Biological Invasions

, Volume 19, Issue 9, pp 2739–2750 | Cite as

Challenges in confirming eradication success of invasive red-eared sliders

  • Pablo García-Díaz
  • David S. L. Ramsey
  • Andrew P. Woolnough
  • Marc Franch
  • Gustavo A. Llorente
  • Albert Montori
  • Xabier Buenetxea
  • Asier R. Larrinaga
  • Matthieu Lasceve
  • Alberto Álvarez
  • José María Traverso
  • Aitor Valdeón
  • Ariñe Crespo
  • Virginia Rada
  • Enrique Ayllón
  • Vicente Sancho
  • J. Ignacio Lacomba
  • José Vicente Bataller
  • Miguel Lizana
Original Paper

Abstract

Confirming eradication success can be notoriously difficult and costly, especially when the species is still present but remains undetected, due to very low population densities and imperfect detection methods. There has been a lack of research on appropriate guidelines and estimation procedures for declaring eradication success for programs aimed at eradicating alien reptiles. Here we develop quantitative rules for confirmation monitoring in eradication campaigns of the red-eared slider turtle (Trachemys scripta elegans). We used a database of slider trapping data from control and eradication campaigns conducted in localities across the Iberian Peninsula and southern France to construct models for inferring appropriate trapping efforts for confirming slider turtle eradication. Basking traps were slightly more efficient than net traps in capturing sliders, although trapping was an inefficient monitoring method given the low capture probabilities estimated. The results of our spatially-explicit eradication scenarios revealed the importance of habitat configuration in declaring eradication success. Declaration of eradication success is contingent on the thresholds set to minimise false positives (i.e., falsely declaring eradication successful), but in any scenario large trapping efforts were required to confirm eradication. Given the low estimated capture probabilities, alternative methods such as eDNA and visual surveys should be considered for monitoring sliders. We suggest that if the costs associated with the impact of alien sliders can be adequately estimated, then eradication can be confirmed by rules minimising both false positive and negative error rates. Otherwise, rules minimising false positive errors would be more appropriate.

Keywords

Alien reptile Capture probability Environmental decision-making False positive Monitoring Trachemys scripta elegans 

Introduction

The eradication of alien species is commonly desirable to eliminate their environmental, social, and economic impacts (Tobin et al. 2014; Hoffmann et al. 2016; Jones et al. 2016). Eradication is a challenging task, which requires careful planning and laying a clear set of goals to be achieved. An especially critical stage in any eradication program occurs at the culmination of the program when it is necessary to ensure the full eradication of the species targeted. Declaring a species eradicated when it is still extant can prompt the recovery of the species, therefore representing program failure and a waste of resources (Ramsey et al. 2011; Rout et al. 2014). The uncertainty due to imperfect detection of the target species means that proof of eradication is difficult and decisions may need to be made with incomplete information (Regan et al. 2005; Guillera-Arroita et al. 2014; Groves and Game 2016). This raises the issue of how much monitoring effort, without detecting the focal species, is required to conclude that the species has been successfully eradicated with a high threshold level of certainty (Ramsey et al. 2009; Guillera-Arroita et al. 2014; Rout et al. 2014).

Quantitative approaches to accurately determine when to declare a species eradicated are available for mammalian predators, invasive plants, and ants, but these tools do not currently exist for alien reptiles (Regan et al. 2006; Ramsey et al. 2011; Burgman et al. 2013; Rout et al. 2014; Ward et al. 2016). Considering the growing numbers of alien reptiles being introduced worldwide (García-Díaz et al. 2016; Seebens et al. 2017), the lack of information hinders evidence-based decision-making in guiding management efforts for alien reptiles. The red-eared slider (Trachemys scripta elegans) (slider, hereafter) is the most common introduced turtle worldwide, and it has caused substantial negative impacts on native turtles in the recipient regions (García-Díaz et al. 2015; Iglesias et al. 2015; Pearson et al. 2015; Héritier et al. 2017). Control and eradication campaigns have been conducted around the world to minimise or eliminate the negative impacts of sliders (O’Keeffe 2009; Ficetola et al. 2012), activities with a need for appropriate methods to assist in declaring success. Otherwise, slider eradication campaigns risk resulting in wrongfully halting eradication efforts preventing the achievement of the conservation gains derived from their removal.

Here, we develop robust statistical rules for quantifying the trapping effort required to ensure successful eradication of invasive sliders. We compiled a database of slider population control and eradication campaigns that we have been conducting in the Iberian Peninsula and southern France (Franch et al. 2007; Valdeón et al. 2010; Lasceve 2014; Sancho and Lacomba 2016). These campaigns varied in the effort and type of devices used to capture and remove turtles from the environment, illustrating a range of situations of interest to environmental managers. We focus on traps as monitoring devices, and our methods apply to the moment when trapping for the purpose of capture fails to catch sliders, and trapping efforts are further extended to confirm the eradication of the species. We also demonstrate approaches for conducting eradication programs in areas where the occurrence of sliders may vary spatially by using methods that account for environmental heterogeneity. Hence, these methods provide great flexibility in both the design and quantitative approach used to confirm eradication success.

Materials and methods

Trapping sliders

We used the results of 44 trapping campaigns where 508 sliders were captured and removed from 31 sites across the Iberian Peninsula and southern France (Tables 1, 2). Trap models employed could be classified into two main types: basking and net traps (including minnow, fyke and crayfish traps; Figs. 1 and A1 in Online Resource 2). Captured turtles were always removed from the environment. The trapping and removal of sliders was always undertaken under the required authorisations issued by the respective jurisdictional government environmental agencies.
Table 1

Summary of slider removal campaigns conducted using net traps considered in this study

Locality

Region

Latitude

Longitude

Year

Occasions

Total trap nights

Total captures

Caserío del Henares

Madrid

40.42

−3.49

2010

5

88

6

Caserío del Henares

Madrid

40.42

−3.49

2011

4

36

4

Caserío del Henares

Madrid

40.42

−3.49

2012

4

32

3

Irati

Navarra

42.71

−1.33

2011

5

32

1

Bardenas

Navarra

42.29

−1.56

2013

7

42

1

Aena

Catalunya

41.30

2.08

2005

4

31

19

Ca l’Arana

Catalunya

41.30

2.13

2006

14

77

2

Ca l’Arana

Catalunya

41.30

2.13

2008

7

36

2

Cal Tet

Catalunya

41.30

2.12

2005

3

25

1

Canal de la Bunyola

Catalunya

41.31

2.11

2004

2

2

4

La Ricarda

Catalunya

41.30

2.11

2004

6

34

1

Llera Nova

Catalunya

41.31

2.12

2006

6

62

1

Parets de la Murtra

Catalunya

41.29

2.02

2008

5

30

7

Remolar-Filipines

Catalunya

41.28

2.06

2003

21

65

73

Sant Daniel

Catalunya

42.00

2.84

2006

2

42

1

Sèquia Major

Catalunya

41.08

1.18

2008

3

45

2

Sèquia Major

Catalunya

41.08

1.18

2010

3

45

2

Sils

Catalunya

41.80

2.74

2011

2

17

3

Terciari

Catalunya

41.31

2.12

2004

2

7

1

Baix Ter

Catalunya

42.05

3.19

2008

17

570

6

Site 1—Vieux Salins

Hyères

43.12

6.23

2013

10

32

30

Site 2—Vieux Salins

Hyères

43.12

6.23

2013

15

44

34

Site 3—Vieux Salins

Hyères

43.12

6.23

2013

3

6

1

Site 4—Vieux Salins

Hyères

43.12

6.23

2013

5

10

8

Arkaute

Basque Country

42.86

−2.64

2013

18

551

6

Betoño

Basque Country

42.86

−2.65

2013

10

210

1

Zadorra

Basque Country

42.86

−2.71

2013

5

85

2

Aramangelu

Basque Country

42.87

−2.68

2014

6

90

1

Arkaute

Basque Country

42.86

−2.64

2014

8

323

1

Zadorra Dep

Basque Country

42.86

−2.71

2014

4

48

2

Zadorra Gobeo

Basque Country

42.86

−2.71

2014

4

8

2

Table 2

Summary of slider removal campaigns conducted using basking traps considered in this study

Locality

Region

Latitude

Longitude

Year

Occasions

Total trap nights

Total captures

Boadilla del Monte

Madrid

40.40

−3.85

2013

5

105

97

El Escorial

Madrid

40.58

−4.14

2013

7

98

14

Manzanares El Real

Madrid

40.72

−3.85

2013

6

120

31

Navalagamella

Madrid

40.47

−4.12

2013

3

76

11

Villaviciosa de Odón

Madrid

40.36

−3.89

2013

13

146

35

Caserío del Henares

Madrid

40.42

−3.49

2012

6

12

7

Caserío del Henares

Madrid

40.42

−3.49

2013

4

8

3

Linares del Arroyo Reservoir

Castilla y León

41.53

−3.56

2014

10

436

38

Pamplona

Navarra

42.83

−1.64

2008

9

79

7

Baix Ter

Catalunya

42.05

3.19

2006 and 2007

12

323

24

Baix Ter

Catalunya

42.05

3.19

2008

18

466

14

Fig. 1

Estimated relationship between the trapping effort per occasion and the capture probability of sliders using two types of traps. Mean and 95% credible intervals were drawn from the posterior distribution of the complementary log–log function. Examples of a net (or crayfish trap) and a basking trap are shown as insets. Photos: Andrés Rodríguez-Pereira (University of Salamanca)

We defined a trapping occasion as the period between consecutive trap checks for captured turtles. For each trapping site and occasion, we recorded the type of trap used, the trapping effort (i.e., the number traps per trapping occasion × number trapping occasions), and the number of sliders captured per occasion. We considered trapping activities conducted at the same site, but in different years, as being independent. This implies that populations were closed while the annual trapping efforts were conducted (i.e., no migration, mortality, recruitment, or release of new individuals), but open in periods between annual trappings. This assumption is appropriate given the short timeframes within which annual trapping efforts were conducted and the geographical independence of the sites. Overall, we analysed data with a cumulative trapping effort of 4594 trap-nights, distributed over 124 occasions across six main geographical regions of the Iberian Peninsula and southern France (Tables 1, 2).

Removal models

We developed closed-population Bayesian removal models for estimating slider population size and capture probabilities during the annual trapping efforts. The number of turtles removed in each trapping occasion i (yi) was dependent on the population size of turtles just before occasion i and the capture probability of the device in use:
$$y_{i} {\sim}Bin(N_{i},p_{i}),$$
(1)
where Ni is the population size just before the removal during the occasion, i and pi is the capture probability during occasion i. The population size at the start of each occasion i was given by:
$$\begin{aligned} N_{i} & = N_{0} - X_{i} \\ N_{0} & {\sim}Poisson(\lambda) \\ \end{aligned}$$
(2)
where N0 is the initial population size of turtles just before the commencement of each of the annual trapping activities, drawn from a Poisson distribution with rate λ, and Xi is the cumulative number of turtles removed by trapping just before occasion i. As the amount of trapping effort per occasion often varied, we extended (1) by modelling the capture probability pi to be a function of trapping effort via a complementary log–log link:
$${\text{cloglog}}\left( {p_{i} } \right) = \alpha + \log \left( {TE_{i} } \right),$$
(3)
where α is the (log) capture rate per trapping effort, and TEi is the trapping effort for occasion i. Under this model, the (log) capture rate per trapping effort α, is assumed to be constant. However, the capture rate may increase or decrease over time due to heterogeneity in capture rates among individual turtles. We estimated this effect using a modified version of (3):
$${\text{cloglog}}\left( {p_{i} } \right) = \alpha + \log \left( {TE_{i} } \right) + \beta \log \left( i \right),$$
(4)
where the parameter β is the effect of the (log) occasion number (i) on the capture rate. We fitted (3) and (4) independently to the net and basking annual trapping data using Markov Chain Monte Carlo methods as implemented through the R statistical environment (R Development Core Team 2015) interface to the JAGS software (Plummer 2003). We compared the support for the models in (3) or (4) using the Deviance Information Criterion (DIC) (Spiegelhalter et al. 2002). We used a relatively uninformative Normal prior for the parameters α and β of the complementary log–log function, N(0, σ2 = 10). We used uniform priors for the (natural log) initial population size for the model for net U(0, 5.5) and basking U(1, 5) traps. Each type of trap was generally used in different sites and conditions (Tables 1, 2), so we used these prior distributions to accommodate such site-specific differences. Models were fitted by using three chains with 250,000 iterations each, and a thinning of five. We obtained the marginal posterior distributions of the parameters of the models after discarding the first 50,000 iterations of each chain (burn-in time), and visually confirming the adequate mixing and convergence of the chains.

We assessed the fit of our removal models to the data using a posterior predictive check to calculate a Bayesian p value (Gelman et al. 1996). The Bayesian p-value was calculated from the posterior distributions of the discrepancy between the observed (y) and the predicted removal data (ypred) given the model, using the sum of the deviance residuals as our test statistic. We also calculated the model coverage, measured as the actual number of turtles removed per occasion that fell within the 95% CIs of the predicted number removed. Finally, we validated our removal model predictions using an independent slider net trapping dataset (detailed in Online Resource 1 and Fig. A3 in Online Resource 2).

‘Stopping rules’ for declaring eradication success

We assumed that trapping is conducted in ‘sessions’ (t) consisting of a particular trapping effort and spatial arrangement set at a particular point in time (e.g., month, year). Multiple sessions will usually be conducted during confirmation monitoring, and here we are concerned with finding the optimal number of trapping sessions t* to undertake to declare successful eradication once there are no new captures.

Spatial heterogeneity in the distribution of the target species and habitat configuration influences the design and outcomes of eradication and control campaigns (Epanchin-Niell et al. 2012; Anderson et al. 2013; Russell et al. 2016). We accounted for spatial variation in risk and trapping effort using the approach described in Anderson et al. (2013). The spatial approach involves dividing the area of interest into grid cells. Alien species establishment risk and trapping effort are then defined for each grid cell (which may be zero), and then the probability of successful eradication is calculated by aggregating results over all cells. A full explanation of this approach is given in Anderson et al. (2013), and here we provide an overview.

The total area subject to eradication activities A is divided into N cells, which are assumed to be of equal area. The probability of capturing a slider anywhere within the area of interest A is then given by:
$$P_{T} = 1 - \left( {1 - P_{ave} \times \frac{n}{N}} \right)^{EPIavg \times N} ,$$
(5)
where PT is the total capture probability (over all cells), Pave is the average capture probability within a grid cell, and n is the number of grid cells subject to monitoring. Pave was empirically estimated for both net and basking traps from the removal models described in the previous section. As we used a complementary log–log link for capture probabilities (3), the capture probability for any particular cell (Pcell) could be calculated as
$$P_{cell} = 1{-}\exp \left( { - \exp \left( { - \alpha + \log \left( {TE_{cell} } \right)} \right)} \right)$$
(6)
where \(\alpha\) is the trapping rate per unit of trapping effort in Eq. (3) and TEcell is the trapping effort deployed in the cell. EPIavg in (5) is the effective probability that a given cell is occupied. It incorporates heterogeneity of risk of slider occurrence and is calculated as:
$$EPIavg = \frac{{P_{u} \sum\nolimits_{k = 1}^{n} {AR_{k} } }}{n},$$
(7)
where Pu is the lower bound estimate of the proportion of grid cells with sliders. Pu constraints the overall level of sensitivity of the system for detecting the species. The required search effort and sensitivity to achieve the target probability of eradication will increase with decreasing Pu. When Pu is set to 1/N, the surveillance sensitivity is the probability of capturing a slider given that only one single grid cell is occupied across the landscape. ARk is the adjusted risk for each grid cell k, which is given by:
$$AR_{k} = N \times {{RR_{k} } \mathord{\left/ {\vphantom {{RR_{k} } {\sum\nolimits_{k = 1}^{N} {RR_{k} ,} }}} \right. \kern-0pt} {\sum\nolimits_{k = 1}^{N} {RR_{k} ,} }}$$
(8)
where RRk is the relative risk for grid-cell k. The RRk are specified with respect to the cell with the lowest risk. If habitat suitability values (HS) are available for each grid cell, then the RRk values can be calculated as:
$$RR_{k} = \frac{{HS_{k} }}{\hbox{min} (HS)},$$
(9)
Finally, given values for (58) above, the probability of successful eradication is given by:
$$P(E|M^{ - } ) = \frac{P(E)}{{1 - P_{T} (1 - P(E))}},$$
(10)
where \(P (E |M^{ - } )\) is the probability that sliders have been eradicated (E) from the area of interest A, given confirmation monitoring fails to detect any individuals (\(M^{ - }\)). P(E) is the prior probability that the species is eradicated (before confirmation monitoring commences) and PT is as defined above (5). For all the calculations, we assumed an uninformative prior Beta(1, 1) for the initial prior probability P(E). This prior was then updated following each trapping session t such that the posterior \(P (E | M^{ - } )_{t}\) becomes the prior for the next session P(E)t+1.
Two stopping rules have been proposed for determining the amount of confirmation monitoring necessary to confidently declare eradication success (Regan et al. 2006; Ramsey et al. 2011). The first stopping rule aims to limit false positive error (falsely declaring eradication successful) whereas the second stopping rule takes a cost-benefit approach to limit both false positive and false negative errors. The second stopping rule requires data on the costs associated with eradication programs, which are the costs of committing either a false positive (wrongly declaring eradication success) or a false negative error (cost of additional monitoring past the point when eradication has already occurred). However, the costs of wrongly declaring eradication of sliders are largely unknown given that it produces non-monetary environmental impacts (Iglesias et al. 2015; Pearson et al. 2015). Given this constraint, we focus on the first stopping rule. The first stopping rule limits the probability of declaring eradication successful when in fact, there are still individuals remaining undetected in the environment (premature declaration). This stopping rule is usually formulated by setting an agreed target for the degree of confidence that a wrong declaration of eradication was made, for example, a stopping rule of 5 or 1% (Ramsey et al. 2009). This equates to 95 and 99% probability that eradication was correctly declared, respectively. We set the level of acceptable error, 5 or 1%, equal to SR (probability that eradication is wrongly declared), and monitoring continues until:
$$P\left( {E|M^{ - } } \right)_{t} \ge \,\left( {1 - SR} \right)$$
(11)
where the expected value (or some quantile) of the posterior distribution is used as the basis for the proposition. If true, then eradication is declared successful after session t.

Case studies

We constructed and simulated two artificial spatially-explicit slider eradication scenarios to estimate the trapping effort, with zero captures, necessary to ensure the eradication of sliders.

Scenario 1, stream

A 3-km × 4-m section of stream inhabited by sliders. For trapping, the stream was divided into 500-m sections, with five net traps deployed per occasion in each section. No basking traps were used for this scenario because we assumed that the stream was not deep enough to set the traps. Net traps were checked every day to minimise non-target impacts. We considered one trapping session to be five consecutive trapping days (i.e., five trapping occasions per session). We chose this setting to reflect resource availability of field operators.

Scenario 2, lake (Fig. A2)

The lake was divided into 100 × 100-m cells for trapping. Cells where the vegetation was considered to be too thick to set traps did not have any monitoring effort (low n/N in [5]). Cells in the centre of the lake also had no monitoring, because we considered this area to be too deep, and thus less suitable for turtles and more difficult for deploying traps (low RR in [9] and low n/N in [5]). The scenario consisted of eight 100 × 100-m cells subject to monitoring (Fig. A2 in Online Resource 2). In this scenario, we simulated trapping using both net and basking traps, as both types of traps are suited for this case. Simulations for net and basking traps were conducted independently. In the case of net traps, we simulated the use of four traps per cell, with traps being checked every day. For basking traps, we ran our simulations using two traps per cell, with traps being checked every five days (i.e., once per trapping session as explained for scenario 1). The design of basking traps ensured that trapped animals could not drown and, thus, they do not need to be checked as frequently as net traps.

We compared our estimates of the trapping effort required to declare successful eradication under stopping rule 1 for scenario 2 (lake), with actual trapping efforts from an independent dataset of four localities in the region of Valencia (Spain). Here, trapping sessions without any captures occurred over the period 2003–2013. However, in most cases after some sessions without captures, sliders were captured again confirming their presence. Nevertheless, a comparison of the predicted amount of trapping required to declare eradication with actual data provides valuable information for managers.

Results

The capture probability increased with the trapping effort for both net and basking traps, with no strong support for a decline in the trapping rate over time (Table 3). Hence, we use results drawn from the models with no decline in trapping rate over time (3) for all further inference. The removal model was an adequate fit to the data (Table 3), and the fit of our model for net traps to the independent validation dataset indicated that the model adequately predicted the observed cumulative number of sliders trapped (Fig. A3 in Online Resource 2).
Table 3

Predictive performance metrics of the Bayesian removal models for sliders captured using net and basking traps

 

Model

DIC

p value

Coverage

Estimates ± SE (95% CI)

α

β

Net traps

Eq. (3)

568

0.25

0.95

−5.35 ± 0.11 (−5.59, −5.15)

 

Eq. (4)

569

0.26

0.95

−5.56 ± 0.19 (−5.96, −5.19)

0.10 ± 0.09 (−0.07, 0.28)

Basking traps

Eq. (3)

500

0.58

0.85

−5.09 ± 0.10 (−5.29, −4.90)

 

Eq. (4)

502

0.58

0.86

−4.93 ± 0.14 (−5.21, −4.66)

−0.17 ± 0.11 (−0.37, 0.04)

DIC Deviance Information Criterion. p value: Bayesian p value for the discrepancy between the deviance residuals for y and yrep. Coverage: the proportion of the actual number of turtles removed per occasion that fell within the modelled 95% credible intervals number of turtles removed on that occasion. Estimates: posterior mean of the detection rate parameters for the relationship between capture probability and trapping effort, α and β in (3) and (4). See "Materials and methods" for further details

Our removal models showed that basking traps were slightly more efficient than net traps at capturing turtles, having a marginally higher capture rate (Fig. 1; Table 3). Nevertheless, the capture probability of turtles per trapping occasion was relatively low regardless of the type of trap employed (Fig. 1). Large trapping efforts were required to offset the low capture probabilities per occasion, with a trapping effort of over 100 basking traps needed to exceed a mean probability of 0.5 (Fig. 1), whereas 100 net traps would only reach a mean probability of 0.38 (Fig. 1).

Consequent with the low estimated probabilities of capture per occasion, our scenario simulations indicated that a large amount of confirmation trapping was required to verify the eradication of turtles, regardless of the type of trap and water body (Figs. 2, 3). For scenario 1 (stream and net traps), eight consecutive trapping sessions (equivalent to 1200 net traps) and nine consecutive trapping sessions (1320 net traps) would be necessary to exceed the 95 and 99% thresholds for stopping rule 1 (t*), respectively (Fig. 2). For scenario 2 (lake), trapping with net traps would need 14 and 16 consecutive trapping sessions (=2240 and 2560 traps, respectively) to reach the thresholds. Using basking traps would require an effort of 17 and 19 consecutive trapping sessions (=1360 and 1520 traps, respectively) for achieving these thresholds (Fig. 3). Comparing our estimates of the trapping effort required for triggering stopping rule 1 for scenario 2 with the trapping data from Valencia revealed that the effort needed for confirming eradication is at least four times higher than has ever been achieved (Fig. 4).
Fig. 2

Estimated probability of eradication of sliders for scenario 1, stream using net traps, as a function of the number of consecutive trapping sessions (one session = five trapping occasions) and the cumulative trapping effort given zero captures. Mean and 95% CIs were drawn from 10,000 iterations. Vertical lines represent the trapping effort necessary to trigger stopping rule 1 (t*); the continuous line represents the 95% threshold, and the dashed line the 99% threshold

Fig. 3

Estimated probability of eradication of sliders for scenario 2, lake using net and basking traps, as a function of the number of consecutive trapping sessions (one session = five trapping occasions) and the cumulative trapping effort given zero captures. Mean and 95% CIs were drawn from 10,000 iterations. Vertical lines represent the trapping effort necessary to trigger stopping rule 1 (t*); the continuous line represents the 95% threshold, and the dashed line the 99% threshold

Fig. 4

Comparison of the cumulative trapping effort with zero captures in Valencia and the estimated thresholds for eradication under stopping rule 1. The continuous line represents the estimated trapping effort to achieve the 95% threshold and the dashed line is the 99% threshold (t*)

Discussion

Our results exemplify the suite of challenges faced in ensuring eradication of sliders, and we have provided methods for informing the planning and decision-making process. Our principal conclusion is that a very large trapping effort, without any single capture of sliders, is necessary to have high confidence about eradication success. This is due to the compounded effects of the low probabilities of capture using traps, the simulated spatial heterogeneity of the water bodies being monitored (stream vs. lake), and the high uncertainty in the posterior estimates of eradication success (in turn influenced by the prior beliefs about the presence or absence of sliders). In Valencia, the observed trapping effort was four times less than the predicted confirmation trapping required to achieve stopping rule 1 (Fig. 4); evidencing how large the required confirmation trapping effort can be. Additionally, it is important to note that our research applies to the moment when it is necessary to confirm the eradication of the species, in turn implying a precedent decision to proceed with eradication. However, if the site is re-invaded during the post-removal confirmation phase, either due to new releases or colonisation from adjoining sites, environmental managers would need to assess whether to start the eradication program from the beginning again or stop the eradication attempts. It is fundamental that decision-makers and environmental managers running eradication campaigns are well aware of these challenges. Having identified the key steps that contribute to the estimated large confirmation monitoring efforts, we can single them out to examine alternatives and potential improvements.

We found that basking traps were slightly more efficient than net traps at capturing sliders, something previously underscored by other authors (Pérez-Santigosa et al. 2006b; Valdeón et al. 2010). Nonetheless, trapping appears to be a relatively inefficient method for confirming the eradication of sliders, given the amount of trapping needed to obtain high probabilities of eradication confirmation. Therefore, other methods should be explored to inform eradication confirmation efforts for sliders. This includes surveying for live basking turtles using camera traps and direct observations, and the potential use of environmental DNA techniques (Lebboroni and Cecchini 2005; Pérez-Santigosa et al. 2006a; Bluett and Cosentino 2013; Davy et al. 2015). However, it is important to realise that confirmation monitoring happens when traps fail to capture new sliders and switching to an alternative monitoring method may not always be worthwhile. Moreover, basking traps are frequently used to provide a platform for sliders to bask and facilitate direct observations (Pérez-Santigosa et al. 2006a). We suggest the use of both trapping and direct observation monitoring from the beginning of the eradication program. When the stage of confirming eradication is reached, the information compiled during eradication efforts could be employed for deciding the best monitoring method available [i.e., adaptive monitoring: Lindenmayer and Likens (2009); Will et al. (2015)]. In line with the principles of adaptive management, acquiring information on the monitoring intensity deployed in other eradication programs could serve to construct informative prior distributions of the probability of eradication, P(E) in (9), that can be used to estimate subsequent monitoring efforts needed to ensure eradication.

Our scenario simulation approach revealed the importance of incorporating spatially-explicit information into the decision-making process for confirming eradication success, echoing previous calls for a better integration of environmental heterogeneity into management practices (Possingham et al. 2005; Anderson et al. 2013; Ward et al. 2016). For example, our results showed that stopping rules for net traps are rather different for the stream and lake scenarios. Spatial heterogeneity in eradication campaigns may arise from several non-exclusive processes, including active habitat selection by the target species and the accessibility of the trapping sites. These processes are likely to be persistent in all eradication campaigns. A crucial first step is to recognise the pervasive effects of spatial heterogeneity (Possingham et al. 2005), and we have provided techniques and examples on how to explicitly account for this heterogeneity.

The large trapping efforts reported here imply great investments, warranting the future evaluation of stopping rule 2 that will provide optimal solutions for confirmation monitoring when adequate estimates of the costs of monitoring and impacts are available (Regan et al. 2006; Rout et al. 2014). The costs of mistakenly declaring eradication would be difficult to estimate for emerging alien species such as sliders, given that their environmental impacts may be difficult to monetise (McIntosh et al. 2009; Kraus 2015). We advise managers to make a comprehensive assessment of the potential impacts associated with the presence of sliders; if this can be reasonably quantified, then we suggest using stopping rule 2. Alternately, if the impacts posed by sliders are largely unknown and eradication is highly desirable, then we suggest eradication success be confirmed by employing stopping rule 1.

Sliders are the most widespread alien turtle in the world (Ficetola et al. 2012; García-Díaz et al. 2015), and our work provides methods and baselines to support environmental managers designing and conducting eradication programs. Reptiles are a group of emerging alien vertebrates worldwide (Kraus 2009, 2015) and we expect the number of eradication attempts to increase concomitantly. The success of eradication programs will depend on being able to make informed decisions, and thus we want to emphasise the fundamental role and current need for developing and testing evidence-based procedures to support the eradication efforts.

Notes

Acknowledgements

This research was conducted thanks to the Department of Economic Development, Jobs, Transport and Resources Invasive Plants and Animals Research Project When to stop: Defining rules for surveillance of red-eared slider turtles (Victoria State Government, Australia). We want to thank C. Ayres, J.V. Ross, A. Rodríguez-Pereira, P. Cassey, N. Ainsworth, two anonymous reviewers, and the editor for their comments, support and help. The slider trapping campaigns were possible thanks to the funding, help and support provided by many institutions and people. Full acknowledgments are available in Online Resource 1.

Supplementary material

10530_2017_1480_MOESM1_ESM.docx (21 kb)
Supplementary material 1 (DOCX 21 kb)
10530_2017_1480_MOESM2_ESM.docx (2.4 mb)
Supplementary material 2 (DOCX 2414 kb)

References

  1. Anderson D, Ramsey D, Nugent G (2013) A novel approach to assess the probability of disease eradication from a wild-animal reservoir host. Epidemiol Infect 141:1509–1521CrossRefPubMedGoogle Scholar
  2. Bluett RD, Cosentino BJ (2013) Estimating occupancy of Trachemys scripta and Chrysemys picta with time-lapse cameras and basking rafts: a pilot study in Illinois, USA. Illinois State Academy of Science. Transactions 106:15–21Google Scholar
  3. Burgman MA, McCarthy MA, Robinson A et al (2013) Improving decisions for invasive species management: reformulation and extensions of the Panetta-Lawes eradication graph. Divers Distrib 19:603–607CrossRefGoogle Scholar
  4. Davy CM, Kidd AG, Wilson CC (2015) Development and validation of environmental DNA (eDNA) markers for detection of freshwater turtles. PLoS ONE 10:e0130965CrossRefPubMedPubMedCentralGoogle Scholar
  5. Epanchin-Niell RS, Haight RG, Berec L et al (2012) Optimal surveillance and eradication of invasive species in heterogeneous landscapes. Ecol Lett 15:803–812CrossRefPubMedGoogle Scholar
  6. Ficetola F, Rodder D, Padoa-Schioppa E (2012) Trachemys scripta (slider terrapin). In: Francis R (ed) A handbook of global freshwater invasive species. Earthscan, Oxon, pp 331–339Google Scholar
  7. Franch M, Llorente GA, Montori A (2007) Primeros datos sobre la biología de Trachemys scripta elegans en sintopía con Mauremys leprosa en el Delta del Llobregat (NE Ibérico). In: GEIB Grupo Especialista en Invasiones Biológicas (ed) Invasiones Biológicas: un factor de cambio global. GEIB Grupo Especialista en Invasiones Biológicas, León, Spain, pp 85–101Google Scholar
  8. García-Díaz P, Ross JV, Ayres C et al (2015) Understanding the biological invasion risk posed by the global wildlife trade: propagule pressure drives the introduction and establishment of Nearctic turtles. Glob Change Biol 21:1078–1091CrossRefGoogle Scholar
  9. García-Díaz P, Ross JV, Woolnough AP et al (2016) The illegal wildlife trade is a likely source of alien species. Conserv Lett. doi:10.1111/conl.12301
  10. Gelman A, Meng X-L, Stern H (1996) Posterior predictive assessment of model fitness via realized discrepancies. Stat Sin 6:733–760Google Scholar
  11. Groves CR, Game ET (2016) Conservation planning: informed decisions for a healthier planet. Roberts, Greenwood VillageGoogle Scholar
  12. Guillera-Arroita G, Hauser CE, McCarthy MA (2014) Optimal surveillance strategy for invasive species management when surveys stop after detection. Ecol Evol 4:1751–1760CrossRefPubMedPubMedCentralGoogle Scholar
  13. Héritier L, Valdeón A, Sadaoui A et al (2017) Introduction and invasion of the red-eared slider and its parasites in freshwater ecosystems of southern Europe: risk assessment for the European pond turtle in wild environments. Biodivers Conserv 1–27. doi:10.1007/s10531-017-1331-y
  14. Hoffmann BD, Luque GM, Bellard C et al (2016) Improving invasive ant eradication as a conservation tool: a review. Biol Conserv 198:37–49CrossRefGoogle Scholar
  15. Iglesias R, García-Estévez JM, Ayres C et al (2015) First reported outbreak of severe spirorchiidiasis in Emys orbicularis, probably resulting from a parasite spillover event. Dis Aquat Organ 113:75–80CrossRefPubMedGoogle Scholar
  16. Jones HP, Holmes ND, Butchart SHM et al (2016) Invasive mammal eradication on islands results in substantial conservation gains. PNAS 113:4033–4038CrossRefPubMedPubMedCentralGoogle Scholar
  17. Kraus F (2009) Alien reptiles and amphibians: a scientific compendium and analysis. Springer, DordrechtCrossRefGoogle Scholar
  18. Kraus F (2015) Impacts from invasive reptiles and amphibians. Annu Rev Ecol Evol Syst 46:75–97CrossRefGoogle Scholar
  19. Lasceve M (2014) Premiers résultats de l’opération de limitation de la population de tortues de Floride sur le site des Vieux Salins, Hyères (Var, France) Scientific Reports of Port-Cros National Park, vol 28, pp 195–201Google Scholar
  20. Lebboroni M, Cecchini A (2005) Basking counts as abundance indices in pond populations of Emys orbicularis. Herpetol J 15:121–124Google Scholar
  21. Lindenmayer DB, Likens GE (2009) Adaptive monitoring: a new paradigm for long-term research and monitoring. Trends Ecol Evol 24:482–486CrossRefPubMedGoogle Scholar
  22. McIntosh CR, Finnoff DC, Settle C et al (2009) Economic valuation and invasive species. In: Keller RP, Lodge DM, Lewis MA, Shogren JF (eds) Bioeconomics of invasive species. Integrating ecology, economics, policy, and management. Oxford University Press, New York, pp 151–179Google Scholar
  23. O’Keeffe S (2009) The practicalities of eradicating red-eared slider turtles (Trachemys scripta elegans). Aliens Invas Species Bull 28:19–25Google Scholar
  24. Pearson SH, Avery HW, Spotila JR (2015) Juvenile invasive red-eared slider turtles negatively impact the growth of native turtles: implications for global freshwater turtle populations. Biol Conserv 186:115–121CrossRefGoogle Scholar
  25. Pérez-Santigosa N, Díaz-Paniagua C, Hidalgo-Vila J et al (2006a) Características de dos poblaciones reproductoras del galápago de Florida, Trachemys scripta elegans, en el suroeste de España. Rev Esp Herpetol 20:5–16Google Scholar
  26. Pérez-Santigosa NP, Paniagua CD, Vila JH et al (2006b) Trampas y plataformas de asoleamiento: la mejor combinación para erradicar galápagos exóticos. Bol Asoc Herpetol Esp 17:115–120Google Scholar
  27. Plummer M (2003) JAGS: a program for analysis of Bayesian graphical models using Gibbs sampling. In: Proceedings of the 3rd international workshop on distributed statistical computing (DSC 2003), Vienna, Austria, pp 20–22Google Scholar
  28. Possingham HP, Franklin J, Wilson K et al (2005) The roles of spatial heterogeneity and ecological processes in conservation planning. In: Lovett GM, Turner MG, Jones CG, Weathers KC (eds) Ecosystem function in heterogeneous landscapes. Springer, New York, pp 389–406CrossRefGoogle Scholar
  29. R Development Core Team (2015) R: A language and environment for statistical computing. R Foundation for Statistical Computing, ViennaGoogle Scholar
  30. Ramsey DS, Parkes J, Morrison SA (2009) Quantifying eradication success: the removal of feral pigs from Santa Cruz Island, California. Conserv Biol 23:449–459CrossRefPubMedGoogle Scholar
  31. Ramsey DS, Parkes JP, Will D et al (2011) Quantifying the success of feral cat eradication, San Nicolas Island, California. New Zeal J Ecol 35:163–173Google Scholar
  32. Regan HM, Ben-Haim Y, Langford B et al (2005) Robust decision-making under severe uncertainty for conservation management. Ecol Appl 15:1471–1477CrossRefGoogle Scholar
  33. Regan TJ, McCarthy MA, Baxter PWJ et al (2006) Optimal eradication: when to stop looking for an invasive plant. Ecol Lett 9:759–766CrossRefPubMedGoogle Scholar
  34. Rout TM, Kirkwood R, Sutherland DR et al (2014) When to declare successful eradication of an invasive predator? Anim Conserv 17:125–132CrossRefGoogle Scholar
  35. Russell JC, Binnie HR, Oh J et al (2016) Optimizing confirmation of invasive species eradication with rapid eradication assessment. J Appl Ecol 54:160–169CrossRefGoogle Scholar
  36. Sancho V, Lacomba JI (2016) Expansion of Trachemys scripta in the Valencian Community (Eastern Spain). Proc Int Symp Fresh Turtles Conserv 1:41–49Google Scholar
  37. Seebens H, Blackburn TM, Dyer EE et al (2017) No saturation in the accumulation of alien species worldwide. Nat Commun 8:14435CrossRefPubMedPubMedCentralGoogle Scholar
  38. Spiegelhalter DJ, Best NG, Carlin BP et al (2002) Bayesian measures of model complexity and fit. J R Stat Soc Ser B (Statistical Methodology) 64:583–639CrossRefGoogle Scholar
  39. Tobin P, Kean J, Suckling D et al (2014) Determinants of successful arthropod eradication programs. Biol Invas 16:401–414CrossRefGoogle Scholar
  40. Valdeón A, Crespo-Diaz A, Egaña-Callejo A et al (2010) Update of the pond slider Trachemys scripta (Schoepff, 1792) records in Navarre (Northern Spain), and presentation of the Aranzadi Turtle Trap for its population control. Aquat Invas 5:297–302CrossRefGoogle Scholar
  41. Ward DF, Anderson DP, Barron MC (2016) Using spatially explicit surveillance models to provide confidence in the eradication of an invasive ant. Sci Rep 6:34953CrossRefPubMedPubMedCentralGoogle Scholar
  42. Will DJ, Campbell KJ, Holmes ND (2015) Using digital data collection tools to improve overall cost-efficiency and provide timely analysis for decision making during invasive species eradication campaigns. Wildl Res 41:499–509CrossRefGoogle Scholar

Copyright information

© Springer International Publishing AG 2017

Authors and Affiliations

  • Pablo García-Díaz
    • 1
    • 2
    • 3
  • David S. L. Ramsey
    • 4
  • Andrew P. Woolnough
    • 5
  • Marc Franch
    • 6
    • 7
  • Gustavo A. Llorente
    • 6
  • Albert Montori
    • 6
  • Xabier Buenetxea
    • 8
  • Asier R. Larrinaga
    • 9
  • Matthieu Lasceve
    • 10
  • Alberto Álvarez
    • 11
  • José María Traverso
    • 11
  • Aitor Valdeón
    • 12
    • 13
  • Ariñe Crespo
    • 13
  • Virginia Rada
    • 13
  • Enrique Ayllón
    • 11
  • Vicente Sancho
    • 14
  • J. Ignacio Lacomba
    • 15
  • José Vicente Bataller
    • 16
  • Miguel Lizana
    • 2
  1. 1.School of Biological Sciences and Centre for Conservation Science and Technology (CCoST)The University of AdelaideNorth TerraceAustralia
  2. 2.Department of Animal BiologyUniversity of SalamancaSalamancaSpain
  3. 3.Landcare ResearchLincolnNew Zealand
  4. 4.Arthur Rylah Institute, Department of Land, Water and EnvironmentHeidelbergAustralia
  5. 5.Department of Economic Development, Jobs, Transport and ResourcesAttwoodAustralia
  6. 6.Section of Zoology and Anthropology, Department of Evolutionary Biology, Ecology and Environmental Sciences, Faculty of BiologyUniversity of BarcelonaBarcelonaSpain
  7. 7.CICGE - Centro de Investigação em Ciências Geo-Espaciais Observatório Astronómico Prof. Manuel de Barros Alameda do Monte da VirgemVila Nova de GaiaPortugal
  8. 8.BOLUE Ingurumen IkerketakGamiz-FikaSpain
  9. 9.eNeBaDaAríns - Santiago de CompostelaSpain
  10. 10.Toulon Provence Méditerranée, Communauté d’AgglomérationToulonFrance
  11. 11.Asociación Herpetológica EspañolaLeganésSpain
  12. 12.Departamento de Geografía y Ordenación del Territorio, Instituto de Investigación en Ciencias Ambientales (IUCA)Universidad de ZaragozaZaragozaSpain
  13. 13.Department of HerpetologyAranzadi Society of SciencesDonostia-San SebastiánSpain
  14. 14.LIFE+Trachemys Project, Wildlife ServiceRegional Ministry of Environment, Generalitat ValencianaValenciaSpain
  15. 15.Direcció General de Medi Natural, Generalitat ValencianaValenciaSpain
  16. 16.Freshwater Species Conservation Centre, Wildlife Service, Regional Ministry of EnvironmentVAERSA-Generalitat ValencianaEl SalerSpain

Personalised recommendations