Florida longitudinal datasets
The AAT remained relatively constant from 2012–2018, but the NSAT was more variable (Fig. 1). The years 2015, 2016 and 2017 were particularly dry and hot years (Fig. 1). The lowest accumulated precipitation occurred in 2015. The year 2016 was both very dry and very hot, and it is estimated that more than half of the C. caretta eggs died. The year 2017 was more typical in April, but it became very hot and dry during the rest of the nesting season until mid-August when typical conditions returned to the rookeries. During the period 2015–2017, the driest years, the variance in sizes was greatest for both species. The hottest nesting season for C. caretta occurred between 2014 and 2017, but the nesting season for C. mydas was, on average, one degree hotter in the same period (Fig. 2). In Florida, C. mydas starts nesting between the months of June and September, and C. caretta starts its nesting season between April and September; consequently, the weather each species experiences, as a whole, differs. The nesting seasons differed for the two species in terms of precipitation as well. Although the driest year was 2012, which was one of the most humid seasons for C. caretta, the NSP was drier for C. mydas than for C. caretta (Fig. 1). The year 2013 was the most humid year for both nesting seasons, and on average, the nesting season of C. caretta experienced less rainfall than that of C. mydas.
In the case of C. caretta, the precipitation during the nesting season (NSP) correlated with the mean hatchling size (SCC) (p = 0.0026, Table 7, Fig. 3A), SCW (p = 0.0055, Table 7, Fig. 3A) and mass (p = 0.0196, Table 7 = Fig. 3A). Still, it was not correlated with the coefficient of variation of the hatchling size metrics (Fig. 3B). The average temperature during the nesting season (NSAT) was also not correlated with any of the hatchling size metrics, i.e., averages and variation (Fig. 3). Although temperature and precipitation are correlated, at the regional level, precipitation is also a consequence of hydric, topographic, and atmospheric factors; the NSAT has a nonsignificant negative correlation (p = 0.5289) with NSP. The AAT is strongly positively correlated with the SCL (p = 0.0131, Table 7, Fig. 3A) and mass (p = 0.0102, Table 7, Fig. 3A). NSP was positively correlated with the average SCL (p = 0.0226, Table 7, Fig. 3A), SCW (p = 0.0055, Table 7, Fig. 3A) and mass (p = 0.0196, Table 7, Fig. 3A) but not with the coefficient of variation (SCL, p = 0.9747, SCW, p = 0.84656, mass, p = 0.2689, Supplementary material), suggesting that precipitation during the nesting season has a significant effect on the hatchling size of C. caretta (Fig. 3). In the case of C. mydas, geoclimatic variables are not as correlated with hatchling size as they are in C. caretta. The AAT was positively correlated with average mass (p = 0.04571, Table 8, 5, Fig. 4A) but not with the coefficient of variation of either of the hatchling size metrics (SCL, p = 0.3836; SCW, p = 0.73455; mass, p = 0.67518; Supplementary material and Fig. 4B). NSP was positively correlated with the average SCW (p = 0.00026, Table 8, Fig. 4A) and with the coefficient of variation of SCW (p = 0.00176, Supplementary material, Fig. 4B). The air temperature during the nesting season (NSAT) was negatively correlated with the coefficient of variation of the SCW (p = 0.02689, Supplementary material, Fig. 4B).
Meta-analysis of worldwide populations
For this analysis, we considered the effects of latitude and longitude on the populations, and the coordinates were standardised by transforming them into Cartesian pairs. The x-axis [\(\text{sin}(lat)\times \text{cos}(lon)\)] runs along points (0,0), the y-axis [\(\text{cos}(lat)\times \text{sin}(lon)\)] runs along points (0,90), and the z-axis [\(\text{sin}(lat)\)] runs through poles (-90,0) and (90,0). Thus, the x- and y-axes represent the positions along the surface of the Earth when viewed from an azimuthal projection, whereas the z-axis represents a location on the sphere (poles, tropics or equator). On the x-y plane, the first quadrant represents the populations sampled from the Mediterranean in the northern hemisphere, namely the populations in Greece, Cyprus, Turkey, as well as the Aden Gulf population in Abul Wadi, and the western Indian Ocean in the southern hemisphere, namely the populations in Tromelin Island, Astove, Aldabra and Europa Island off the east coast of Africa. The second quadrant represents populations from the western Pacific Ocean, with the populations of the South China and East China Seas in the northern hemisphere, and the populations from the Coral Sea in the southern hemisphere. The third quadrant represents the eastern Pacific, with only one population represented, the one in the French Frigate Shoals. The fourth quadrant represents the western Pacific coasts in South America and the northern portion of the Atlantic Ocean, which includes the Caribbean Sea and the Gulf of Mexico and the populations of Cape Verde in the northern hemisphere, and the population from the Ascension Island in the southern hemisphere (Fig. 2).
In an initial inspection of the dataset for C. caretta, several variables displayed a significant correlation (Table 5). The z-axis, a proxy for latitude, is positively correlated with the triplet SCL-SCW-mass. The AAT is correlated with the SCL and mass, whereas the precipitation is positively correlated with the SCL, SCW and mass. A PCA was performed on a covariance matrix in PAST 4.15 [62].
The first two eigenvalues explained 76% of the variance, with principal component 1 (PC1) containing mostly morphometric measurements (SCL-SCW-mass) and PC2 comprising mainly AAT and NSAT (Fig. 5A). Precipitation, SCL, SCW and mass are positively correlated. Precipitation and NSAT are likely not correlated. The latitude (z-axis) is more strongly positively correlated with the mass than with the SCL and SCW (Fig. 5).
When plotting PC1 against PC3, which has an eigenvalue of 0.84656, the same geography-based clusters were obtained (Fig. 5B). PC3 is mostly composed of NSP. The biplot shows that the precipitation at the nesting site is positively correlated with the annual temperature and the distance along the poles (z-axis). In contrast, the NSAT is closely correlated with size (SCL-SCW-mass) (Fig. 5B). The Mediterranean population was somewhat separated from the other populations, probably due to the very dry regime in the region (Fig. 5B). The initial inspection of the dataset revealed that latitude was positively correlated with SCL (p = 1 × 10–5, Table 7), SCW (p = 0.0002, Table 7) and mass (p = 4 × 10–6, Table 7). AAT could also be correlated to the latitude itself, as the temperature regimes change along the coasts where the populations nest. The significant correlation could be due to its not being decomposed into three vectors, as was the case with geographic coordinates. Thus, the correlation of AAT to hatchling size metrics may be an artefact of the turtle populations having different sizes along long coastlines.
In an initial exploration of the C. mydas dataset (Table 7), the z-axis was positively correlated with SCL (p = 0.04013, Table 8), whereas the AAT was positively correlated with mass (p = 0.0457, Table 8), and NSP was strongly positively correlated with SCW (p = 0.00026, Table 8). The first three components had eigenvalues greater than 1.0 and accounted for 74.96% of the variance, with PC1 mainly composed of hatchling size metrics (SCW-SCL-mass) (Fig. 6), PC2 of AAT and NSAT, and PC3 of NSP. When plotting PC1 vs. PC2 (Fig. 6A), SCL was strongly positively correlated with NSP. SCL and SCW are more strongly correlated with each other than with mass. The AAT and NSAT are likely not correlated with SCW or SCL but are slightly negatively correlated with mass. The z-axis is positively and strongly correlated with mass but less strongly correlated with SCL and SCW. Unlike in C. caretta, the clusters are not as clear. The Mediterranean populations are clustered together but are distinctly isolated from the rest of the populations. The Coral Sea populations and Indo-Pacific seas and Indian Ocean populations form overlapping clusters, whereas the Caribbean and Gulf of Mexico populations partially overlap with the former two. The offshore South Atlantic populations are nested within the main overlapping clusters, whereas the mid-tropical North Pacific Ocean is detached from all the clusters. Interestingly, the eggs were collected from the Caribbean and Gulf of Mexico populations in Tortuguero, Costa Rica, in 1980; the eggs were transported to the US and hatched under experimental conditions, and their placement too far from the cluster that corresponds to the Caribbean Sea and the Gulf of Mexico may reflect their growth without the precipitation regime in situ. When plotting PC1 vs. PC3 (Fig. 6B), the geographic clustering became clearer, with a Mediterranean cluster becoming increasingly detached. The z-axis is positively correlated with mass and SCL, but it is likely not related to SCW. The monthly average temperature during the nesting season was negatively correlated with mass and SCL, but the annual air temperature and precipitation were positively correlated with SCW.
In terms of coefficients of variation, the correlation matrix shows that the coefficient of variation of mass within the population is correlated with its distribution in the Eastern and Western Hemispheres (x-axis [p = 0.089] and y-axis [p = 0.054] negatively and positively correlated, respectively, albeit not significantly). The PCA from the C. caretta dataset produced two principal components that accounted for 64.25% of the variation (Fig. 7A), whereas the PCA from the C. mydas dataset produced two principal components that accounted for 54.85% of the variation (Fig. 7B).
In the case of C. caretta, PC1 contains mainly morphometric variables, whereas PC2 contains mainly geoclimatic variables. The coefficients of variation of mass, SCL and SCW are negatively correlated with the geoclimatic variables (Fig. 7A). The variation in mass is positively correlated with the variation in SCL and SCW. According to a global meta-analysis of C. caretta, the populations from the Mediterranean (Turkey [31, 38, 40, 43, 44], Northern Cyprus [41], Cyprus [38] and Greece [39]) are in a cluster opposite to the populations from the Atlantic coast of the USA [3, 42]. Although size (SCL-SCW-mass) appears to be related to temperature, i.e., the warmest temperatures occur closer to today, this is likely a geographic artefact where the Mediterranean populations are smaller than the American ones. The Mediterranean populations were sampled between 1978 and 2006, whereas the populations on the American Atlantic coast were sampled between 2002 and 2018 (Fig. 7A). This geographic pattern is clearly observed in the three samples taken from 2014, with the samples from the USA clustering togeher and those from the Japanese beaches straddling in the middle of the plot (Fig. 7A).
For C. mydas, the PCA of the geoclimatic variables against the coefficient of variation of hatchling size metrics does not show neatly separated groupings when considering the biogeographic realm [63]. The populations from the Indo-Pacific seas and Indian Ocean [2, 51,52,53], the Coral Sea, the Caribbean and Gulf of Mexico, and the offshore South Atlantic overlap with each other and have a similar spread over the PCA biplot when plotting PC1 vs. PC2 (Fig. 7B). The populations from the Mediterranean were the only ones separated from the rest, suggesting different variation trends when compared to the rest (Fig. 7B).
The influence of precipitation on the PCAs is greater for PC4, which has an eigenvalue less than 1.0. However, it explains the distribution of points along the PCA plot (Figs. 6 and 7). For instance, the two samples from 1973 were collected from two different islands in the Indian Ocean, namely, Europa [52] and Tromelin Island [53]. Europa Island is to the southwest of the coast of Madagascar. It receives less rainfall during the nesting season than does Tromelin Island to the northwest of the coast of Madagascar, which receives nearly three times as much monthly average cumulative rainfall. The European Plateau is to the left, where populations from Turkey [56] Northern Cyprus [41, 43], Yemen [49] and Australia [6, 55] are located, as well as the outlier removed from Tortuguero, Costa Rica, which, under laboratory conditions, was not subjected to any rainfall regime [54]. The in-situ study from Tortuguero, Costa Rica, clustered with the other Atlantic populations to the right, which were exposed to more rainfall [50]. Overall, this clustering shows that the geographic distribution plays a larger role in explaining the variance, i.e., reflecting distinct turtle populations with their own growth variability and their distinct response to the geoclimatic variables than the geoclimatic variables themselves on the species.
Cabo Verde study on C. caretta hatchlings
The eggs were laid between mid-July and mid-August, and they hatched between early September and mid-October 2020. August and September are often the hottest months, while July and August are the driest months. PCA (Fig. 8) revealed that three main components had eigenvalues greater than 1.0 and accounted for between 58.5% and 79.3% of the variance. According to the longitudinal data, the hatchling size and the triplet SCL-SCW mass are strongly positively correlated. According to the PCA plot, Porto Ervatão is more diverse in terms of hatchling size, whereas Ponta Benguinho has more similarly sized hatchlings. Most of the hatchlings from Ponta Cosme clustered near to each other.
When plotting PC1 (27.9% of the variance) vs. PC2 (16.2%), mass, SCL and SCW showed a weak positive correlation against geoclimatic variables. Of the two precipitation values, the precipitation two days before the mid-incubation period, P(b) in Fig. 8, was the driest, with a mean value of 0.27 mm, and showed a strong positive correlation with the hatchling size measurements, albeit weak. The other geoclimatic variables, precipitation in the middle of the incubation period, P(a) in Fig. 8, and temperature in the middle of the incubation period and two days before the mid-incubation period, T(a) and T(b) in Fig. 8, are likely not correlated with hatchling measurements. A stronger correlation between geoclimatic variables, namely, T(b) and P(b), is observed when plotting PC1 vs PC3 (13.4% of the variance). Furthermore, clutch size showed a strong positive correlation with P(b) in this biplot. Finally, PC4 (11.4% of the variance) is mostly composed of the estimated percentage of females in the nest, which suggests a weak correlation with any of the other variables. It is thus likely that the estimated number of females in the nests differed from the real number.
In C. caretta, hatchling size showed a strong negative correlation with precipitation (NSP) [corr(NSP,SCL) = -0.78, p = 0.0397; corr(NSP,SCW) = -0.91, p = 0.0047; corr(NSP,mass) = -0.86, p = 0.014] during the nesting season, whereas weight was strongly positively correlated with the average monthly air temperature during the nesting season (NSAT) [corr(NSAT,mass) = 0.84, p = 0.017]. The annual air temperature (AAT) is also strongly positively correlated with hatchling size, but this may be a confounding variable, as the average annual air temperature overall reflects the hydric and thermic conditions throughout the year [corr(AAT,SCL) = 0.85, p = 0.0142, corr(AAT,SCW) = 0.88, p = 0.0084, corr(AAT,mass) = 0.81, p = 0.0261]. In C. mydas, the environmental variables did not significantly correlate with hatchling size (Supplementary material). Other environmental factors may need to be investigated for C. mydas, such as the response of vegetation to drier and warmer conditions [64,65,66].
Add Comment