Range expansion of the alien red-eared slider Trachemys scripta (Thunberg in Schoepff, 1792) (Reptilia, Testudines) in Eastern Europe, with special reference to Latvia and Ukraine

An increasing number of thermophilic invasive species are spreading and becoming naturalized in Eastern Europe, at least partially due to recent climate change. This can be exemplified by current expansion of the red-eared slider, Trachemys scripta , in Latvia and Ukraine. We collected 44 records of the species in Latvia and 79 in Ukraine. Two of the three subspecies have been found – T. s. elegans and T


Introduction
Climate change in recent years has allowed more thermophilic alien species to naturalize and actively invade Eastern Europe (Pupins and Pupina 2011;Nekrasova 2013;Cerasoli et al. 2019;Kuybida et al. 2019;Nekrasova et al. 2019aNekrasova et al. , b, c, 2021a, b;, b;Marushchak et al. 2021).One of the reasons for the emergence of new alien species is the trade in exotic animals and their uncontrolled release into nature.Freshwater turtles are among the most traded reptiles.Until the 1990's, the trade of freshwater turtles focused mostly on the American red-eared slider Trachemys scripta (Thunberg in Schoepff, 1792).In the period 1989-1997, more than 52 million red-eared sliders were exported from the USA (Telecky 2001).Today red-eared sliders are one of only two reptiles included in the list of the "World's Worst Invasive Alien Species" (Global Invasive Species Database 2021) and considered one of the 100 worst invasive alien species in Europe (Scalera 2009).In Ukraine, several alien turtle species have been discovered, T. scripta, Testudo horsfieldii, Mauremys rivulata, M. caspica (Nekrasova 2013;Kukushkin et al. 2017) and, perhaps Testudo graeca (Nekrasova and Tytar 2012).In Latvia T. scripta, M. rivulata, M. caspica, Pelodiscus sinensis, and T. horsfieldii (Pupins and Pupina 2011) were recorded.
Among the noted species T. scripta represents the greatest threat for local biodiversity because this species can transmit diseases to which native turtles are susceptible, occupy similar ecological niches of native turtles, competing for their food resources, displacing them from their favored basking sites, participating in their breeding-courtship attempts between the autochthonous turtles, and overall reducing their survival (Cadi and Joly 2004;Pupins 2007;Semenov 2009;Cerasoli et al. 2019;Espindola et al. 2019;Pupins et al. 2019;Nekrasova et al. 2021b).According to the Handbook of global freshwater invasive species (Ficetola et al. 2012), pond sliders are recorded in Finland, Latvia and Lithuania (but not in Estonia), Poland, Slovakia, Hungary, Romania, Bulgaria, Russia and other countries (Cadi and Joly 2004;Semenov 2009;Pupins and Pupina 2011;Kukushkin et al. 2017;Kornilev et al. 2020).Numerous findings of exotic turtles continued to appear at the beginning of the 21 st century in the regions of Ukraine -Odesa, Crimea, Transcarpathia areas next to the border with Hungary and Romania (Nekrasova 2013;Кurtyak and Kurtyak 2013;Kukushkin et al. 2017).
Recently, the red-eared slider has been found in a number of regions in Ukraine due to breeding in captivity and release into various wetlands (mostly urban; Nekrasova et al. 2021b).Despite the availability of some distribution maps at different scales (Rödder et al. 2009;Banha et al. 2017) there are no distribution maps concerning the species in Eastern Europe, particularly Latvia and Ukraine.Yet, for biodiversity conservation and species management purposes, it would be important to predict how far the red-eared slider can advance.The availability of spatially explicit maps of risk of establishment may allow to set up specific preventive measures in different regions, like trade regulation or appropriate communication campaigns (Masin et al. 2014).In this study we attempted to summarize findings of the red-eared slider in outdoor settings both in Latvia and Ukraine and, by using a species distribution modeling (SDM) approach, to identify areas where the species could survive under current climatic conditions.
To explore the potential distribution of T. scripta in our study area we employed Bayesian Additive Regression Trees (BART), a machine learning technique consisting of a Bayesian approach to Classification and Regression Trees (CART), capable of producing highly accurate predictions without overfitting to noise or to particular cases in the data.Models of this method estimate the probability of a given output variable (a binary classification of habitat suitability or species presence) based on decision "trees" that split predictor variables with nested, binary rule-sets (Carlson 2020).Running SDMs with BARTs has recently been greatly facilitated by the development of an R package, "embarcadero".The algorithm computes habitat suitability values ranging from 0, for fully nonsuitable habitat, to 1, for fully suitable habitat.Model performance was assessed using measures of accuracy: the area under the receiver-operator curve (AUC) (Fielding and Bell 1997), and the true skills statistic (TSS) (Allouche et al. 2006).
As input, SDMs require georeferenced biodiversity observations and geographic layers of environmental information.Because the use of restricted data (similar to not capturing the full species' environmental range) reduces strongly the combinations of environmental conditions under which the models are calibrated and reduces the applicability of the model for predictive purposes (Pearson and Dawson 2003;Thuiller et al. 2004), we used the full set of European invasive records from the GBIF data base (Trachemys scripta, GBIF 2021), updated by our personal field investigations.These occurrence points (n = 1,147) varied in spatial density due to variable sampling intensity.As a result, and to avoid overemphasizing heavily on sampled areas, the BART algorithm selects points for model calibration using subsampling to reduce sampling bias and spatial autocorrelation, which would produce models of lower rather than higher quality (Beck et al. 2013).The niche for the species (not discriminating for subspecies) was described based on climate.We transformed 30 climate data from the CliMond global dataset (Kriticos et al. 2014, see Table S3).This dataset consists of climatic variables in raster format covering temperature, rainfall, solar radiation, and seasonality conditions that reportedly influence reptile ranges (Martínez et al. 2020); following Araújo et al. (2006), a 10′ grid resolution was chosen.Because collinearity among environmental predictors will increase uncertainty in model parameters and decrease statistical power we used principal components analysis (PCA) in SAGA GIS (Conrad et al. 2015) to reduce collinearity (Petitpierre et al. 2017), giving a new, simpler environmental space defined in fewer, and fully orthogonal axes.Significance of the components was identified using the broken stick method (Jackson 1993).PCA reduced collinearity amongst the 30 variables from the CliMond dataset (Table S3); the first four components were identified as significant, which together accounted for almost 90% of the variation (according to standard techniques https://www.climond.org/BioclimData.aspx,Kriticos et al. 2014, Tables S4-S5).To differentiate between suitable and non-suitable environments in terms of invasion risk two thresholds were used: the minimum training presence (Pearson et al. 2007) and the more conservative one percentile threshold.These are the thresholds at which all or all but 1% of the training presences are required to be included within projected suitable environments, respectively.Contour lines depicting the threshold values separating suitable from unsuitable areas for the species were produced in SAGA GIS.Maps of habitat suitability in the GeoTIFF format were processed and visualized in SAGA GIS.

Results and discussion
We collected 123 findings of T. scripta in Latvia and Ukraine (Figure 1; Tables S1-S2).Turtles were found mostly in stagnant water bodies.In Latvia, pond sliders were first found in 1999 (Table S1) and in 2006, when six adult animals were recorded in Nitaure, a village in Amata municipality, where they had successfully over-wintered.In addition to the records of Pupins (2007) and Pupins and Pupina (2011), 11 new findings of T. s. scripta (Schoepff, 1792) and 33 of T. s. elegans (Wied-Neuwied, 1839) were added to our database (Figure 1; Table S1).
In Ukraine, the turtle was found in drainage waters of Kyiv (Ukraine) in the late 1990s, namely at the Bortnychi sewage water treatment plant (50.3837N; 30.6642E),where juveniles and adults were recorded (Table S2).Interestingly, at any time of the year various exotic species of fish and plants could also be found in the warm waters (about +19-+20 °С) of this sewage water treatment plant, for instance guppy (Poecilia reticulata Peters, 1859) (Nekrasova et al. 2021a, b).
Two of the three subspecies of T. scripta were found in Latvia and Ukraine -T.s. elegans and T. s. scripta.The more common one was T. s. elegans -75% of the records in Latvia and 97.5% of the records in Ukraine.It is possible that the T. s. troostii subspecies also occurs in both study areas, but it is rather difficult to identify it.
In Ukraine the red-eared slider is often kept in captivity and is a fairly popular pet.This may be evidenced by a simple Google Search query using the key words "buy" and "red-eared slider" (in Ukrainian) that returned 1,350 results (4 th of August 2021).Unfortunately, turtles are commonly released into the wild to nearby ponds and lakes, thus invading urban ecosystems.In the wintertime, turtles in these habitats usually don't survive and this becomes noticeable in the spring.For instance, dead turtles were found on the 21 st of March 2013 in ponds of Myrhorod (Poltava Region) and on the 13 th of April 2015 in fish ponds of Didorivka (a suburb of Kyiv).
Occasional records have been made of egg-laying and courtship behavior.In Latvia (Daugavpils) an individual was recorded laying eggs after successfully over-wintering in a fenced outdoor pool (Figure 2A); these eggs were fertilized but the turtles did not hatch.Similarly, a female T. scripta was observed laying eggs on the 28th of June 2017 in Mariupol (Donetsk Region) (personal communication -O.Lazarenko, Figure 2B; Table S2), but the fate of the clutch is unknown.Courtship behavior in Ukraine has been recorded in ponds of the Athletic Park in Odesa on the 21st of June 2020 (Figure 2С).
In urban ecosystems T. scripta was found alongside with a native species, Emys orbicularis (Linnaeus, 1758).In these ecosystems in southern Ukraine T. scripta is numerically dominant over E. orbicularis.For instance, of the 80 turtles found on the 8 th of May 2020 in the Athletic Park 94% were T. scripta.Within this system of lakes in the park, koi fish (a domesticated variety of the common carp) are also found.These exotic fish, together with red-eared slider, are assumed to have been successfully wintering here for at least 10 years.In Europe, strong competition occurs between T. scripta and E. orbicularis, since they occupy similar habitats and ecological niches (Cadi and Joly 2004).For Europe habitat suitability maps showed a strong correlation (Pearson's r = 0.68) between the species, meaning high chances of co-occurrence and the potential for interspecific competition (Pupins et al. 2019).
These PCA-derived variables were used for the modelling instead of the original set of environmental information (Tables S3-S5).In terms of discrimination accuracy the BART model showed acceptable performance: AUC = 0.825 and TSS = 0.513 (Swets 1988;Jiménez-Valverde et al. 2011).
Analyzing the models, we came to the conclusion that the minimum training presence logistic threshold, which provides minimum requirements for the species climate preferences, indicated all sites in Latvia and Ukraine as suitable in terms of the bioclimatic niche for the red-eared slider, although on the average habitat suitability (HS) conditions for the species are higher in Ukraine (HS 0.219 against HS 0.106 in Latvia), meaning the risk of invasion is potentially higher in Ukraine.Employing the more conservative one percentile threshold shows that in Latvia the areas more suitable for the species are in the south and south-west of the country, whereas in Ukraine these are located primarily in the west and south.Moreover, this species successfully winters in the south of Ukraine.Therefore, specific preventive measures should be planned and undertaken here.These include both legal measures to control pet trade and campaigns to explain the problems caused by imported turtles and to encourage people to change their attitudes towards nature and support biodiversity conservation (Ficetola et al. 2012;Masin et al. 2014).

Figure 1 .
Figure 1.Distribution of records of the red-eared slider in Latvia (LV green dots) and Ukraine (UA red dots); numbers correspond to information in the Tables S1-S2 of the Supplementary material.

Figure 3 .
Figure 3. Predicted habitat suitability for the red-eared slider.Warmer colours indicate higher bioclimatic suitability for the species.The contour line in color fuchsia depicts the one percentile threshold: A -Latvia; B -Ukraine.Areas of the highest habitat suitability (> 0.3-0.5)are colored in red and areas of the lowest (< 0.2) -in blue (SagaGis).