Long-Term Variation in Survival of A Neotropical Freshwater Turtle: Habitat and Climatic Influences

Few long-term demographic studies have been conducted on freshwater turtles of South America, despite the need for this type of inquiry to investigate natural variation and strengthen conservation efforts for these species. In this study, we examined the variation in demography of the Chocoan River Turtle (Rhinoclemmys nasuta) based on a population from an island locality in the Colombian Pacific region between 2005 and 2017. We calculated survival, recapture, and transition probabilities, and the effects of stream substrate and El Niño–Southern Oscillation (ENSO) phases (El Niño, Neutral, La Niña) on these variables using a multi-state model. We found differences in survival probabilities between ENSO phases, likely as a consequence of an increase in flood events. In addition, we found support for survival being greater in muddy streams than rocky streams, possibly because it is easier to escape or hide in mud substrates. Recapture probabilities varied by life stages; differences in the probability of recapture between size classes were associated with the high fidelity to territories by adults. The present increases in frequency and severity of El Niño and La Niña may exacerbate the consequences of climatic regimes on natural populations of turtles by increasing the mortality caused by drastic phenomena such as floods.


Introduction
In recent decades, many turtle species have shown declines in population size [1,2]. The most recent overview estimates that 54% of all turtle species are threatened, exceeding the threat levels of other vertebrate groups such as parrots, primates, or frogs, which are considered as examples sensitive to extinction [3]. There are a variety of causes that affect population viability among chelonians, including habitat loss and degradation, introduction of invasive species, environmental pollution, disease, and human exploitation [4,5]. These threats are intensified by factors inherent to the biology of turtles, such as the slow recovery of populations after declines and delayed sexual maturity [6,7].
Extreme climatic variation and climate warming have been identified as other factors that potentially affect turtle populations [8][9][10]. With respect to temperature, warming may skew sex ratios in populations of the many turtle species that have temperature-dependent sex determination [11][12][13][14], as well as alter the phenology of nesting and nest emergence [9,10]. On the other hand, variation in precipitation can change net primary productivity (NPP), thereby affecting the food supply, body condition, and reproductive output in populations of different species [15], including turtles and tortoises [16,17]. Changes in temperature, precipitation, NPP, and sea level will also modify habitats and are projected to alter the geographic ranges of a significant number of turtle species under various The study site is within one of the wettest continental areas on the planet, coinciding with landfall of the Intertropical Convergence Zone (ITCZ) in the eastern Pacific [36,37]. Average climatic factors are equable with high amounts of precipitation, between 7200 to 8500 mm annually, high mean relative humidity of approximately 90%, and a mean annual temperature of 23.5 to 25.7 °C [38]. Despite this, the area is subject to the climatic phenomenon El Niño-Southern Oscillation (ENSO), which differentially affects weather conditions along the Colombian Pacific coast [39,40]. During the warm phase of ENSO (El Niño), positive anomalies in the sea surface temperatures (SSTs) of the Pacific Ocean lead to a deficit of precipitation affecting the central zone of the Colombian Pacific coast. On the other hand, during the cold phase of ENSO (La Niña), the SSTs have negative anomalies and an excess of precipitation is recorded in the central zone of the Colombian Pacific coast [40]. The third phase of ENSO (Neutral) is when SSTs are near average, as are precipitation amounts in the region. El Niño and La Niña represent environmental changes that are extreme in intensity and duration, so all flora and fauna associated with the coastal and continental areas are affected [41][42][43][44][45]. The study site is within one of the wettest continental areas on the planet, coinciding with landfall of the Intertropical Convergence Zone (ITCZ) in the eastern Pacific [36,37]. Average climatic factors are equable with high amounts of precipitation, between 7200 to 8500 mm annually, high mean relative humidity of approximately 90%, and a mean annual temperature of 23.5 to 25.7 • C [38]. Despite this, the area is subject to the climatic phenomenon El Niño-Southern Oscillation (ENSO), which differentially affects weather conditions along the Colombian Pacific coast [39,40]. During the warm phase of ENSO (El Niño), positive anomalies in the sea surface temperatures (SSTs) of the Pacific Ocean lead to a deficit of precipitation affecting the central zone of the Colombian Pacific coast. On the other hand, during the cold phase of ENSO (La Niña), the SSTs have negative anomalies and an excess of precipitation is recorded in the central zone of the Colombian Pacific coast [40]. The third phase of ENSO (Neutral) is when SSTs are near average, as are precipitation amounts in the region. El Niño and La Niña represent environmental changes that are extreme in intensity and duration, so all flora and fauna associated with the coastal and continental areas are affected [41][42][43][44][45].

Field Data Collection
Sampling was conducted annually between June 2005 and March 2017, except for the years 2008, 2013, and 2014. Sampling was performed in four streams of first and second order (widths less than 4 m and depths less than 1.5 m), with variable lengths between 100 and 300 m; the total area for all streams was 0.36 ha. Specimens were collected between 20:00 and 00:00 by direct visual encounter in the clear water and manual investigation of crevices, overhangs, and leaf litter; the animals collected were marked at the site of capture using a triangular file. We used Cagle's [46] scute notching method to individually mark turtles, with some species-specific modifications [34]. For each individual, the carapace length was measured with a Vernier caliper as an indicator of the life stage (age) category to which it corresponds.

Data Handling and Analyses
Multi-state mark-recapture models [47] were implemented using the program MARK version 8.1 [48] to estimate annual survival, recapture (individual detectability), and transition (recruitment) probabilities for turtles in the three different life stages: hatchlings (soft plastron, carapace length < 100 mm); juveniles (carapace length > 100 mm and < 140 mm); and adults (carapace length > 140 mm) [31,49]. The capture history of each turtle is included in the Supplementary file. Model selection was based on the information theoretic approach (Akaike information criterion), with the most supported models having the lowest values [50]. We treated type of stream (rocky vs. muddy) as a categorical factor to examine the relationship of survival and transition probabilities with these types of habitats. Another factor included in the analysis was the ENSO phase-we checked the phase when the capture event occurred and treated El Niño, La Niña, and Neutral phase as a factor to test whether this climate phenomenon influenced survival or transition probability. We first evaluated a global model that included interactions among all four factors (year, ENSO, life stage, stream) for survival, recapture, and transition probabilities; however, due to limitations of the data, the program only used 69 parameters in the calculation of results and not the full 145 parameters. For this reason, we substituted the next most-inclusive model (142 parameters) as the global model with interactions among three of the factors (model 1) and a dot model where all three factors were constant (model 2). Then, maintaining a global model structure for survival and transition, we evaluated support for all single factors (year, life stage, stream, ENSO; models 3-6) and maintain recapture probability as constant (model 7). Subsequently, we selected the most supported model structure for recapture probabilities and repeated this process for transition and survival (models [8][9][10][11][12][13][14][15][16][17][18]. In cases where the differences between models were ∆AIC c < 2.0, we tested additive and multiplicative effects between variables.

Results
During the 10 visits to Isla Palma, we captured 650 different individuals of R. nasuta (hatchlings: 170; juveniles: 142; adults: 338) 850 times. Of the set of 18 candidate models used to assess the influence of year, life stage, stream type (muddy vs. rocky), and ENSO phase on survival, recapture, and transition probabilities, the most supported models included ENSO affecting survival, and life stage determining recapture and constant transition probabilities (w = 0.47, Table 1). According to this model, annual survival probabilities were lower for La Niña phase (µ = 0.64, ± CI 95% = 0.13) and Neutral phase (µ = 0.70, ± CI 95% = 0.10) than El Niño phase (µ = 0.80, ± CI 95% = 0.11; Figure 2a). Recapture probabilities were higher in adults (µ = 0.073, ± CI 95% = 0.023), followed by juveniles (µ = 0.018, ± CI 95% = 0.008), and, finally, hatchlings (µ = 0.005, ± CI 95% = 0.003; Figure 2b). Transition probabilities were constant between the different stages with wide variation (µ = 0.449, ± CI 95% = 0.227). Only one other model was considered competing (∆AIC c = 1.2); in this model (15), stream type affected survival, which was lower in rocky streams (µ = 0.686, ± CI 95% = 0.09) than in muddy streams (µ = 0.799, ± CI 95% = 0.122; Figure 3). There were no additive or multiplicative effects identified among ENSO phases or stream types. In addition, we tested whether or not the differences between survival probabilities were significant in the two best models (16 and 15) using a test for confidence intervals. We found that in the first model (16), the survival probability was significantly higher in El Niño years than in La Niña years (T 1540 = 1.950; p = 0.049); in the other two comparisons for this model we did not find significant differences (El Niño vs. Neutral; La Niña vs. Neutral; p > 0.05). For the second model (15), we did not detect a significant difference in the survival probabilities between streams with different substrates at p < 0.05; rather, the value lies between 0.05 and 0.10 (T 1540 = 1.652; p = 0.099). Table 1. Rankings of multi-state models estimating survival (Φ), recapture (p), and transition (Ψ) probabilities for the Chocoan River Turtle on Isla Palma. AIC c = Akaike information criterion corrected for small sample sizes; K = number of parameters; w = AIC c weight; El Niño-Southern Oscillation (ENSO) = phases (La Niña, Neutral, El Niño); stages = hatchlings, juveniles, and adults; type of stream = muddy, rocky. Models are ranked based on AIC c . The order in which the models were computed is shown in parentheses.

Discussion
Our results suggest that survival in this insular population of R. nasuta is more affected by the ENSO oscillations than by the type of stream this species inhabits or the variation among the different size classes. The probability of recapture, as predicted, varies among size classes, which is possibly associated with the fidelity to territory exhibited by adults [31]. The probability of transition is constant; surprisingly, the differential growth rate among size classes [31] does not seem to result in variation of this probability. It is possible that the constant transition result is affected by the few individuals that made transitions between size classes.
Previous studies have shown that climatic conditions affect the viability of turtle populations by affecting sex ratio and nesting success [51]. On the other hand, a study in the Mediterranean tortoise Testudo hermanni showed that this species had greater survival when precipitation increased due to the positive effects on plant growth and forage for the young tortoises [16]. In our tropical species, we found the opposite pattern because the mechanism by which the rain affects the populations is different and has nothing to do with diet. We propose that climatic conditions directly affect the survival of R. nasuta when catastrophic events such as floods occur, which generate the expulsion of some individuals from the streams and change the habitat structure and water dynamics of the streams; this type of effect has been related to variation in population parameters in some turtle populations [25,52,53]. Anecdotally, we have observed this ejection of individuals during sampling when some individuals were found on the beaches or in the intertidal area after strong storms, and we have seen smaller individuals being washed downstream during torrential rainstorms.
In addition to climatic factors, it has been amply demonstrated that freshwater turtles exhibit a preference for particular habitats [54][55][56] and that variation in survival probability can be produced by differences in predation levels [57,58]. The differences found at the level of survival in this population in streams with different substrates may be associated with the fact that the muddy substrates afford a reduction in predation risk because it is easier to escape or hide in mud, a pattern that has been reported in other species of freshwater turtles [59,60]. It may also be the case that the muddy substrates afford some protection from the scouring action of the torrential rains during the wetter ENSO phases-we have observed juveniles and adults that dug their own body pits in mud along the margins of the streams.
In general, the capture rate of turtle hatchlings in demographic studies is very low because typical turtle capture methods, such as trapping, are not effective with small individuals [61,62]. However, in our Isla Palma study streams, the method of capture is manual and the R. nasuta population exhibits a high percentage of hatchlings [63]-they are not missing. Differences in the recapture probability between size classes may be associated with the high fidelity to relatively small territories by adults, in particular, the females [31,34]. Another possibility is that, merely based on size, there are more small spaces to hide and more small, wet depressions where hatchlings and juveniles can evade detection [64]. Due to the lack of a pattern in variation of the transition probabilities, it is difficult to draw conclusions. The low proportion of individuals that made a transition (6.3%) may be the cause of large variation and the lack of a pattern. This low transition rate is probably associated with the low annual growth rate in this species, with growth rates in juveniles of about 5 mm per year [31]. This means that for an individual to move from hatchling to juvenile it would need, on average, 7.5 years. However, it is possible that the low recapture probability during hatchling or juvenile stages did not allow for the frequent detection of transitions.
The IPCC has identified basic aspects of climate change that are expected to impact biodiversity, such as the global warming trend, extremes of temperature, drying trends, and precipitation, as well as two geographic areas of particular concern, namely, terrestrial ecosystems in the tropics and northern latitudes [65]. As has been suggested, it may not be changes in average climatic conditions that may be determinative of a species geographic range and, presumably, its likelihood of extinction, but, rather, the extreme climatic events to which it is exposed [66]. Climate projections for the probability of extreme weather events, like floods and droughts, are greater than the projections of change in average conditions [67]. Although a flooding event has previously been suggested to influence reproduction in turtles by impacting survival [68], there appears to be no previous suggestion of an association in turtles between a documented climatic fluctuation, such as ENSO and survival, something that has been predicted for fishes [69].
Current research shows that in turtle populations where human disturbance and the rate of depredation are low, they are nevertheless affected by climatic changes. The increase in the severity of El Niño and La Niña may exacerbate the consequences of climatic regimes on natural populations of turtles in the tropics as this study suggests. The proximate cause of the increased mortality appears to be extreme climatic phenomena, in this case, floods associated with extreme rainfall during the wettest phases of ENSO.