To ensure good resolution within the δ13C, δ15N, and δ34S isoscapes of the sampled green turtle diet sources, twelve geographically separate sites were chosen along roughly 1,100 km of the NSW coastline (Fig. 1). Green turtle tissue samples were collected from multiple management and conservation organizations responsible for cataloguing stranded green turtles along the NSW coastline. A total 36 green turtle tissue samples were provided, from turtles that stranded between 2018 and 2021, spanning 18 stranding locations, which were assigned to four major stranding regions: north coast (n = 8), mid-north coast (n = 15), central coast (n = 11), and south coast (n = 2).
Green turtle diet source sampling
Green turtle diet sources (seagrass, algae, and mangrove) were collected from multiple locations along the NSW coastline (Fig. 1), with three to five coastal or estuary sites (5 km2 radius) within each location selected for sampling, based on accessibility and depth (< 3 m). Habitat selection included intertidal and subtidal seagrass meadows, coastal rocky headlands, mangrove stands, and artificial formations such as sea-walls and boat ramps, representing typical foraging habitats for green turtles (Limpus 2009, Bjorndal et al. 1997). Diet source selection was confined to taxa known to be consumed by green turtles within Australian waters (Prior et al. 2015; Coffee 2020), with known distributions in NSW. Seagrass and algae were collected during Autumn months (2021), and commonly ingested mangrove species were sampled during Winter months, 2021 (Fig. 1) (Amorocho & Reina 2007; Arthur et al. 2008; Prior et al. 2015; Coffee 2020). Mangrove sampling was restricted to the northern NSW coastline because local COVID-19 restrictions prevented further sampling. All algae groups (Rhodophyta, Chlorophyta, Phaeophyta) were combined, as there was no significant difference between isotopic values for δ13C (one-way ANOVA, δ13C: Df = 2, F = 2.954, P = 0.06, δ15N: Df = 2, F = 0.008, P = 0.99, and δ34S: Df = 2, F = 8.351, P = 0.63). The omission of animal diet-items from this study was a result of COVID-19 restrictions restricting repeat access to sampling locations.
Turtle tissue sampling
Epidermis samples were taken from posterior and anterior regions of the ventral surface, usually around the hind or front flippers (Reich et al. 2008), using a sterile biopsy punch with a diameter of 4 mm, or using a scalpel, and following methods described in (Arthur et al. 2008). The straight carapace length (CCL) of each turtle was measured between the nuchal notch and posterior tip of the carapace (Limpus et al. 1994). All sampled turtles where considered sub adults (SCL 37–68 cm), as defined by previous literature (Summers et al. 2017). Due to the opportunistic sampling associated with the stranded green turtles used in this study, it is possible that isotope values may be affected by rates of decomposition during the days prior to sampling. In temperatures exceeding 20°C, bacterial decomposition is known to enrich both δ13C and δ15N values after 3–4 days, with significant enrichment after 14 days (Burrows et al. 2014; Keenan & DeBruyn 2019). However, the storage of these samples in ethanol (> 70%) likely inhibited the enrichment of isotope values resulting from decomposition, as ethanol is known to greatly hinder bacterial growth (Ingram 1989; Dyrda et al. 2019).
Stable isotope sample preparation
Green turtle diet and epidermis samples were lightly scrubbed and rinsed with distilled water and cleared of attached debris and epiphytes. Diet samples were stored frozen (-18°C) for preservation post-sampling (Barrow et al. 2008; Burrows et al. 2014). Epidermis samples were either frozen, or stored in 70% ethanol, then frozen, in line with methodology set out by Barrow et al. (2008). The effect of lipid extraction via hydrochloric acid treatment (1 N HCL) on the isotope values of epidermis values has been examined in previous literature (Post et al. 2007; Vander Zanden et al. 2012; Bergamo et al. 2016; Turner Tomaszewicz et al. 2017), and deemed unnecessary here, as acid treatment is also known to alter δ34S values in plants, algae and animals (Connolly & Schlacher 2013). Therefore, isotope analysis was performed on non-acid-treated epidermis samples during this study. Diet samples were oven-dried for 24–48 hours and ground to a fine consistency using a mortar and pestle (Arthur et al. 2008; Reich et al. 2008; Arthur et al. 2009). Standardized sample sizes (12–15 mg for plant and algae items; 6–10 mg for epidermis) were then placed into tin capsules, pelletized, and placed into an elemental analyzer for combustion (per Fry 2019). Samples were analyzed for carbon, nitrogen and sulfur stable isotopes in an IsoPrime isotope ratio mass spectrometer (with N2 and CO2 gases chromatographically separated) at the Stable Isotope Laboratory at Griffith University, Nathan Campus, Queensland (QLD), Australia. Elemental analysis provided δ13C, δ15N, and δ34S isotope values as a relative deviation (‰) from conventional standards, and expressed in delta (δ) notation in parts per thousand:
δ = ([Rsample / Rstandard] -1) x 1000 (1)
Where Rsample is the ratio of heavy to light isotopes in the sample, and Rstandard is the isotope ratio of the corresponding international standard, Pee-Dee Belemnite for carbon, atmospheric air for nitrogen, and Canyon Diablo Troilite for sulphur. Laboratory calibrated glycine and spinach (for diet sources), and bovine liver (for epidermis samples) were used during each run for control purposes. Standard deviation of the mean glycine, spinach and bovine liver samples confirmed the precision of analysis for diet and epidermis samples, respectively.
Generating isoscapes
Spatial analyses for generating elemental isoscapes for green turtle diet sources were run in R version 4.1.1 (R Core Team, 2017). Three sets of elemental isoscapes for δ13C, δ15N, and δ34S were generated for three major green turtle diet sources: seagrass (n = 90), algae (n = 63), and mangrove (n = 21). Isoscapes were generated using spatial and mapping packages in R, with data from each sampling location, including specific geocoordinate locations associated with sampled foraging areas. A 20 km wide ‘buffer zone’ raster layer was constructed to map coastal geographic isotopic variations to enhance visualization of isotopic trends at a large scale, and with spatial resolution set to 500 x 500-meter pixels to account for sufficient levels of detail given the amount of, and distance between, sample data values.
Each isoscape was modelled using a gaussian generalized additive mixed model (GAMM), with an identity link function (Wood 2017), fit by restricted maximum likelihood using the R package ‘mgcv’ (Wood 2003). These models included a one-dimensional gaussian process (kriging) for Northing, and a random intercept for Species that accounted for isotopic differences among sampled diet groups (e.g., Zostera muelleri for ‘Seagrass’). Site-fold cross-validation (CV) was used to compare each model (excluding those for mangroves, which had insufficient data) to two other candidate models: a model with an additional random intercept for Site, and model that used a thin plate spline for Northing instead of the gaussian process. This investigation revealed that our model outperformed the other two models at predicting interpolated data in four out of six cases, based on the root-mean-square error. Gaussian processes that allowed for different levels of complexity were fit to ensure that the models adequately captured spatial relationships without overfitting the data. For each of the final models, the assumptions of normality and heteroscedasticity were investigated visually using Q-Q plots and standardized residual plots, respectively, which revealed no major concerns with the specification of these models. For each isotope value within each of the diet groups, there was a strong linear correspondence between values predicted by the models, and observed values from the sampling dataset, indicating the models fit the data relatively well.
Following model selection and validation, the models were used to spatially interpolate areas between sampling locations, which allowed for the construction of isoscape maps along the entire NSW coastline. Further, GAMM models were restricted to predictions within the isotopic range of sampled diet sources, leading to only realistic values being predicted in the isoscapes. The relationships for isotope values between diet sources, as well as diet source isotope values between sampling locations, was tested statistically using a one-way ANOVA (α: 0.05) with location and diet source as factors.
Stable isotope mixing models
The relative contribution of seagrass, macroalgae, and mangrove was assessed for sampled turtles within and between all four stranding regions, with the availability of each diet source being assumed as continuously available to all study turtles. All statistical analyses were conducted using R software version 4.1.1 (R Core Team 2017). Isotope discrimination factors for green turtle epidermis (0.17 \(\pm\) 0.03 o for δ13C, 2.80 \(\pm\) 0.11 % for δ15N) were sourced from Seminoff et al. (2006), and were applied to account for turtle-diet isotopic discrimination (Fry 2006). Stable isotope mixing models for diet source contributions were run in SIMMR (stable isotope analysis in R; Parnell & Inger 2021), using the Bayesian Framework. The SIMMR model utilizes iterations of Markov Chain Monte Carlo algorithms for probability distributions (Moore & Semmens 2008). The model produced probability density distributions for potential proportions of diet sources (Parnell et al. 2010). Normal distributions of the source data were confirmed, and the mean values for diet sources were examined for normality and homogeneity of variance, and were incorporated into the model to estimate the most probable contribution of each food source to the overall diet of each turtle. A one-way ANOVA (α: 0.05) was used to test the relationship between isotope values of sampled turtles between locations, and the relationship between the isotope values in epidermis samples and the most prominent diet sources.