Introduction
The Atrato slider Trachemys medemi is a freshwater turtle endemic to the lower basin of the Atrato River in north-west Colombia (Vargas-Ramírez et al., Reference Vargas-Ramírez, Del Valle, Ceballos and Fritz2017). Formerly identified as Trachemys venusta uhrigi (McCord et al., Reference McCord, Joseph-Ouni, Hagen and Blanck2010), a subspecies of the Meso-American slider, it was recognized as a new species in 2017 (Vargas-Ramírez et al., Reference Vargas-Ramírez, Del Valle, Ceballos and Fritz2017). This reclassification led to uncertainties regarding its biology, ecology and distribution (Medem, Reference Medem1962; Pritchard & Trebbau, Reference Pritchard and Trebbau1984; MAVDT, 2009; Bock et al., Reference Bock, Páez and Castaño-Mora2012; Ceballos & Brand, Reference Ceballos and Brand2014). It is therefore important to determine the distribution of the species and improve our understanding of its evolutionary history, dispersal patterns, population dynamics and physiological requirements, to inform conservation planning (Mota-Vargas & Rojas-Soto, Reference Mota-Vargas and Rojas-Soto2012).
The probable distribution of T. medemi in Colombia has changed over time. In 2009 the Colombian Ministry of Environment, Housing and Territorial Development (Ministerio de Ambiente, Vivienda y Desarrollo Territorial, MAVDT) suggested that its range extended from Punta Arboletes (Antioquia department) in the north to the Quito River (Chocó department) in the south, and from the Mulatos River in the east to the Panama border in the west. Vargas-Ramírez et al. (Reference Vargas-Ramírez, Del Valle, Ceballos and Fritz2017) modelled the distribution of T. venusta from literature reports (Williams, Reference Williams1956; Pritchard & Trebbau, Reference Pritchard and Trebbau1984; Castaño-Mora, Reference Castaño-Mora1992; MAVDT, 2009; Bock et al., Reference Bock, Páez and Castaño-Mora2012) and suggested a range for T. medemi that covered the lower basin of the Atrato River in the municipalities of Acandí, Unguía, and Ríosucio in Chocó, and Chigorodó and Turbo in Antioquia. However, there was doubt about the species’ presence at some sites including Punta Arboletes (MAVDT, 2009), Necoclí (Pritchard & Trebbau, Reference Pritchard and Trebbau1984; Turtle Taxonomy Working Group, Reference Rhodin, Iverson, Bour, Fritz, Georges, Shaffer and van Dijk2021) and the Quito River (upper Atrato, Chocó; MAVDT, 2009), and other sites were excluded altogether such as the municipality of Mutatá (Bock et al., Reference Bock, Páez and Castaño-Mora2012). Through examination of the grey literature and interviews with local researchers, we obtained photographic evidence of T. medemi from new areas including the Salado Swamp in Necoclí (M. Duarte, pers. comm., 2019), Vigía del Fuerte and Murindó (E. Ortiz, pers. comm., 2019). These findings underlined the need to update the geographical range of the species by verifying its presence in the new areas and assessing presence in previously occupied areas.
The known threats to T. medemi include hunting, habitat degradation and habitat fragmentation, and the species is provisionally categorized as Vulnerable by the IUCN Tortoise and Freshwater Turtle Specialist Group (Turtle Taxonomy Working Group, Reference Rhodin, Iverson, Bour, Fritz, Georges, Shaffer and van Dijk2021). Reports indicate extensive harvesting of wild individuals in Carepa, Turbo and Necoclí municipalities, particularly during the Holy Week when many people only eat white meat by tradition (Bock et al., Reference Bock, Páez and Castaño-Mora2012; Ceballos & Brand, Reference Ceballos and Brand2014; Corpourabá, 2019; this study). Communities in Urabá subsist on agriculture and mining, resulting in contamination of soils and water with heavy metals and agrochemicals, and unsustainable use of forest resources (Bock et al., Reference Bock, Páez and Castaño-Mora2012; Ceballos & Brand, Reference Ceballos and Brand2014; Vallejo-Toro et al., Reference Vallejo-Toro, Vásquez, Correa, Bernal, Alcántara-Carrió and Palacio2016). This aquatic and terrestrial habitat degradation and fragmentation is likely to have a negative impact on T. medemi by reducing feeding areas and possibly disrupting reproductive cycles, potentially limiting the species’ dispersal capacity (Bodie, Reference Bodie2001; Bowne et al., Reference Bowne, Bowers and Hines2006).
Climate change poses additional threats (Steffen et al., Reference Steffen, Richardson, Rockström, Cornell, Fetzer and Bennett2015). In Colombia the temperature has increased by up to 0.3 °C per decade from 1971 to 2000, with projected increases of 1.4 ± SD 0.4 °C between 2011 and 2040 in the country’s Caribbean hydrographic area (Ruíz-Murcia, Reference Ruíz-Murcia2010; Ruíz-Murcia et al., Reference Ruíz-Murcia, Gutiérrez, Dorado, Mendoza, Martínez and Rojas2015). Although there is no evidence of a direct effect on T. medemi, impacts have been proven on other species of aquatic turtles, including climate-induced metabolic changes (Ligon & Peterson, Reference Ligon and Peterson2002; Litzgus & Hopkins, Reference Litzgus and Hopkins2003; Stephens & Wiens, Reference Stephens and Wiens2009). These can affect growth rates and reproductive patterns, including sex ratios in species with temperature-dependent determination of sex such as slider turtles Trachemys spp. (Neuwald & Valenzuela, Reference Neuwald and Valenzuela2011; Ihlow et al., Reference Ihlow, Dambach, Engler, Flecks, Hartmann and Nekum2012).
This study had two main objectives. Firstly, we aimed to update the potential geographical range of T. medemi using ecological niche modelling to correlate environmental variables with species presence, and species distribution modelling to identify environmental requirements (Peterson & Soberón, Reference Peterson and Soberón2012; Baker et al., Reference Baker, Maclean, Goodall and Gaston2022). Secondly, we set out to estimate the potential effects of climate change on the species’ distribution and to identify future trends in its range. This information is needed to update the conservation status of T. medemi (IUCN Standards and Petitions Committee, Reference Standards and Committee2022; Páez et al., Reference Páez, Bock, Alzate-Estrada, Barrientos-Munoz, Cartagena-Otalvaro, Echeverry-Alcendra and Vallejo-Betancur2022) and to support effective conservation action to protect the species.
Study area
The Urabá region in Colombia is a tropical humid forest with swamps, canals and streams connected to the Atrato River basin. The mean annual rainfall is 2,000 mm, with two wet season peaks (April–June, August–October) and a dry season (December–March). The mean altitude of the area is 28 m and mean monthly air temperature ranges from 20 to 35 °C throughout the year (Pabón et al., Reference Pabón, Zea, León, Hurtado, González, Montealegre and Leyva2001).
Methods
We used ecological niche modelling to investigate the potential aquatic distribution of T. medemi based on data collected in the field and from published records. We then used the modelling results to predict future potential distributions under two different climate change scenarios.
Occurrence records
During January 2022–February 2023, we conducted eight field trips to the Atrato River and its tributaries to search for evidence of T. medemi (Fig. 1). Specifically, in the north of the study area we visited the Suriquí and León Rivers (Turbo, Antioquia) and the Salado Swamp (Necoclí, Antioquia); in the east we visited the Chontaduralito River (Mutatá, Antioquia); and in the south the Napipí Swamp (Bojayá, Chocó), the Bajirá River (Chocó) and Vigía del Fuerte in the middle Atrato River (Antioquia). In addition, we visited the Ciénaga 37 Swamp in the San Juan River (Arboletes, Antioquia), and the Tierralta Swamp in the Sinú River (Córdoba). We did not visit the upper Atrato region (Chocó) because of security problems associated with the ongoing armed conflict. Specific search sites were selected based on previous anecdotal reports by local people, habitat suitability, accessibility and safety conditions. Turtles were trapped by two methods used simultaneously: muddling and hoop net traps. In the muddling method, two local guides waded into shallow water and systematically searched the underwater vegetation, catching turtles by hand (Bury et al., Reference Bury, Ashton, Germano, Karraker, Reese, Schlick, Bury, Welsh, Germano and Ashton2012). We had a main guide with prior experience capturing this species, and a second guide was chosen at the specific searching site. Both guides, one of the authors and a field assistant worked as a team in flooded areas up to 2 m deep to catch the turtles. At each site, we surveyed accessible river sections covering 4–10 km (by water and foot) per day, over a period of 4–5 days. Searching was during daylight hours, until sunset. For the hoop net trapping method, we installed 8–10 traps per locality in slow-flowing sections of the river chosen at random, but generally where the water was deeper than 2 m. Traps were attached to nearby vegetation to prevent displacement, and baited with plantain, guava, chicken or tuna. They were checked every 24 h for 4–5 consecutive days, while we searched turtles with the muddling technique in the nearby area as described above. The total (in all eight field visits) capture effort of the muddling technique was 276 person-hours, and 179 trap-days of the hoop net trap technique. The capture success was defined as the number of turtles captured by hand divided by the total person-hours of active searching across all sites for the muddling technique. The capture success for the hoop net trap technique was defined as the number of turtles captured divided by the total trap-days across all sites. Once a turtle was captured, we identified it to species level following Vargas-Ramírez et al. (Reference Vargas-Ramírez, Del Valle, Ceballos and Fritz2017) based on physical appearance, notably size, carapace shape and colouration. Other species in the area included the Colombian wood turtle Rhinoclemmys melanosterna, the white-lipped mud turtle Kinosternon leucostomum and the South American snapping turtle Chelydra acutirostris. Each turtle was sexed, weighed, measured for linear morphometrics (length and width of the carapace and plastron) and marked with notches on the carapace (Cagle, Reference Cagle1939) to allow recognition upon recapture. Following measurement and processing, all turtles were released at their capture site, and the search continued along the same waterbody as access permitted. Additionally, five turtles were voluntarily provided by residents who had kept them in captivity. They agreed to the collection of morphometric data but declined to release the animals. The capture sites for these turtles were recorded based on the information provided and were validated based on local knowledge and field verification. Lastly, we examined field records (S. Cuadrado, pers. comm., 2023), the scientific literature and the Colombian Biodiversity Information Facility (SiB Colombia) for genetic evidence or phenotypic descriptions of the species (Williams, Reference Williams1956; Castaño-Mora, Reference Castaño-Mora1992; Bock et al., Reference Bock, Páez and Castaño-Mora2012; Vargas-Ramírez et al., Reference Vargas-Ramírez, Del Valle, Ceballos and Fritz2017; Borja-Acosta & Galeano, Reference Borja-Acosta and Galeano2023). We conducted literature searches using Google Scholar (Google, USA), Scopus (Elsevier, The Netherlands) and Web of Science (Clarivate, USA), as well as the University of Antioquia’s database, with the keywords Trachemys medemi, Trachemys venusta, Atrato River slider, distribution and Colombia. No date limits were applied.
Freshwater bodies in the Atrato River basin, north-west Colombia, showing (a) habitat suitability for the Atrato slider Trachemys medemi determined using ecological niche modelling (cloglog output from MaxEnt), and (b) potentially suitable aquatic conditions based on the minimum training presence threshold rule (see text). Blue lines indicate the water bodies where the species may be present. Red points indicate field records (this study and Cuadrado, pers. comm.) and black points are presence records from published literature (Williams, Reference Williams1956; Castaño-Mora, Reference Castaño-Mora1992; Bock et al., Reference Bock, Páez and Castaño-Mora2012; Vargas-Ramírez et al., Reference Vargas-Ramírez, Del Valle, Ceballos and Fritz2017; Borja-Acosta & Galeano, Reference Borja-Acosta and Galeano2023). White-filled points indicate sites where T. medemi was not detected, and a white-filled square marks a doubtful presence in the Sinú River. There were multiple records in close proximity in the middle Atrato basin (Napipí Swamp, Bojayá River, and Atrato River at Vigía del Fuerte); these are grouped together as location (4). Locations with confirmed presence in 2022–2023 are numbered: (1) Salado Swamp (Antioquia), (2) Suriquí River (Antioquia), (3) Bajirá River (Antioquia), (4) middle Atrato basin including Napipí Swamp, Bojayá River (Chocó) and Atrato River at Vigía del Fuerte (Antioquia), (5) Salaquí River (Chocó), (6) Truandó River (Chocó). (Readers of the printed journal are referred to the online article for a colour version of this figure.)
Ecological niche modelling
We characterized the ecological niche of T. medemi from 26 climate and hydrological variables (Supplementary Table 1; Nori & Rojas-Soto, Reference Nori and Rojas-Soto2019; Arango-Lozano et al., Reference Arango-Lozano, Patiño-Siro and Toro-Cardona2023). We obtained values of these variables at a spatial resolution of 0.0083° (c. 1 km2) from Worldclim 2.0 (15 climate variables; Fick & Hijmans, Reference Fick and Hijmans2017), EarthEnv (10 freshwater variables; Domisch et al., Reference Domisch, Amatulli and Jetz2015), and an evapotranspiration variable obtained from the CGIAR-CSI Global Aridity and PET Database (PET; Murphy et al., Reference Murphy, Tuberville, Maerz and Andrews2016; Trabucco & Zomer, Reference Trabucco and Zomer2019). We did not use four of the 19 variables available in WorldClim (Bios 8, 9, 18 and 19) because they exhibit spatial anomalies such as odd discontinuities between neighbouring pixels (Escobar et al., Reference Escobar, Lira-Noriega, Medina-Vogel and Peterson2014; Campbell et al., Reference Campbell, Luther, Moo-Llanes, Ramsey, Danis-Lozano and Peterson2015; Booth, Reference Booth2022).
Ecological niches can be estimated using correlative methods that link data on species occurrence to associated environmental characteristics. Some algorithms such as MaxEnt require the establishment of an environmental background over a calibration area (Barve et al., Reference Barve, Barve, Jiménez-Valverde, Lira-Noriega, Maher and Peterson2011). One approach to defining the calibration area involves identifying areas historically accessible to the species, considering opportunities and constraints on species movement such as dispersal areas and historical barriers (Barve et al., Reference Barve, Barve, Jiménez-Valverde, Lira-Noriega, Maher and Peterson2011). River basins represent biogeographical entities that limit the distribution of aquatic species. Thus, we estimated the potential geographical range of T. medemi by restricting its habitat to freshwater bodies (Nori & Rojas-Soto, Reference Nori and Rojas-Soto2019; Arango-Lozano et al., Reference Arango-Lozano, Patiño-Siro and Toro-Cardona2023). We selected river basins with at least one historical record of T. medemi, including the Atrato River, the San Juan River, which floods and connects to the Atrato riverbed during the rainy season, and the Sinú River, which is connected to a tributary of the Atrato River (Carepa River) by human-made channels. We also included the Tuira, Chucunaque and Cartí Rivers in Panama, as they are < 20 km from the Salaquí River and Acandí where T. medemi has been recorded (this study; Bock et al., Reference Bock, Páez and Castaño-Mora2012).
We cleaned the species occurrence data by removing duplicate records in each raster pixel (c. 1 km2; Cobos et al., Reference Cobos, Peterson, Barve and Osorio-Olvera2019). For those records that did not directly coincide with a freshwater body as a result of small differences in georeferences, we calculated the Euclidean distance to their surrounding pixels and moved them manually to the closest pixel within a waterbody using ArcGIS Pro 2.5 (Esri, 2021). Then, using the occurrence and environmental variables database, we assessed multicollinearity using the Spearman rank order correlation in R 4.3.0 (R Core Team, 2023) for variables with non-normal distributions, with a threshold of 0.7 to reduce overfitting (Dormann et al., Reference Dormann, Elith, Bacher, Buchmann, Carl and Carré2013). Subsequently, we evaluated the variance inflation factor (VIF) with a threshold of 7 using the usdm package in R (Naimi et al., Reference Naimi, Hamm, Groen, Skidmore and Toxopeus2014). In this way we generated five sets of variables (Supplementary Table 1). Sets 1 and 3 were based on the Spearman correlations of the presence points of all variables and the results of freshwater variables only. Sets 2 and 4 were based on the VIF correlation results of the presence points of all variables and the results of freshwater variables only. Finally, set 5 was derived from the Spearman correlation result using 10,000 random points within the calibration area.
We utilized the MaxEnt algorithm (Phillips et al., Reference Phillips, Anderson and Schapire2006) with the kuenm package in R (Cobos et al., Reference Cobos, Peterson, Barve and Osorio-Olvera2019) to model the potential distribution of T. medemi. This algorithm allows modelling with presence records only and evaluates multiple sets of variables (Elith et al., Reference Elith, Phillips, Dudík, Chee and Yates2011). We randomly selected 80% of the occurrences for training and 20% for model evaluation. The model was parameterized with linear, quadratic and product responses, excluding hinge and threshold, to avoid complex model responses (Elith et al., Reference Elith, Phillips, Dudík, Chee and Yates2011). We tested regularization multiplier values from 0.1 to 1 in intervals of 0.1, and from 2 to 8 in intervals of 1, with 100 replicates and 1,000 iterations. We then selected the best ecological niche model using kuenm metrics to evaluate candidate models on: (1) statistical significance, assessed through the partial area under the receiver operating curve ratio (AUCr) using independent random subsets of presence data (Peterson et al., Reference Peterson, Papes and Soberon2008); (2) predictive performance, considering omission rate (OR) derived from independent random subsets of presence data, and minimum training presence (MTP) thresholds adjusted to the calibration data (Anderson et al., Reference Anderson, Lew and Peterson2003); and (3) model fit and simplicity, evaluated using the Akaike information criterion adjusted for small sample size (AICc) as adapted for MaxEnt by Warren & Seifert (Reference Warren and Seifert2011). In kuenm, models are selected as optimal when statistically they are significantly better than random in their predictions of independent subsets of occurrence data (AUCr > 1, P ≤ 0.05), they present a low omission rate (false-negative rate ≤ MTP value), and are within 2 units of the minimum AICc value among significant and high-performing models (Cobos et al., Reference Cobos, Peterson, Barve and Osorio-Olvera2019). To enable evaluation, analysis and interpretation of distribution in terms of presence/absence, we transformed the cloglog output from MaxEnt into a binary (presence/absence) potential distribution map using the Spatial Analyst Reclassify tool in ArcGIS (Esri, 2021). We applied the MTP as a threshold rule because of our confidence in the veracity of the records, assuming that the least suitable habitat where the species was known to occur represents the minimum suitability value and ensuring that no training sample was excluded.
Effects of climate change on the potential distribution of T. medemi
We employed Worldclim 2.0 environmental variables (Fick & Hijmans, Reference Fick and Hijmans2017) to estimate the effects of climate change on the potential distribution of T. medemi and to explore trends under future climate scenarios. We obtained variables for the periods 2041–2060 and 2061–2080 (hereafter referred to as 2050 and 2070, respectively) under two standardized Shared Socioeconomic Pathway scenarios: SSP245 (intermediate scenario; assumes moderate efforts to mitigate emissions) and SSP585 (high-emissions scenario; assumes continued fossil fuel use and limited climate policy). We used the following three databases: the Model for Interdisciplinary Research on Climate version 6 (MIROC6; Shiogama et al., Reference Shiogama, Abe and Tatebe2019; Tatebe et al., Reference Tatebe, Ogura, Nitta, Komuro, Ogochi and Takemura2019), Meteorological Research Institute Earth System Model version 2.0 (MRI-ESM2.0; Yukimoto et al., Reference Yukimoto, Kawai, Koshiro, Oshima, Yoshida and Urakawa2019) and Australian Community Climate and Earth System Simulator, Coupled Model version 2 (ACCESS-CM2; Bi et al., Reference Bi, Dix, Marsland, O’Farrell, Sullivan and Bodman2020).
We used the same calibration area to build a new potential distribution model and compared four sets of variables (Supplementary Table 2). Sets 1 and 2 were based on the Spearman correlations (threshold=0.7) and the VIF (threshold=7) of the values of the occurrence records. Sets 3 and 4 were based on the Spearman correlations and the VIF of the values extracted from 10,000 points chosen at random from the calibration area of the initial model. We maintained the parameterization of the initial model without extrapolation (Berriozabal-Islas et al., Reference Berriozabal-Islas, Ramírez-Bautista, Torres-Ángeles, Mota Rodrigues, Macip-Ríos and Octavio-Aguilar2020). Subsequently, we identified the best ecological niche model for T. medemi based on kuenm metrics (AUCr, OR and AICc). Then, using the Multivariate Environmental Similarity Surfaces (MESS) analysis, we assessed climatic similarity amongst current and future scenarios to identify regions lacking climatic analogy and where the species response remains uncertain (Elith et al., Reference Elith, Kearney and Phillips2010). Finally, we projected the current environmental niche to future scenarios and transformed the logistic MaxEnt output into a binary (presence/absence) potential distribution map with the Spatial Analyst Reclassify tool in ArcGIS (Esri, 2021), and using the MTP threshold.
Results
Occurrence
We recorded 32 T. medemi in eight field surveys across the Atrato River and its tributaries in the municipalities of Turbo, Necoclí, Vigía del Fuerte (Antioquia), Bajirá and Bojayá (Chocó) during January 2022–February 2023. The species was confirmed in the Suriquí River (n = 24), the Salado Swamp (n = 2), and in the middle Atrato River basin at Napipí Swamp (n = 3), near Vigía del Fuerte (n = 2), and Bajirá River (n = 1; Fig. 1). We found no evidence of T. medemi in Arboletes, Apartadó, Mutatá (Antioquia), and Tierralta (Córdoba).
Of the 32 records obtained in the field, 24 turtles were captured by hand (capture success=0.0932). Five turtles provided by local residents were associated with the Napipí River (n = 3) and Vigía del Fuerte (n = 2), based on information from the owners and subsequent field verification. No additional individuals were found in these areas during later surveys, probably because of high water levels at the time. Three additional individuals were identified visually while basking, two in the Suriquí and one in the Bajirá River, with diagnostic features such as carapace shape and colour pattern allowing confident identification before they submerged. No turtles were captured using hoop net traps. We used two records of T. medemi from field reports (Cuadrado, pers. comm., 2023), corresponding to the Truandó and Salaquí Rivers in Riosucio (Chocó). In addition, we identified seven records from five localities reported in the scientific literature and the Colombian Biodiversity Information Facility (SiB Colombia), corresponding to Turbo, Chigorodó and Necoclí (Antioquia), and to Acandí and Unguía (n = 3; Chocó; Fig. 1).
Ecological niche modelling
The total dataset included 41 presence records of T. medemi (32 from our field surveys plus nine records provided by colleagues and from literature sources; see Methods). Following spatial and environmental cleaning, we retained 27 records for further analysis. The best performing environmental niche model (AUCr=1.81, OR=0%, AICc=469.3) was developed using a subset of the environmental variables (Set 2, Supplementary Table 1). The variables that contributed most to this model were minimum air temperature of the coldest month (Bio_6, 67.8% contribution, mean=22.9 °C, range=21.8–23.1 °C), mean hydrological slope (Slope_04, 10.8% contribution, mean=2.03°, range=0.07–7.60° slope gradient) and maximum upstream water temperature in the warmest month (Hydro_5, 9.5% contribution, mean=31.5 °C, range=30.0–32.8 °C). We transformed the suitability map (Fig. 1a) into a potential suitability map for T. medemi (Fig. 1b) using the logistic MTP with a threshold of 0.0187.
The map of potential distribution (Fig. 1b) encompassed the upper (Chigorodó River at the height of Istmina, Chocó), middle and lower sections of the Atrato River, covering c. 9,527 km2 of water. In the west, it extended to the Salaquí River (location 5, Fig. 1b), as well as the Chucunaque and Cartí Rivers within the Emberá-Wounaan and Guna Yala Indigenous regions in Panama. We confirmed the presence of T. medemi in the Salado Swamp in Necoclí (location 1, Fig. 1b), which was in doubt previously (Vargas-Ramírez et al., Reference Vargas-Ramírez, Del Valle, Ceballos and Fritz2017). Additionally, the model extended the area of potential suitability north-east to include the basins of the San Juan and Sinú Rivers, yet we neither captured nor sighted T. medemi during our field surveys there, and local people did not report its presence either. This area warrants further exploration given our limited sampling effort.
Effects of climate change on the potential distribution of T. medemi
The best performing model (AUCr=1.87, OR=0%, AICc=443, Supplementary Fig. 1) was developed using a subset of the environmental variables (Set 1, Supplementary Table 2). The variables that contributed most were minimum air temperature of the coldest month (Bio_6, 80.4% contribution) and maximum air temperature of the warmest month (Bio 5, 8.1% contribution); the remaining variables contributed < 5% each. The MESS analysis revealed non-analogous climates in current and future scenarios, particularly in the lower and middle Atrato River, leading us to dismiss these regions as having potential for occupation by T. medemi in future. We estimated the difference between the current and projected potential distribution areas to quantify the potential decrease in the geographical range. Our analysis demonstrated that occurrence would decline by 59% (intermediate scenario) and 69% (high-emissions scenario) by 2050 (Fig. 2), and by 71% (intermediate) and 86% (high-emissions) by 2070 (Fig. 3). This significant loss of suitable habitat primarily affected the main riverbed of the Atrato River, with a trend towards displacement to peripheral areas, leaving the population severely fragmented. These results are conservative with habitat loss projected to be even higher in the MRI-ESM2.0 (Supplementary Fig. 2) and ACCESS-CM2 models (Supplementary Fig. 3).
The potential distribution of the Atrato slider T. medemi in the Atrato River basin, north-west Colombia, modelled using the Model for Interdisciplinary Research on Climate Version 6 (MIROC6) and environmental variables from Worldclim 2.0, EarthEnv and PET to project occurrence in 2050 under (a) an intermediate (SSP245) and (b) a high-emissions (SSP585) climate change scenario. Blue lines indicate freshwater bodies, red points indicate field records (this study and Cuadrado, pers. comm.), and black points are presence records from published literature (Williams, Reference Williams1956; Castaño-Mora, Reference Castaño-Mora1992; Bock et al., Reference Bock, Páez and Castaño-Mora2012; Vargas-Ramírez et al., Reference Vargas-Ramírez, Del Valle, Ceballos and Fritz2017; Borja-Acosta & Galeano, Reference Borja-Acosta and Galeano2023). White-filled points indicate sites where T. medemi was not detected. Some points overlap as there were multiple records in close proximity. Locations with confirmed presence in 2022–2023 are indicated by 1–6 (see Fig. 1). (Readers of the printed journal are referred to the online article for a colour version of this figure.)
The potential distribution of the Atrato slider T. medemi in the Atrato River basin, north-west Colombia, modelled using MIROC6 and environmental variables from Worldclim 2.0, Earthenv and PET to project occurrence in 2070 under (a) an intermediate (SSP245) and (b) high-emissions (SSP585) climate change scenario. Blue lines indicate freshwater bodies, red points indicate field records (this study and Cuadrado, pers. comm.), and black points are presence records from published literature (Williams, Reference Williams1956; Castaño-Mora, Reference Castaño-Mora1992; Bock et al., Reference Bock, Páez and Castaño-Mora2012; Vargas-Ramírez et al., Reference Vargas-Ramírez, Del Valle, Ceballos and Fritz2017; Borja-Acosta & Galeano, Reference Borja-Acosta and Galeano2023). White-filled points indicate sites where T. medemi was not detected. Some points overlap as there were multiple records in close proximity. Locations with confirmed presence in 2022–2023 are indicated by 1–6 (see Fig. 1). (Readers of the printed journal are referred to the online article for a colour version of this figure.)
Discussion
We found that the geographical range of T. medemi encompassed the lower, middle and upper Atrato River of Colombia. The most important variables influencing the species’ distribution were terrestrial and freshwater temperature, which was also evident in the marked effect of climate change on the availability of suitable habitat in future decades.
Geographical range
Our ecological niche model aligns with Vargas-Ramírez et al. (Reference Vargas-Ramírez, Del Valle, Ceballos and Fritz2017) but updates the estimated geographical range based on field data and new records of occurrence (i.e. Vigía del Fuerte, Bajirá, Bojayá). Our species distribution model encompassed the area to the south of Quibdó and the potential range included not only the lower Atrato River but also the middle and upper regions.
The previous geographical range covered 19,644 km2 (Turtle Taxonomy Working Group, Reference Rhodin, Iverson, Bour, Fritz, Georges, Shaffer and van Dijk2021) including both terrestrial and aquatic habitats, whereas our model estimated a potential area of 9,527 km2 of aquatic habitat. We restricted our distribution model to water bodies because T. medemi is highly aquatic and its dispersal relies on connectivity between suitable habitat (Stanford et al., Reference Stanford, Iverson, Rhodin, van Dijk, Mittermier and Kuchling2020). In reality, the area of suitable aquatic habitat may be even smaller because T. medemi has not been detected in the San Juan and Sinú Rivers in Colombia and in rivers in Panamá even though the environmental conditions appear suitable. A survey of these rivers is needed but access restrictions may present challenges (Mauri, Reference Mauri2021; Fritz et al., Reference Fritz, Kehlmaier, Scott, Fournier, McCranie and Gallego-García2023).
Environmental requirements
We identified air temperature, water temperature and hydrological slope as primary limiting variables in our model of potential geographical range. Although the thermal tolerance limits of T. medemi are unknown, our field records and distribution models suggest a preference for a mean of 29 °C and a minimum of 23 °C for air temperature, and water temperatures below 31 °C. In freshwater turtles, temperatures below physiological limits can induce lethargy, loss of appetite and even decreased cardiac activity, whilst high temperatures can lead to muscle tremors and dehydration (Geng et al., Reference Geng, Dong, Wu and Lu2018; Tatum, Reference Tatum2020). Even within physiological limits, temperature fluctuations can affect the survival of offspring, growth rate and population sex ratio in species with temperature-dependent sex determination (Les et al., Reference Les, Paitz and Bowden2009; Doody & Moore, Reference Doody and Moore2010). It is likely that the sex ratio of T. medemi is affected by ambient temperature as incubation temperature determines sex in other Trachemys species, such as T. scripta (Czerwinski et al., Reference Czerwinski, Natarajan, Barske, Looger and Capel2016), T. dorbigni (Fagundes et al., Reference Fagundes, Bager and Cechin2010) and T. venusta (Vogt, Reference Vogt and Gibbons1990). Finally, hydrological slope (the slope gradient of the land influencing water movement) ranged between 0.07 and 7.60 degrees, and turtles were captured in areas below 100 m. This intolerance of hydrological slope supports the hypothesis that the Serranía de Abibe (100–1,500 m) in the Andean western mountain range restricts the dispersal of T. medemi from the Atrato River to the Sinú River basin (Pritchard & Trebbau, Reference Pritchard and Trebbau1984; Lynch & Suárez-Mayorga, Reference Lynch, Suárez-Mayorga and Orlando2004; Vargas-Ramírez et al., Reference Vargas-Ramírez, Del Valle, Ceballos and Fritz2017). It may also explain the allopatry of the closely related T. venusta callirostris.
Climate change
We suggest a significant decline in areas of suitable habitat as a consequence of climate change within 30–50 years (Figs 1, 2 & 3). Given that environmental changes associated with global warming are anticipated to occur within a short period (< 30 years), that turtles exhibit low vagility (Butler, Reference Butler2019; Berriozabal-Islas et al., Reference Berriozabal-Islas, Ramírez-Bautista, Torres-Ángeles, Mota Rodrigues, Macip-Ríos and Octavio-Aguilar2020) and that T. medemi is highly dependent on freshwater, the decrease in the potential area of suitable habitat is concerning. We acknowledge, however, that the niche transfers to future scenarios in our models did not include aquatic variables, which implies that the predicted geographical range of T. medemi applies to areas with similar climatic conditions only. Although climate transfers are valuable and commonly used, caution must be exercised when interpreting the results, as other significant variables, such as biotic variables (Anderson, Reference Anderson2016), have not been considered. In this regard, the models may over- or underestimate the area of potential future occurrence.
Our study estimated the extent of occurrence to be < 9,527 km2, which categorizes T. medemi as Vulnerable under the IUCN Red List Criterion B1, in agreement with the Turtle Taxonomy Working Group (Reference Rhodin, Iverson, Bour, Fritz, Georges, Shaffer and van Dijk2021). However, our climate change modelling predicted a drastic reduction in the extent of occurrence of 59% by the year 2050 (to 3,906 km2) in the best scenario and 86% by the year 2070 (to 1,334 km²) in the worst scenario, which would categorize T. medemi as Endangered (Criterion B1, < 5,000 km2). In addition to a predicted reduction in the availability of suitable habitat, T. medemi faces anthropogenic threats from hunting, trade and consumption, which has persisted for at least the last 30 years and is unlikely to cease in the future (Castaño-Mora, Reference Castaño-Mora1992; Bock et al., Reference Bock, Páez and Castaño-Mora2012; Ceballos & Brand, Reference Ceballos and Brand2014; Pedroza-Escobar & Valencia-Palacio, Reference Pedroza-Escobar and Valencia-Palacio2016). We recorded evidence of local people hunting T. medemi for food in our field study, which was confirmed by local wildlife officials.
Conservation of T. medemi
We presented our findings at the 2023 IUCN Species Survival Commission Tortoise and Freshwater Turtle Specialist Group meeting in Bolivia as evidence of the need to categorize T. medemi as Endangered on the IUCN Red List, which is under review (Turtle Taxonomy Working Group, Reference Rhodin, Iverson, Fritz, Gallego-García, Georges, Shaffer and van Dijk2025).
There is an urgent need to conserve this species in existing protected areas under the Natural Protected Areas System of Colombia. In particular, we advocate a focus on the National Forest Protecting Reserve of the Leon River (38,853 ha), Los Katios National Natural Park (700,000 ha), and the Integrated Management District of the Rionegro Ensenada, Marimonda and El Salado Wetlands (25,735 ha). However, because of ongoing and rapid habitat degradation, these areas will not be sufficient to protect the regional biodiversity. Community-based conservation programmes on unprotected lands are an alternative approach that has been effective elsewhere and could be encouraged in Colombia (Horwich and Lyon, Reference Horwich, Lyon and Jacobson1995). In addition, we suggest empowering local ecotourism programmes that already protect community lands around the Atrato drainage basin, such as the Natural Reserve Surikí (300 ha) in Antioquia. These projects require resources and education to balance the needs of wildlife with those of people. Finally, we strongly suggest that compensation measures carried out by regional private companies as a licensing requirement (Colombian Decree 2820 of 2010) imposed by the National Authority of Environmental Licences (ANLA), should include reforestation and land restoration of degraded areas. These are needed to facilitate connectivity between freshwater habitats in the future (Thiem et al., Reference Thiem, Birnie-Gauvin, Opperman, Franklin, Richter and Baumgartner2023), which will be crucial to the survival and persistence of T. medemi across its geographical range under a warming climate.
Author contributions
Study design: JG, CPC; data collection: JG; data analysis and writing: JG, CPC, OR.
Acknowledgements
We thank C. Mota-Vargas, D. Valencia-Rodríguez, the Bioclimatology lab at INECOL, E. Noguera-Urbano and M. Vargas-Ramírez for help with ecological niche modelling; S. Cuadrado for sharing records of T. medemi; and the Universidad de Antioquia (UdeA) veterinary students for help during data collection. J. Gaviria-Hernández received a Doctoral Scholarship from Ministerio de Ciencia, Tecnología e Innovación of Colombia; J. Gaviria-Hernández and C. P. Ceballos received a research grant from Re:wild-Turtle Conservation Fund (Grant TCF0866) and an Internal Grant from IUCN; J. Gaviria-Hernández received additional funding from the UdeA GAMMA research group (Convenio ES84190042-243-2022). We thank the local communities for permission to conduct research in their territories and share their knowledge of T. medemi.
Conflicts of interest
None.
Ethical standards
This research abided by the Oryx guidelines on ethical standards and was conducted with the approval of the Ethics Committee for Experimentation with Animals of the University of Antioquia, No. 142, 5 October 2021.
Data availability
The data supporting the findings of this study are available within the article and the Supplementary Material.
Supplementary material
The supplementary material for this article is available at doi.org/10.1017/S0030605324001625
Open access
