Population Viability Analysis of a Long-Lived Freshwater Turtle, Hydromedusa maximiliani (Testudines: Chelidae)
Abstract
Maximilian's snake-necked turtle, Hydromedusa maximiliani, is an endemic and vulnerable long-lived freshwater turtle from eastern Atlantic mountainous rainforest regions of Brazil. Three scenarios (real population, optimistic population, and pessimistic population) based on the species' life-history data and carrying capacity estimates collected throughout 13 yrs were generated with VORTEX and subjected to sensitivity tests to verify the influence of demographic (mortality rate and inbreeding depression) and environmental (fire and deforestation catastrophes) parameters on population viability. The results showed significant differences among the viability scenarios, with extinction probability ranging from 6% to 99%. These analyses illustrate that environmental and demographic stochasticity can negatively impact populations of H. maximiliani and emphasize the necessity of protecting areas for long-lived organisms inhabiting regions impacted by humans.
Biodiversity monitoring programs are useful for priority actions, including species management and vulnerability assessment (Fagan and Holmes 2006). Population viability analyses (PVA) are important tools for animal population research and have been used to estimate extinction probabilities when species are faced with potential population fluctuations in response to environmental, demographic, and genetic stochasticity over a given time (Lindenmayer et al. 1993; Akçakaya and Sjögren-Gulve 2000; Lacy et al. 2009). Since PVAs integrate complex and often uncertain information, sensitivity analyses must be applied to explore the influence of parameters and assumptions that can result in a more precise model (Akçakaya and Sjögren-Gulve 2000; Naujokaitis-Lewis et al. 2009). Sensitivity analyses can identify which variables or assumptions are important or those with sensible scenarios and enhance transparency in model predictions, helping prioritize costly research for guiding the recovery of species at risk (McCarthy et al. 1995; Mills and Lindberg 2002).
Recent improvements in PVA analysis have been developed and new software packages have been added for higher precision of simulations and more accurate modeling of real-world scenarios (Lacy 1993; Lindenmayer et al. 1995; Beissinger 2002). An improvement of this technique, conducted on a quantitative framework, is the possibility of threat detection before the presumed threats are confirmed, with the benefit of more efficient management programs (Akçakaya and Sjögren-Gulve 2000; Brito and Fernandez 2000a; Lacy et al. 2009). These advantages have been addressed in several studies, from plants to animals, including turtles (Doak et al. 1994; Rivera and Fernández 2004; Enneson and Litzgus 2009; Snover and Heppell 2009; Bulté et al. 2010).
The Brazilian Atlantic rain forest is one of the most threatened ecosystems in the world, a hot spot deserving integrated studies encompassing biodiversity and habitat management planning (Brooks et al. 2002). Brazil is ranked among the top 5 most turtle-rich countries in the world (van Dijk et al. 2011). Although preliminary records on population ecology for some Brazilian Chelidae have accumulated in recent years (Brito et al. 2009; Bujes and Verrastro 2009; Famelli et al. 2011; Fraxe Neto et al. 2011), there are few long-term studies, a mandatory step for simulation models in population viability analyses. Because they are long-lived organisms, determination of turtle life-history traits as well as demographical approaches must rely on long-term studies (Gibbons 1997; Martins and Souza 2008, 2009).
The Maximilian's snake-necked turtle, Hydromedusa maximiliani, is an endemic freshwater turtle from the eastern part of the Atlantic rain forest mountain regions in Brazil and is classified as vulnerable according to International Union for Conservation of Nature (IUCN) threatened categories (Martins and Souza 2009; IUCN 2011). This species has become a model species in Neotropical freshwater turtle research, with intensive studies at a local population level ranging from natural history to molecular ecology over the past 2 decades (see references in Souza and Martins 2009). Data from these studies offer significant contributions to the development of a PVA for this population. We report the findings from the first PVA developed for a Neotropical freshwater turtle. Our results emphasize the necessity of conservation units for the survival of long-lived species.
METHODS
Study Area
The Parque Estadual Carlos Botelho (PECB) is an approximately 38,000-ha conservation unit located in southern São Paulo State in southeastern Brazil (lat 24°00′00″–24°15′00″S, long 47°45′00″–48°10′00″W). Together with other conservation units in its vicinity, it includes 120,000 ha of pristine Atlantic rain forest, representing one of the most representative remnants of this ecosystem in Brazil (Guix 2002). Geographically, the PECB exhibits a complex system of ridges and valleys with altitude ranging from 30 to 1000 m, defining a hydrological web with hundreds of temporary and perennial rivers and streams (Pfeifer et al. 1986; Domingues and Silva 1988). Mean annual precipitation is 1800 mm, and average temperature is 20°C. Despite the predominant native vegetation characteristics, some points (usually the borders) exhibit anthropogenic impacts, such as reforested fragments with exotic tree species (Eucalyptus spp. and Araucaria angustifolia; Camargo et al. 1972; Dias et al. 1995), clandestine palm-tree harvesting, and the transformation of soil use for pastures and banana plantation.
Simulation Extinction Processes with VORTEX
Hydromedusa maximiliani population viability analysis was performed with VORTEX 9.99 software (Lacy et al. 2009). VORTEX conducts population risk analyses based on Monte Carlo simulations, considering population models in probabilistic results (Lacy 1994; Lacy et al. 2009). Random processes generate fluctuations in population demography as well as environmental stochasticity, such as fire and deforestation events (Lacy 1994; Lacy et al. 2009). Thus, the more reliable the species' biological parameters are (gene flow, survival and fecundity rates, annual hatching, and sex ratio), the more accurate the resulting population viability scenarios are. The output data are summarized in statistical analyses of population growth rate, population extinction rate, and population extinction time, considering the time (years) defined by the user (Lacy 1994; Lacy et al. 2009).
VORTEX Scenarios
The scenario used as a model for population simulations in VORTEX was based on H. maximiliani natural history records from over 20 yrs in PECB (Yamashita 1990; Guix et al. 1992; Souza and Abe 1995, 1997a, 1997b; Martins and Souza 2008, 2009) and was referred to as the real population (RP) scenario. Two additional population scenarios were generated by adjusting carrying capacity (K) for an area of 350 ha: an optimistic population (OP; K = 1842) and a pessimistic population (PP; K = 101; Martins and Souza 2009; Table 1).
The mortality rates used in these simulations apply to adult turtles (Martins and Souza 2009). VORTEX requires information on the mortality rates for individuals belonging to age classes from youth to maturity. Since these data are unavailable for juveniles of H. maximiliani and other Neotropical species, juvenile mortality was based on literature records for Chrysemys picta (Frazer et al. 1990, 1991). Adult mortality rates were based on existing data measured for H. maximiliani (Martins and Souza 2009).
Catastrophic events such as fire (type 1) and deforestation (type 2) are considered the most important environmental threats in Atlantic rain forest (Fundação S.O.S. Mata Atlântica and Inpe 2002) and have been proven to influence small mammal population dynamics in several forest fragments southeastern Brazil (Brito and Fernandez 2000b; Brito and Grelle 2004; Brito and Fonseca 2006). Intense fires can make forest populations more susceptible to decline given the habitat change from a complex to a simplified structure, although biological community responses to this environmental factor are still controversial (Ashton and Knipps 2011; Arthur et al. 2012). The impact of catastrophes on population fluctuations was measured by changes in reproductive and survival rates.
The default value for inbreeding depression was used as a genetic parameter. Inbreeding depression was modeled in VORTEX as a loss of viability of inbred animals during their first year (Lacy et al. 2009). For each scenario, 1000 random iterations were performed, considering a 100-yr period. Populations were considered extinct when only 1 sex remained.
Data Analyses
The data generated in VORTEX simulations were 1) population growth rate (r; mean ± SD), the exponential growth rate or the mean generation time for males and females, and the stable age distribution used to initialize the starting population (see Lacy et al. 2009); 2) extinction probability (mean ± SE); 3) population size (mean ± SD); 4) extinction time (mean ± SD); and 5) decrease in genetic variability, expressed as expected heterozygosity (He) or gene diversity. VORTEX models loss of genetic variation in populations by simulating the transmission of alleles from parents to offspring and monitoring how many of the original alleles remain within the population as well as the average heterozygosity and gene diversity (He) relative to the starting levels (Lacy et al. 2009). The extinction probability is the proportion of simulations ending with extinction (Lacy et al. 2009).
Sensitivity tests were performed to test the influence of life-history and environmental parameters on predetermined scenarios (McCarthy et al. 1995). The scenarios were examined in relation to sensitivity variation in life-history parameters assumed to be important for determining the likelihood of a population's persistence. The parameters tested were mortality rate, frequency of catastrophes (together and individually), and inbreeding depression.
Baseline scenarios results from the RP, OP, and PP simulation scenarios were compared with analysis of variance using PAST Statistics 2.15 (Hammer et al. 2001). The significance of the difference in output between the basic assumptions scenario and changed models (sensitivity analysis) were tested using a Student 2-tailed t-test (Zar 1999) using Statistica 6.0 (StatSoft 2001).
RESULTS
Baseline Scenarios
Populations from the 3 scenarios responded differently to threats depending on the carrying capacity established for each one. We found a stronger difference among persistence scenarios (Table 2; Fig. 1; F2,297 = 86.32; p < 0.0001). There was a significant difference between all pairs of scenarios for all parameters compared, except that He did not differ between RP and PP (Table 2; Fig. 1).
Citation: Chelonian Conservation and Biology 11, 2; 10.2744/CCB-1001.1
VORTEX simulations showed a 31% mean extinction probability for the RP scenario, with a mean extinction time of 67 yrs. Considering a population survivorship throughout the 100-yr period, the final mean population size was 1090 individuals, with 11% mean annual population growth rate. The mean expected heterozygosity was 77%. For the OP scenario, the population went extinct in a mean period of 73 yrs with a mean extinction probability of 6%. Population growth rate was 12% annually. Population size could reach 1720 individuals with an expected heterozygosity around 82%. In the PP scenario, there was a 99.9% probability of extinction in a mean time of 31 yrs. Final population size was only 59 individuals, with an 11% mean annual growth rate (Table 2; Fig. 1).
Sensitivity Analysis
The probability of extinction for the RP scenario (31%) was significantly affected by changes in almost every parameter simulated; increased deforestation rate (type 2 catastrophe) was the only exception. Inbreeding depression resulted in an increased probability of extinction (33%), but there was an increase in expected heterozygosity for the real population. Only mortality and both catastrophes were significant influences on the mean time to extinction in the RP scenario. Frequency of both catastrophes also significantly influenced final population size (Table 3).
In the OP scenario, outcomes were similar to the RP scenario (Table 4). While the increase in mortality (81%) and in both catastrophes (81%) had a negative impact on original expected heterozygosity (82%), decreasing both catastrophic trends in the OP scenario significantly improved He by 87%. Only the increased frequency of both catastrophes had a negative and significant influence on the final population size in the OP scenario (from 1720 to 1687 individuals). The genetic parameters had significant influence when inbreeding depression was included in the He simulation, as mortality increased and the frequency of catastrophes was changed (Table 4).
Mean time to extinction in the PP scenario was 31 yrs and would be just 1 yr later (32 yrs) if the mortality rate were to decrease by 10% (Table 5). Final population size was not significantly affected when there was a decrease in deforestation rate and an increase in the frequency of both catastrophes. The expected heterozygosity did not show significant influence on any data parameters in the PP scenario (Table 5).
DISCUSSION
The PVA results suggest a probability of 30% for extinction within 100 yrs for the real population scenario, a high value according to IUCN criteria, which define a demographically unviable population when this value is below 10% (IUCN 2011). Considering this criterion, only in the optimistic scenario would the population remain viable. With changes in parameters used in sensitivity analysis, the population persistence probability remained approximately 30%. The mean extinction probability was 3 times higher (99%) for the pessimistic scenario. Populations inhabiting nonprotected areas are more susceptible to demographic stochasticity due to nest predation, more catastrophe recurrence, and competition with invasive species (Miller 2001; Rivera and Fernández 2004). In relation to environmental stochasticity, changes in catastrophe probability were associated with fluctuations in probability of population extinction. In the RP scenario, an effective decrease in population extinction probability would be achieved by reducing deforestation by more than 10%, suggesting that environmental stochasticity can act in H. maximiliani population demography while catastrophe monitoring should be emphasized as a management practice. The H. maximiliani population studied exhibits some demographical parameters and life-history traits that deserve attention in management programs, such as a female-biased sex ratio (Martins and Souza 2009) and a small clutch size (1–3 eggs; Yamashita 1990; Guix et al. 1992; Famelli 2009). The species lives in areas characterized by shallow rocky streams normally with small waterfalls, clear water, and dense vegetation along the banks and a predominance of broad-leafed plants and palm trees (Souza 2005; Souza and Martins 2009). In these habitats, resource partitioning is verified with adults frequently found in the main river course, hatchlings in shoreline areas with little flow and an accumulation of dead leaves, and juveniles also exploring shoreline areas (Souza and Abe 1998; Souza 2005). Action also must be taken in relation to aquatic habitat management. Deforestation can directly affect river and stream systems, reducing suitable habitat for turtles (Bodie 2001) as well as modifying river flow dynamics. River alterations, such as channelization, riparian zone alteration, and drainage, are associated with changes in demographic characteristics of freshwater turtles, including age structures and sex ratios (Chessman 2011; Usuda et al. 2012). The comparison between the real and pessimistic populations indicates how fluctuations around carrying capacity and demographic characteristics are important to population persistence.
The H. maximiliani population growth rate (r) simulation did not differ between the real and optimistic scenarios. However, final population size after a 100-yr period was significantly different between scenarios since other variables were incorporated into the models, including carrying capacity, growth, mortality rates, catastrophic events, and loss of genetic variability. A smaller carrying capacity could represent a smaller population recovering capacity (or a small suitable area for the species) in years with pronounced environmental and demographic stochasticity, such as inbreeding depression and population decline, which in turn would lead to higher extinction probability (Wedekind 2003).
Sensitivity tests can be helpful in detecting which environmental or demographic parameters represent potential influences on population viability (McCarthy et al. 1995). The application of sensitivity analysis within a PVA will identify parameters that consistently influence the dynamics of particular parameters in real-world populations (Naujokaitis-Lewis et al. 2009). For the RP scenario, all variables were important in sensitivity tests, changing significantly the population persistence that in turn influenced the baseline scenario (McCarthy et al. 1995; Lacy et al. 2009).
Changes in mortality rates altered H. maximiliani mean population extinction probability. The available data for the present study apply only to mature males and females (Martins and Souza 2008), although population dynamics could change if variable juvenile mortality rate were included in viability simulations (Miller 2001). Analyses of a Clemmys guttata population from Georgian Bay, Canada, using stage-classified matrices indicated that population persistence requires high juvenile survivorship, indicating that these life stages must be considered in conservation strategies (Enneson and Litzgus 2008). Moreover, sources of extinction risk that increase mortality rates of reproductive individuals and perturb the balance between fecundity and longevity can be particularly harmful for species like turtles that have long life spans and delayed maturation (Gibbons et al. 2000; Carrete et al. 2009). Despite more than 10 yrs of research on H. maximiliani in PECB, information about fecundity and juvenile mortality does not exist. In fact, the lack of information on demographical parameters can produce misinterpretations of the output of the PVA (Boyce 1992) and even PVA invalidation if one is not confident that the data captured the distribution of population growth rates or vital rates (Coulson et al. 2001). Given that a 10% variation in mortality rates resulted in significant differences across PVA scenarios, the actual H. maximiliani population could even be more fragile if faced with an accentuated juvenile mortality. An increase or decrease in population growth rate represented a change in extinction probability in the OP and PP scenarios. However, PP exhibited a more profound effect of this variable.
One of the main problems in biological conservation is the impact of habitat fragmentation on natural populations (Hoffmann et al. 2010; Pereira et al. 2010). Research focused on detecting these impacts on biodiversity is important for conservation planning and management strategies. Although the present research was conducted in a protected area, this is not the situation for H. maximiliani populations throughout the species' geographical distribution. In fact, several populations can be found outside protected areas (Souza and Martins 2009), configuring the actual hypothetical OP and PP scenarios. Population viability analysis is an interesting tool to examine population scenarios in wild populations. The interesting and consistent results obtained here must be viewed as a first practical approximation to verify how apparently well-established populations can be prone to extinction even in protected areas and show the importance of long-term research, particularly for turtles, given the difficulties in tracking long-lived organisms throughout their entire life spans.
Comparison of the persistence between baseline scenarios for the freshwater turtle Hydromedusa maximiliani in a 100-yr time frame.
