.1111/1365-2664.12219 (2014).t 28.t Gilman, E., Chaloupa, M., Swimmer, Y. & Piovano, S. A cross-taxa assessment of pelagic longline by-catch mitigation measures: conicts and mutual benets to elasmobranchs. Fish Fish. 17, 748–784, https://doi.org/10.1111/faf.12143 (2016).t 29.t Gasparrini, A., Armstrong, B. & enward, M. G. Multivariate meta-analysis for non-linear and other multi-parameter associations. Stat. Med. 31, 3821–3839, https://doi.org/10.1002/sim.5471 (2012).t 30.t Core Team. : A language and environment for statistical computing. Foundation for Statistical Computing, Vienna, Austria. http://www.-project.org/ (2017).t 31.t Wood, S. N. Generalized additive models: an introduction with (CC Press, 2017).t 32.t Wicham, H. ggplot2: Elegant graphics for data analysis. (Springer, 2016).t 33.t Higgins, J. & ompson, S. G. Quantifying heterogeneity in a meta-analysis. Stat. Med. 21 1539–58, https://doi.org/10.1002/ sim.1186 (2002).t 34.t Viechtbauer, W. & Cheung, M. W-L. Outlier and influence diagnostics for meta-analysis. es. Synth. Methods 1 https://doi. org/10.1002/jrsm.11 (2010).

PAGE 9

9 t 35.t Viechtbauer, W., Lpez-Lpez, J. A., Snchez-Meca, J. & Marn-Martnez, F. A comparison of procedures to test for moderators in mixed-eects meta-regression models. Psychol. Methods 20, 360–374, https://doi.org/10.1037/met0000023 (2015).t 36.t Peters, J. L., Sutton, A. J., Jones, D. ., Abrams, . & ushton, L. Contour-enhanced meta-analysis funnel plots help distinguish publication bias from other causes of asymmetry. J. Clin. Epidemiol. 61, 991–996, https://doi.org/10.1016/j.jclinepi.2007.11.010 (2008).t 37.t Groom, ., Griths, A. & Chaloupa, M. Estimating long-term trends in abundance and survival for nesting atbac turtles in aadu National Par, Australia. Endang. Species es. 32, 203–211, https://doi.org/10.3354/esr00795 (2017).t 38.t Limpus, C. J. Estimation of tag loss in marine turtle research. Wildlife es. 19, 457–469, https://doi.org/10.1071/W9920457 (1992).t 39.t Balazs, G. H. Factors aecting the retention of metal tags on sea turtles. Marine Turtle Newsletter 20, 11–14 (1982).t 40.t Bjorndal, A. Demography of the breeding population of the green turtle, Chelonia mydas at Tortuguero, Costa ica. Copeia 1980, 525–530, https://doi.org/10.2307/1444530 (1980).t 41.t Chua, T. H. Nesting population and frequency of visits in Dermochelys coriacea in Malaysia. J. Herpetol. 22 192–207, https://doi. org/10.2307/1563998 (1988).t 42.t endall, W. L. & Bjorland, Using open robust design models to estimate temporary emigration from capture-recapture data. Biometrics 57, 1113–1122, https://doi.org/10.1111/j.0006-341X.2001.01113.x (2001).t 43.t Converse, S. J., endall, W. L., Doherty, P. F. Jr. & yan, P. G. Multistate models for estimation of survival and reproduction in greyheaded albatross (alassarche chrysoma ). Au 126, 77–88, https://doi.org/10.1525/au.2009.07189 (2009).t 44.t Mazaris, A. D., Schoeld, G., Gazinou, C., Almpanidou, V. & Hays, G. C. Global sea turtle conservation successes. Sci. Adv. 3 e1600730, https://doi.org/10.1126/sciadv.1600730 (2017).t 45.t Plotin, P. T. Biology and Conservation of idley sea turtles (e Johns Hopins University Press, 2007).t 46.t Chaloupa, M. & Balazs, G. H. Using Bayesian state-space modelling to assess the recovery and harvest potential of the Hawaiian green sea turtle stoc. Ecol. Model. 207, 93–109, https://doi.org/10.1016/j.ecolmodel.2007.02.010 (2007).t 47.t Quesnelle, P. E., Lindsay, E. & Fahrig, L. Low reproductive rate predicts species sensitivity to habitat loss: a meta-analysis of wetland vertebrates. PLoS One 9, e90926, https://doi.org/10.1371/journal.pone.0090926 (2014).t 48.t Bentez-Lpez, A. et al e impact of hunting on tropical mammal and bird populations. Science 356, 180–183, https://doi. org/10.1126/science.aaj1891 (2017).t 49.t Hays, G. C., Mazaris, A. D. & Schoeld, G. Dierent male vs. female breeding periodicity helps mitigate ospring sex ratio sews in sea turtles. Front. Mar. Sci. 1, 43, https://doi.org/10.3389/fmars.2014.00043 (2014).t 50.t Lee, P. L. M., Schoeld, G., Haughey, I., Mazaris, A. D. & Hays, G. C. A review of patterns of multiple paternity across sea turtle rooeries. Adv. Mar. Biol (in press) https://doi.org/10.1016/bs.amb.2017.09.004 (2017).t 51.t Limpus, C. J., Parmenter, C. J. & Chaloupa, M. Monitoring of Coastal Sea Turtles: GapAnalysis 5. Flatbac turtles, Natator depressus in the Port Curtis and Port Alma egion. eport produced for the Ecosystem esearch and Monitoring Program Advisory Panel as part of Gladstone Ports Corporation’s Ecosystem esearch and Monitoring Program http://gpcl.com.au/EnvironmentDocuments/ EHPMonitoringofCoastalSeaTurtlesGapAnalysis5Flatbac%20turtles.pdf (2013).t 52.t Guinea, M., Chaloupa, M. & Limpus, C. Bare Sand Island atbac turtle capture-mar-recapture program (1996–2014). eport prepared for use in Department of Environment and Energy (Commonwealth of Australia) recovery plan for marine turtles in Australia (Canberra, Australia) (2015).t 53.t Chaloupa, M. Using multistate open robust design statistical modeling approach for estimating ey demographic parameters for the atbac nesting population on Barrow Island, Western Australia (2005/06–2013/14). Society of Petroleum Engineers International Conference on Health, Safety and Environment SPE 156747 (2014).t 54.t Chaloupa, M. Using multistate open robust design statistical modeling approach for estimating ey demographic parameters for the flatbac nesting population at Mundabullangana, Western Australia (2009/10–2013/14). Society of Petroleum Engineers International Conference on Health, Safety and Environment SPE 156747 (2015).t 55.t Semino, J., Jones, T. T., esendiz, A., Nichols, W. J. & Chaloupa, M. Y. Monitoring green turtles (Chelonia mydas ) at a coastal foraging area in Baja California, Mexico: multiple indices describe population status. J. Mar. Biol. Assoc. U 83, 1355–1362, https:// doi.org/10.1017/S0025315403008816 (2003).t 56.t Chaloupa, M. Phase 1Assessment of suitability of Queensland Pars & Wildlife Service Sea Turtle Data for use in models of the population dynamics of the Southern Great Barrier eef Green Turtle Stoc. Great Barrier eef Marine Par Authority esearch Publication 74, 1–49 (2002).t 57.t Limpus, C., Jones, & Chaloupa, M. Fibropapilloma disease in marine turtles in eastern Indian Ocean South Western Pacic Ocean. NOAA Tech Memo NOAA-TM-NMFS-PIFSC-054 (2016).t 58.t Chaloupa, M. & Limpus, C. Estimates of sexand age-class-specic survival probabilities for a southern Great Barrier eef green sea turtle population. Mar. Biol. 146, 1251–1261, https://doi.org/10.1007/s00227-004-1512-6 (2005).t 59.t Piacenza, S. E., Balazs, G. H., Hargrove, S. ., ichards, P. M. & Heppell, S. S. Trends and variability in demographic indicators of a recovering population of green sea turtles. Endang. Species es. 31, 103–117, https://doi.org/10.3354/esr00753 (2016).t 60.t Trong, S. & Chaloupa, M. Variation in adult annual survival probability and remigration intervals of sea turtles. Mar. Biol. 151, 1721–1730, https://doi.org/10.1007/s00227-007-0611-6 (2007).t 61.t Campbell, C. L. & Lagueux, C. J. Survival probability estimates for large juvenile and adult green turtles (Chelonia mydas ) exposed to an artisanal marine turtle shery in the western Caribbean. Herpetologica 61, 91–103, https://doi.org/10.1655/04-26 (2005).t 62.t Solow, A. ., Bjorndal, A. & Bolten, A. B. Annual variation in nesting numbers of marine turtles: the eect of sea surface temperature on re-migration intervals. Ecol. Lett. 5, 742–746, https://doi.org/10.1046/j.1461-0248.2002.00374.x (2002).t 63.t Garca-Cruz, M. A. et al Population trends and survival of nesting green sea turtles Chelonia mydas on Aves Island, Venezuela. Endang. Species es. 29, 103–116, https://doi.org/10.3354/esr00695 (2015).t 64.t Penaloza, C. Demografa y viabilidad de las poblacin de tortuga verde, Chelonia myda s, en Isla deAves. Universidad Simn Bolvar (2000).t 65.t ichardson, J. I., Bell, & ichardson, T. H. Population ecology and demographic implications drawn from an 11-year study of nesting hawsbill turtles, Eretmochelys imbricata at Jumby Bay, Long Island, Antigua, West Indies. Chelonian Conserv. Biol. 3 244–250 (1999).t 66.t Prince, I. T. & Chaloupa, M. Estimating demographic parameters for a critically endangered marine species with frequent reproductive omission: hawsbill turtles nesting at Varanus Island, Western Australia. Mar. Biol. 159, 355–363, https://doi. org/10.1007/s00227-011-1813-5 (2012).t 67.t Bell, I., Schwarzopf, L. & Manicom, C. High survivorship of an annually decreasing aggregation of hawsbill turtles, Eretmochelys imbricata found foraging in the northern Great Barrier eef. Aquat. Conserv. 22, 673–682, https://doi.org/10.1002/aqc.2245 (2012).t 68.t Limpus, C. & Chaloupa, M. Moreton Bay loggerhead turtle capture-mar-recapture program (1990–2008). eport produced for the Queensland Department of Environment and Heritage (2008).t 69.t Heppell, S. S., Limpus, C. J., Crouse, D. T., Frazer, N. B. & Crowder, L. B. Population model analysis for the loggerhead sea turtle, Caretta caretta in Queensland. Wildlife es. 23, 143–159, https://doi.org/10.1071/W9960143 (1996).t 70.t Chaloupa, M. & Limpus, C. Survival probability estimates for the endangered loggerhead sea turtle resident in southern Great Barrier eef waters. Mar. Biol. 140, 267–277, https://doi.org/10.1007/s002270100697 (2002).t 71.t ichardson, J. I. & ichardson, T. H. An experimental population model for the loggerhead sea turtle (Caretta caretta) in Biology and conservation of sea turtles (ed. Bjorndal, A.) 165–176 (Smithsonian Institution Press, 1995).

PAGE 10

10 t 72.t Pfaller, J. B. et al Accounting for imperfect detection is critical for inferring marine turtle nesting population trends. PLoS One 8 e62326, https://doi.org/10.1371/journal.pone.0062326 (2013).t 73.t Lamont, M. M., Fujisai, I. & Carthy, . Estimates of vital rates for a declining loggerhead turtle (Caretta caretta ) subpopulation: implications for management. Mar. Biol. 161, 2659–2668, https://doi.org/10.1007/s00227-014-2537-0 (2014).t 74.t Mon, M. H., Berson, J. & ivalan, P. Estimating demographic parameters for loggerhead sea turtles using mar-recapture data and a multistate model. Popul. Ecol. 53, 165–174, https://doi.org/10.1007/s10144-010-0205-x (2011).t 75.t TEWG (Turtle Expert Woring Group). An assessment of the loggerhead turtle population in the western North Atlantic Ocean. NOAA Tech Memo NMFS-SEFSC-575 (2009).t 76.t Frazer, N. B. Survivorship of adult female loggerhead sea turtles, Caretta caretta nesting on Little Cumberland Island, Georgia, USA. Herpetologica 39, 436–447 (1983).t 77.t orson, J. T., Punt, A. E. & Nel, Evaluating population recovery for sea turtles under nesting beach protection while accounting for nesting behaviors and changes in availability. J. Appl. Ecol. 49, 601–610, https://doi.org/10.1111/j.1365-2664.2012.02143.x (2012).t 78.t Taylor, B. L. Population viability analysis for the Cumberland Island loggerhead turtle population. NOAA Tech Memo NMFSSEFSC-361 (1995).t 79.t Henry, J. D., Phaneuf, P. J. & Sprayberry, T. H. Survivorship estimates for female loggerhead sea turtles, Caretta caretta nesting on Wassaw and Pine Islands, Georgia. NOAA Tech Memo NMFS-SEFSC-415 (1998).t 80.t Phillips, ., Manseld, L., Die, D. J. & Addison, D. S. Survival and remigration probabilities for loggerhead turtles (Caretta caretta ) nesting in the Eastern Gulf of Mexico. Mar. Biol. 161, 863–870, https://doi.org/10.1007/s00227-013-2386-2 (2014).t 81.t ivalan, P., Prvolt-Julliard, A. C. & Girondot, M. First estimation of the annual female survival rate based on tagging data from French Guiana. NOAA Tech Memo NMFS-SEFSC-567 (2008).t 82.t ivalan, P. et al Trade-o between current eort and delay to next reproduction in the leatherbac turtle. Oecologia 145, 564–574, https://doi.org/10.1007/s00442-005-0159-4 (2005).t 83.t Stewart, . et al Increased nesting, good survival and variable site delity for leatherbac turtles in Florida, USA. Biol. Conserv. 176, 117–125, https://doi.org/10.1016/j.biocon.2014.05.008 (2014).t 84.t Dutton, D. L., Dutton, P. H., Chaloupa, M. & Boulon, H. Increase of a Caribbean leatherbac turtle Dermochelys coriacea nesting population lined to long-term nest protection. Biol. Conserv. 126, 186–194, https://doi.org/10.1016/j.biocon.2005.05.013 (2005).t 85.t Pilcher, N. & Chaloupa, M. Using community-based monitoring to estimate demographic parameters for a remote nesting population of the Critically Endangered leatherbac turtle. Endang. Species es. 20, 49–57, https://doi.org/10.3354/esr00480 (2013).t 86.t Tomillo, M. P. S. Factors aecting population dynamics of eastern Pacic leatherbac turtles (Dermochelys coriacea). Drexel University (2007).t 87.t Dutton, D. L., Dutton, P. H. & Boulon, ecruitment and mortality estimates for female leatherbacs nesting in St. Croix, U.S. Virgin Islands. NOAA Tech Memo NMFS-SEFSC-443 (2000).t 88.t Spotila, J. ., eina, D., Steyermar, A. C., Plotin, P. T. & Paladino, F. V. Pacic leatherbac turtles face extinction. Nature 405, 529530, https://doi.org/10.1038/35014729 (2000).t 89.t SEATUTLE.OG Maptool. SEATUTLE.OG, Inc. http://www.seaturtle.org/maptool/ (9 November 2017).AcknowledgementsWe are grateful to the dedicated research teams that have undertaken the long-term studies from which we have derived demographic data. Without these long-term studies, any analyses such as ours would not be possible. We thank the compliers and maintainers of the Sea Turtle Document Library (seaturtle.org) and the Sea Turtle Online Bibliography (Archie Carr Center for Sea Turtle Research, University of Florida).We thank Chevron Australia for use of the Barrow Island and Mundabullangana data.Publication of this article was funded in part by the University of Florida Open Access Publishing Fund.Author ContributionsJ.B.P. and M.C. collected and managed data. All authors conceived and designed the study, analyzed the data and wrote the paper.††‹–‹‘ƒŽfˆ‘”ƒ–‹‘Supplementary information accompanies this paper at https://doi.org/10.1038/s41598-018-24262-w Competing Interests : e authors declare no competing interests. Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional aliations. Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made. e images or other third party material in this article are included in the article’s Creative Commons license, unless indicated otherwise in a credit line to the material. If material is not included in the article’s Creative Commons license and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this license, visit http://creativecommons.org/licenses/by/4.0/. e Author(s) 2018