The raccoon (Procyon lotor) was introduced to Japan from North America, and has been implicated in the population declines or local extinctions of native freshwater turtles. We conducted a capture–recapture study of the Japanese pond turtle (Mauremys japonica) at two sites in the Boso Peninsula, Japan, to investigate the effects of P. lotor predation on the demography of native turtle populations, during the period from 2014 to 2017, shortly after an invasion of P. lotor. The study revealed a steady decline in turtle population size, along with changes in age structure, characterized by a skew toward older age classes. We also found a number of living individuals with amputated limbs, which we suspect to be evidence of P. lotor predation. Such injuries highlight the importance of the early detection of population declines and identification of related warning signals or demographic characteristics, which in turn will enable the timely implementation of appropriate conservation measures to prevent local extinctions of freshwater turtles.
The Japanese pond turtle (Mauremys japonica) is endemic to Japan, where it is distributed in the islands of Honshu, Shikoku, Kyushu, and the adjacent small Japanese islands (Yasukawa et al., 2008). Although the eggs, hatchlings, and juveniles of M. japonica are preyed upon by native predators, such as Japanese striped snake (Elaphe quadrivirgata), Japanese giant salamander (Andrias japonicus), weasel (Mustela itatsi), crows (Corvus spp.), and raccoon dog (Nyctereutes procyonoides) (Yabe, 2002; Matsukubo, 2005; Kosuge and Kobayashi, 2015), mature and large adult turtles are generally at little risk from predation by native predators (Yabe, 2002; Yasukawa et al., 2008). Consequently, Ma. japonica used to be a common and sometimes abundant freshwater turtle in Japan (Yasukawa et al., 2008).
However, anthropogenic threats including habitat loss, river alteration, resource competition (e.g., for food and basking sites) with the red-eared slider (Trachemys scripta elegans), hybridization with Reeves' pond turtle (Ma. reevesii), and overexploitation for the pet trade (Yasukawa et al., 2008; Usuda et al., 2012; Yabe, 2014) all threaten this species. In recent predation by the feral raccoon Procyon lotor native to North America, has also been recognized as a novel threat to Ma. japonica populations (Kosuge and Kobayashi, 2015; Ogano et al., 2015a). Although there is circumstantial evidence that Ma. japonica underwent rapid declines in local and restricted areas where P. lotor has appeared (Kosuge and Kobayashi, 2015; Ogano et al., 2015a), the processes and mechanisms underlying these population declines have yet to be established. In Japan naturalists and herpetologists suspect that the recent spread of P. lotor is a prominent causal factor in the recorded population declines and local extinctions of Ma. japonica in formerly P. lotor-free freshwater habitats (Kosuge and Kobayashi, 2015; Ogano et al., 2015a).
Procyon lotor preys not only on the eggs and juveniles of freshwater turtles but also on adults (Seigel, 1980; Tucker et al., 1999; Feinberg and Burke, 2003; Karson et al., 2019). The findings of previous studies (e.g., Congdon et al., 1993, 1994; Enneson and Litzgus, 2008) have indicated that populations of freshwater turtles with slow life-history strategies are vulnerable to even small reductions in adult survivorship, with as little as a 2–3% reduction in survivorship resulting in severe population declines (Keevil et al., 2018). Therefore, P. lotor may cause substantial reductions in native turtle populations in countries such as Japan where it has been introduced (Kosuge and Kobayashi, 2015; Ogano et al., 2015a).
To obtain basic information on the demographic risk posed by P. lotor to native freshwater turtles, we conducted a capture–recapture study of Ma. japonica during the initial period when feral P. lotor invaded the study area, in order to identify the warning signals of P. lotor predation. We detected a steady population decline, characterized by an age structure skewing toward older age classes and an increasing number of living individuals with amputated limbs, which we considered to be evidence of P. lotor predation (Kosuge and Kobayashi, 2015; Ogano et al., 2015b; Suzuki et al., 2015). On the basis of these findings, we discuss strategies for the early detection of serious declines in the populations of freshwater turtles, thereby enabling the implementation of timely and appropriate conservation measures to prevent local extinctions.
Materials and Methods
During the period between late July and late September in the years from 2014 to 2017, we conducted a capture–recapture study of Ma. japonica along approximately 3 and 1.5 km reaches of two small rivers in Tateyama City (the Sunomiyagawa and Sanogawa rivers, respectively) at the southern tip of the Boso Peninsula, Chiba Prefecture, Japan (Fig. 1). The study area is located within a region of Chiba Prefecture containing habitats favorable to Ma. japonica (Ogano et al., 2015a), owing to an absence of anthropogenic disturbance, such as extensive river alteration practices. Moreover, though hybridization between Ma. japonica and the non-native Ma. reevesii has occurred (Kagayama et al., 2017), there is an absence of potentially detrimental biotic impacts associated with the presence of the non-native competitor T. s. elegans. Although the study area is inhabited by potential native and nonnative predators of the eggs, hatchlings, and juveniles of Ma. japonica (e.g., Japanese striped snakes, raccoon dogs, weasels, crows, and feral cats), there are no major predators of adult female Ma. japonica.
Given these aforementioned characteristics, the study area could be regarded as an important area in which to study the temporal dynamics of the detrimental effects of P. lotor on native turtle populations. In the area around Tateyama City, no P. lotor individuals were captured prior to 2009 (Asada and Shinohara, 2009; Asada, 2011), and the number of captures remained low until 2012 (Asada, 2014). In the Boso Peninsula, P. lotor initially established breeding populations in the late 1990s in the cities of Isumi and Onjuku on the southeastern side of the peninsula, which are assumed to have originated from individuals that were either accidentally or intentionally released (Ochiai et al., 2002). Further founding populations of P. lotor were detected in the cities of Kimitsu and Futtsu, located in the northwestern part of Boso mountain (Yoshida et al., 2020). The time-lag in the appearance of P. lotor in Tateyama could be ascribed to the time required for individuals to disperse from one or both of these locations.
In each of the administrative units in Chiba Prefecture, the programs to control and reduce the agricultural damage caused by P. lotor have been implemented in each fiscal year by capturing individual using baited traps, and catch per unit effort (CPUE: the number of removed individuals/100 trap-days) data have been compiled by Chiba Prefecture since 1999. For the purposes of the present study, we used CPUE data for Tateyama City collected during the period from 2013 to 2017 (fiscal years) to examine causal relationships between trends in the local populations of Ma. japonica and P. lotor in the study area.
Turtles were captured using traps (72 cm× 44 cm×55 cm) baited with fish, which incorporated a 1.7-m-long bag-like passage for the trapped turtles to breath air (Fig. 2). The traps were prepared in the afternoon of the day prior to deployment and inspected the following morning. Each year, we conducted four to seven capture–recapture sessions using 14 or 18 traps in the Sunomiyagawa River and three to five capture–recapture sessions using 12 or 14 traps in the Sanogawa River (Table 1). Furthermore, during the day, we walked along a channel (approximately 20 m in length) connected to the Sunomiyagawa River and captured all observed turtles by hand. These turtles were marked to identify individuals following the method described by Kobayashi (2008). For each of the captured turtles, we recorded sex, age, carapace length (CL), and limb amputation status. Sex was categorized as female, male, or undetermined (unsexed), based on the relative position of the cloaca (Yabe, 1989). Turtle age was estimated by counting the annuli on the plastron scutes (Sexton, 1959). All individuals with effaced carapaces and uncountable plastron scutes and/or estimated to be over 11 years of age were classified as old individuals. To determine trends in the population dynamics of male, female, and juvenile turtles, males with a CL>80 mm (approximately >three years old) and females with a CL>150 mm (approximately >eight years old) were defined as adult turtles (Yabe, 1989, 1992; Yasukawa et al., 2008; Kagayama, 2020a). In contrast, zero- (hatchlings) to four-year-old turtles (including unsexed individuals and females, but excluding males; see the following section) were defined as juveniles. Turtles that had at least one amputated limb were recorded as limb-amputated individuals.
Number of sampling sessions and traps used per year.
Population size estimation
To estimate the population size of Ma. japonica in the Sunomiyagawa and Sanogawa rivers from 2014 to 2017, we applied the Jolly–Seber (JS) model (Jolly, 1965; Seber, 1965), based on the state-space formulation of Bayesian hierarchical models and the Just Another Gibbs Sampler (JAGS) code described by Kéry and Schaub (2012a, b). We made extensive use of parameter-expanded data augmentation (Royle et al., 2007; Kéry and Schaub, 2012a; Royle and Dorazio, 2012), which entails augmenting a dataset by adding a large number of potentially unobserved individuals, all with zero-only encounter histories. The key idea is to fix the dimension of the parameter space in the analysis by augmenting the observed data with a large number of all-zero capture histories, resulting in a larger dataset with a fixed dimension M. This augmented dataset is then analyzed using a reparameterized (zero-inflated) version of the model that could be applied if the superpopulation size (the number of individuals that survived during the study) is known (Kéry and Schaub, 2012a). Moreover, we adopted the superpopulation parameterization approach for the JS model (see details in Kéry and Schaub, 2012a). This model can be used to estimate the superpopulation size (Nsuper), population size in each year (N), capture probability (P), survival rate (Phi), number of entries (B), entry probability (b), and inclusion probability (Psi). Psi is the probability that a member of the augmented dataset, M (M individuals), is included in Nsuper.
Given that the survival rate of turtles is highly dependent on their body size and life-history stage (Iverson, 1991), we used separate models for adults and juveniles to estimate their respective population sizes from 2014 to 2017. Furthermore, as the survival rates of males and females of different body sizes can potentially differ, we assumed that survival rate of adults is sex-dependent. We also assumed a constant capture probability (P) and time-dependent entry probabilities (b). In this model, we used the capture histories of 34 adult males (Sunomiyagawa River: 19 individuals; Sanogawa River: 15 individuals) and 47 adult females (Sunomiyagawa River: 33 individuals, Sanogawa River: 14 individuals). For both males and females, the datasets were augmented with 500 individuals with all-zero capture histories. The JS model for the adults was constructed based on the JAGS code described by Kéry and Schaub (2012b) (Appendix I).
In contrast, those individuals marked (identified) when they were between zero (hatchlings) and four years old were defined as juveniles (including unsexed individuals and females). Given that males attain adulthood at approximately three years of age, we excluded male turtles from our estimations of juvenile population sizes. In this model, we assumed constant survival (Phi), capture probabilities (P), and time-dependent entry probabilities (b). We used the capture history of 80 juveniles, and augmented the model with 1000 individuals with all-zero capture histories. The JS model for juveniles was constructed based on the JAGS code described by Kéry and Schaub (2012a) (Appendix II). However, given the small number of samples obtained for the Sanogawa River, we did not estimate the population size of juveniles in this river.
We estimated the parameters of the JS model using Markov Chain Monte Carlo (MCMC) methods in JAGS ver. 3.4.0 (Plummer, 2003) via the R2jags package (Su and Yajima, 2015) in R ver. 3.5.0 (R Core Team, 2018). We discarded the initial 50,000 MCMC samples as burn-in, and performed an additional 50,000 MCMC iterations, saving every 100th iteration to reduce serial autocorrelation among the adult turtle samples. In contrast, for juvenile turtle samples, we discarded the initial 100,000 MCMC samples as burn-in and performed an additional 100,000 MCMC iterations, saving every 200th iteration to reduce serial autocorrelation. We confirmed convergence by assessing whether the R value was smaller than 1.1 for all estimates. We summarized the posterior distributions of all parameters using the mean of all MCMC samples as a point estimate and used the 2.5 and 97.5 percentiles of the MCMC samples as a 95% credible interval (95% CI).
Current status of Ma. japonica in a P. lotor invasion area
We marked 181 Ma. japonica individuals (identified individuals: 144 and 37 in the Sunomiyagawa and Sanogawa rivers, respectively). Turtles of all life-history stages, from hatchlings and juveniles to adults, were captured from both rivers. However, during the study period the age structure of females in the Sunomiyagawa River tended to be skewed toward older adults (Fig. 3). During the survey period from 2014 to 2017, the number of captured turtles declined from 57 to 35 in the Sunomiyagawa River, and from 18 to 11 in the Sanogawa River, as the proxy for P. lotor abundance (CPUE) increased (Fig. 4A). Correspondingly, the number of turtles recorded with amputated limbs increased from two in 2014 to six in 2017 in the Sunomiyagawa River, and from one in 2014 to four in 2017 in the Sanogawa River (Fig. 4B). Of a total of 232 turtles examined, limb amputation was recorded in 29 individuals (Table 2). The incidence of limb amputation tended to increase with age (Table 2).
Population dynamics of adult and juvenile turtles
All parameters estimated in the adult population dynamics model are shown in Appendix III. For the Sunomiyagawa River, the estimated mean population size of adult female and male turtles decreased from 35.28 (95% CI: 25.00–50.00) to 26.65 (14.00–47.00) and 20.46 (9.00–37.00) to 10.12 (5.00–21.00), respectively, during the period from 2014 to 2017 (Fig. 5). Similarly, the estimated mean population size of adult female and male turtles in the Sanogawa River decreased from 22.65 (95% CI: 13.00–43.00) to 10.42 (5.00–25.00) and 19.93 (9.00–38.53) to 13.77 (5.00–29.53), respectively (Fig. 5). The rate of population decline for the past three years was calculated to be 0.66 ([26.65+10.12]/ [35.28+20.46]) for the Sunomiyagawa River and 0.57 ([10.42+13.77]/[22.65+19.93]) for the Sanogawa River. These rates of decline corresponded to the annual changes of population size (R) of 0.87 (a product of R3=0.66) and 0.83 (a product of R3=0.57), respectively. The detection probabilities of adult turtles were estimated to be 0.44 (0.27–0.62) and 0.36 (0.17–0.58), for the Sunomiyagawa and Sanogawa River populations, respectively. The annual adult survival rates of females and males were estimated to be 0.76 (0.59–0.93) and 0.31 (0.06–0.69) in the Sunomiyagawa River, and 0.58 (0.34–0.82) and 0.68 (0.37–0.94) in the Sanogawa River.
Summary of the number of Mauremys japonica individuals with amputated limbs recorded during the period 2014–2017. The same individuals that were recaptured in different years were counted as different individuals to account for differences in age.
All parameters estimated for the juvenile population dynamics model are shown in Appendix IV. Although the estimated size of the juvenile turtle population was higher in 2016 than in other years of the study period, it showed an overall slight reduction [from 49.31 (36.00–66.00) to 33.67 (24.00–46.00)] from 2014 to 2017 (Fig. 6). The juvenile detection probability and survival rate were 0.23 (0.01–0.75) and 0.30 (0.18–0.44), respectively. All parameters converged as the R value did not exceed 1.1.
Our study revealed a steady decline in turtle population sizes (0.66 and 0.57 for the three years) at the southern tip of the Boso peninsula, where the feral raccoons recently appeared. The annual rates of population change estimated for adult turtles were 0.87 and 0.83, which were lower than 1.0 of the assumed stable population sizes. If the sizes of Ma. japonica populations continue to decline at this rate, the populations are likely to undergo severe declines (to approximately 10% of the current population sizes) within 15 years. Similarly, Kosuge and Kobayashi (2015) reported mass mortalities and declines in the populations of Ma. japonica and Ma. reevesii in 2008 in a river in Kimitsu City. Kosuge and Kobayashi (2015) suspected that the mass mortalities (the finding of dead turtles), living turtles with amputated limbs, and rapid population declines recorded in Kimitsu were attributable to P. lotor predation. Therefore, the steady decline in the population of Ma. japonica in Tateyama City concomitant with the appearance and increase in CPUE of P. lotor in Tateyama since 2009 (Ochiai et al., 2002; Asada and Shinohara, 2009; Asada, 2011, 2014) is compatible with the hypothesis that P. lotor can be a source of marked declines in the populations of Ma. japonica. In this regard, however, it should be acknowledged that we obtained no direct evidence of P. lotor predation, such as P. lotor gut contents or photographic records of predation events. In the following paragraphs, we discuss the causal factors and mechanisms associated with limb amputation (Kosuge and Kobayashi, 2015; Ogano et al., 2015b; Suzuki et al., 2015; Tagami et al., 2019) and the skewed population age-size structure, as proxies of predation impacts by P. lotor on native freshwater turtle populations.
Reduced recruitment of newly matured adults from juveniles would potentially skew the population structure toward older-aged individuals. Low reproductive output through low egg production, high egg mortality, or the high mortality of juveniles (including hatchlings) due to predation would all contribute to reducing the recruitment of newly matured individuals (Gibbons, 1968; Wilbur, 1975; Tinkle et al., 1981; Browne and Hecnar, 2007; Dorcas et al., 2007; Chen and Lue, 2009). However, even though we annually recorded reasonable numbers of adult females capable of producing large clutches and numerous newly hatched turtles, the higher survivorship of adult turtles in general (Wilbur, 1975; Iverson, 1991; Congdon et al., 1993, 1994) would not compensate for the low annual survival of hatchlings and reduced recruitment of newly matured adults to sustain a stable population size.
In the present study, we found that those turtles characterized by amputated limbs were generally adult males and females rather than juveniles (Table 2). However, a low incidence of juvenile turtles with injured limbs is not necessarily indicative of a low rate of predation, as it is more likely to reflect the high mortality rate of predated individual, and that predators tend to leave no traces of dead individuals (Ogano et al., 2015b). The turtles could also lose limbs because of lawnmower activities. When the turtles were injured by lawnmower operation, severe scars were left on turtle carapaces (Kagayama, 2020b). The fact that no such scarring was detected in the limb-amputated turtles identified in the present study would thus tend to indicate that limb amputation was caused by predation rather than by human disturbance. Furthermore, although further confirmatory data are required, we did find some evidence of skewed limb amputation rates in the turtle population, indicating size-specific predation on smaller turtles. This could reflect the fact that large and heavy turtles may escape P. lotor predation, as these P. lotor generally consume smaller prey items that they can conveniently grasp with their forefeet (Kosuge and Kobayashi, 2015), and it is assumed that P. lotor would experience difficulties in handling larger prey items (particularly adult female Ma. japonica).
Ogano et al. (2015a) noted that although feral P. lotor population management in Japan has been conducted mostly from the perspective of agricultural damage control, measures should also be taken to eradicate feral P. lotor for the purposes of biodiversity conservation. Even in its native range in North America, P. lotor has purposely been removed from certain areas for the conservation of freshwater and marine turtles (e.g., diamondback terrapins and loggerhead turtles), in order to reduce predation pressure at vulnerable lifestages for native turtles (Garmestani and Percival, 2005; Barton and Roth, 2007; Munscher et al., 2012). However, whether it be for agricultural control or biodiversity conservation, early detection of P. lotor predation is pivotal to the planning and implementation of conservation programs for native turtle populations in Japan.
We would like to thank Chiaki Iida, Hiroko Iida, Kazuya Yoshida, Masamu Tsujii, Tomoko Nishibori, and Megumi Kondo for their assistance during the field study. Additionally, we thank Chiba Prefecture for granting permission to use the CPUE data for non-native P. lotor and deploy fish-baited traps in the study area. We are also grateful to the anonymous reviewers for their valuable comments regarding our manuscript. This study was funded by the Environment Research and Technology Development Fund (4-1705) of the Environmental Restoration and Conservation Agency.
Posterior distribution of the estimated parameters of the Jolly-Seber model for adult turtles in (A) the Sunomiyagawa River and (B) the Sanogawa River.
Nm: actual male population size, Nf: actual female population size, Nsuperm: size of the male superpopulation, Nsuperf: size of the female superpopulation, Bm: number of male entries, Bf: number of female entries, b: entry probability, Phim: annual survival rate of males, Phif: annual survival rate of females, P: mean detection probability, and Psi: inclusion probability.
Posterior distribution of the estimated parameters of the Jolly Seber model for juvenile turtles in the Sunomiyagawa River. N: actual population size of juvenile turtles, Nsuper: size of the juvenile superpopulation, B: number of juvenile turtle entries, b: entry probability of juveniles, Phi: annual survival rate of juvenile turtles, P: mean detection probability, and Psi: inclusion probability.