Herpetological Conservation and Biology 13(1):58–69.

Submitted: 3 September 2016; Accepted: 16 January 2018; Published 30 April 2018.

Distribution and Conservation Status of Andinobates

virolinensis (Dendrobatidae), a Threatened Andean Poison
Frog Endemic to Colombia
Eliana Ramos1, Fabio Leonardo Meza-Joya1,3, and Carlos Hernández-Jaimes1,2
1
Colombia Endémica, Asociación para el Estudio y la Conservación de los Recursos Naturales,
Bucaramanga, Colombia
2
Grupo de Investigación en Biotecnología Industrial y Biología Molecular, Escuela de Biología,
Universidad Industrial de Santander, Piedecuesta, Santander, Colombia
3
Corresponding author, e-mail: [email protected]

Abstract.—Detailed information of geographic distribution is critical for the conservation and management of
endangered and endemic taxa. Such knowledge is limited for the Santander Poison Frog (Andinobates virolinensis),
a threatened frog endemic to the Cordillera Oriental of the Colombian Andes. Here, we use new and historical data
to model the potential distribution of this species and estimate its extent of occurrence. Our model predicted that
suitable habitat exists on the western slope of the Cordillera Oriental in Santander, Boyacá, and Cundinamarca
departments in Colombia. The occurrence of this species was strongly, positively associated with precipitation of
the driest month, and positively, but more weakly related to mean diurnal temperature range and isothermality.
Our models suggest the low elevations of the Chicamocha and Sogamoso canyons and the high elevations of the
Cordillera Oriental constitute unsuitable habitats for this species. We identified 10,828 km2 of suitable habitat, of
which about 3.5% is inside protected areas. Our findings suggest that A. virolinensis should be re-categorized from
Endangered to Vulnerable in the Red List of the International Union for the Conservation of Nature. Improving
protective measures, collaboration with local farmers, and expanding the network of national protected areas are
likely to benefit A. virolinensis and other species from this Andean region.

Key Words.—endemic species; extent of occurrence; geographic distribution; species distribution model

Introduction The Santander Poison Frog, Andinobates virolinensis

(Ruiz-Carranza and Ramírez-Pinilla 1992), is a small
The lack of basic natural history and distribution dendrobatid (Fig. 1) found on the northwestern slope
data represents a challenge for conservation of many of the Cordillera Oriental of the Andes in Colombia,
amphibian species (Chunco et al. 2013). This holds with confirmed records in Cundinamarca and Santander
especially true for rare, threatened, and endemic species, departments (Ruiz-Carranza and Ramírez-Pinilla 1992;
and impedes the assessment of their conservation Stuart et al. 2008; Brown et al. 2011). This diurnal frog
status (Guisan et al. 2006; Kumar and Stohlgren 2009; inhabits primary and secondary cloud forests and some
Kamino et al. 2012; Groff et al. 2014; Foggi et al. 2015). traditional agroecosystems (Stuart et al. 2008; Meza-
Species distribution modeling has enormous potential Joya et al. 2015; Fig. 2A-C). The species is included in
for conservation planning because it can improve the the Andinobates bombetes species group (Brown et al.
understanding of geographic distribution and habitat 2011), a cluster of species threatened by ongoing loss of
suitability for data-poor species (Raxworthy et al. habitat due to agricultural expansion (Brown et al. 2011;
2003; Gaston and Fuller 2009; Kamino et al. 2012; Amézquita et al. 2013; Fig. 2D).
Khafagi et al. 2012; Fourcade et al. 2014). The extent The IUCN Red List of Threatened Species was
of occurrence (EOO) and area of occupancy (AOO) are designed to assess the extinction risk of species
two approaches to determining geographic distribution (Mace et al. 2008). Following the IUCN Categories
of species and both are used by the International and Criteria, A. virolinensis is listed as Endangered
Union for Conservation of Nature (IUCN) to evaluate B1ab(iii) because its estimated EOO is < 5,000 km2, it
conservation status (IUCN 2014). The EOO is the area was known to occur at no more than five threat-defined
within the outermost geographic limits of the occurrence locations (i.e., a geographically or ecologically distinct
of a species, whereas AOO is the area within the EOO area in which a single threat event will soon affect all
where the species is currently known to occur (Gaston individuals of a given taxon; IUCN 2014), and there
and Fuller 2009). is a continuing decline in the quality and extent of its

Copyright © 2018. Eliana Ramos 58

All Rights Reserved.
Ramos et al.—Distribution and conservation of Andinobates virolinensis.

Figure 1. Adult male of Santander Poison Frog (Andinobates virolinensis) in Santander Department, Serranía de los Yariguíes,
municipality of San Vicente de Chucurí, vereda La Colorada, Colombia. Note one tadpole on back being transported to phytotelmata in
bromeliads. (Photographed by Carlos A. Hernández).
natural habitat (Amézquita and Rueda-Almonacid between 0800 and 1600. We identified specimens based
2004). Most information about the assessment of the on the original description of the species (Ruiz-Carranza
species is anecdotal. Consequently, understanding of and Ramírez-Pinilla 1992) and comparison with the
current conservation status and the nature of threats is key provided by Brown et al. (2011). We deposited
incomplete for A. virolinensis. This species is known specimens (UIS-A 5505, UIS-A 5506) in the Colección
from one national protected area (Santuario de Fauna y Herpetológica of Universidad Industrial de Santander.
Flora Guanentá Alto Río Fonce; Amézquita and Rueda-
Almonacid 2004), but the size of its local range in this Distribution data.—We compiled distribution
and other protected areas is unknown. data from our field surveys, published literature, and
Herein, we provide new locality records and develop specimens housed at Colección Herpetológica of
a species distribution model (SDM) for A. virolinensis. Universidad Industrial de Santander, Colombia (UIS);
We use our data and model to estimate the EOO for this Colección Herpetológica of Grupo de Ecofisiología del
species. We also review the representation of the species Comportamiento y Herpetología of Universidad de Los
in protected areas within its distributional range and Andes, Colombia (GECOH); the online catalogue of
propose an update to its conservation status following Instituto de Ciencias Naturales (ICN) of Universidad
IUCN guidelines. Lastly, we discuss conservation and Nacional de Colombia (http://www.biovirtual.unal.
research priorities that should contribute to the long- edu.co [Accessed 5 November 2015]); and Colección
term survival of A. virolinensis. de Vertebrados of Instituto Alexander von Humboldt,
Colombia (IAvH) through the SiB Colombia (http://
Materials and Methods www.sibcolombia.net [Accessed 5 November 2015]).
We assigned latitude/longitude to localities that lacked
Surveys.—We surveyed for the presence of A. coordinates based on site descriptions by the collectors
virolinensis in 24 localities in Santander Department and plotted their locations with Global Gazetteer Version
(Colombia). We selected these localities to include 2.3 (http://www.fallingrain.com [Accessed 11 January
areas where the species is known to be present, 2016]) and Google Earth (Google Inc., Mountain
unsurveyed areas within its previously hypothesized View, California, USA; Table 1). Although there is
range (Amézquita and Rueda-Almonacid 2004), and uncertainty around these coordinates, we expect them to
areas near but beyond its range limits as currently fall near or in the correct pixel of the environmental data
understood. We conducted surveys between January (pixel size is 30 arc-seconds, or about 1 km2; see next
2013 and March 2016 (8 mo in 2013, 5 mo in 2014, 6 paragraph). We used the Spatially Rarefy Occurrence
mo in 2015, and 1 mo in 2016), totaling 1,462 sampling analysis (package SDMtoolbox; Brown 2014) to reduce
hours. We performed diurnal visual encounter surveys sampling biases via spatial filtering (Anderson and Raza
(Crump and Scott 1994) and opportunistic observations 2010; Boria et al. 2014). This approach reduced our

59
Herpetological Conservation and Biology

Figure 2. Habitats in the range of Santander Poison Frog (Andinobates virolinensis), Colombia. (A) Primary forest at type-locality,
municipality of Charalá, corregimiento of Virolín. (B) Secondary forest in municipality of Florián, vereda La Vueltiada. (C) Traditional
agroecosystem of mature mixed culture of native-shaded coffee and plantain trees in municipality of San Vicente de Chucurí, vereda
La Colorada. (D) Intensively grazed pastures near the type-locality (Santander Department). (Photographed by Carlos A. Hernández).

data to 12 independent occurrence records that were ≥ (www.cs.princeton.edu/~schapire/maxent [Accessed

10 km away from one another (Table 1). Using higher 3 February 2016]). To select the variables used in
filtering values (20 km) left too few occurrence points the final models, we followed the process outlined by
for model building, whereas decreasing filtering values Warren et al. (2014). We began by running a model
(5 km) increased the effects of sampling bias. including all bioclimatic variables and elevation, and
calculated the contribution scores for each variable.
Climate and elevation data for modeling.—We To do so, MaxEnt employs two metrics: percentage
used data for elevation and 19 bioclimatic variables contribution, which is a heuristic approach to estimate
(Appendix A; O’Donnell and Ignizio 2012) from the the contribution values of the corresponding variable by
WorldClim Project for the years 1960–1990 (Hijmans et the increase in gain (a measure of goodness-of-fit) in the
al. 2005; WorldClim. 2014. WorldClim - Global Climate model, and permutation importance, which is a measure
Data. Free climate data for ecological modeling and that determines the contribution of each variable by
GIS. Available at http://www.worldclim.org/. [Accessed randomly permuting each variable among the presence
1 December 2015]). We used two methods to define and background training points and measuring the
the limits of our study region (Fig. 3A), following resulting drop in the area under the curve (AUC). To
Anderson and Raza (2010). In Method 1, we calibrated get alternate estimates of which variables are most
the model to a rectangular region encompassing the important in the model, we also ran a jackknife test.
known localities for the species. Method 2 included This approach generates a series of models in an iterative
mainly the Andean Mountains of the study area from process, excluding one variable at a time, retaining
Method 1, which is recognized as the habitat of this each variable in isolation, and using all variables in
species (Amézquita and Rueda-Almonacid 2004). The conjunction, to provide information on how important
selection of these methods seems appropriate because it each variable is and how much unique information each
excludes large regions where the species is likely absent variable provides for the model (see online tutorial for
(i.e., areas where the species has not been collected MaxEnt at www.cs.princeton.edu/~schapire/maxent
historically [Brotons et al. 2004], or Inventory Pseudo- [Accessed 18 December 2017]). We calculated the
absences [Elith and Leathwick 2007]). spatial correlations (Pearson coefficient) between
variables using ENMTools 1.3 (Warren et al. 2010).
Modeling strategy.—We generated SDMs using We used contribution scores and scores from spatial
the software MaxEnt (Phillips et al. 2006) v. 3.3.3k correlations to reduce predictors from the full model.

60
Ramos et al.—Distribution and conservation of Andinobates virolinensis.

Table 1. Locality data (sorted from south to north) for Santander Poison Frog (Andinobates virolinensis) from Colombia. Date (month-
year) is for surveys conducted during this study. Dates in bold represent sites we did not survey for this study. Locality data (after
Spatially Rarefy Occurrence analysis) used in species distribution models are denoted (1). Coordinates inferred from Global Gazetteer
and Google Earth are marked (2). Coordinates are in decimal degrees (WGS-84 datum). Type Locality (sensu Ruiz-Carranza and
Ramírez-Pinilla 1992) includes coordinates based on subsequent surveys of Valderrama-Vernaza et al. (2010). Acronyms for museum
specimens are as in the text and elevation in meters above sea level.

Date Locality: Department, Municipality, Site Coordinates Elevation Source

10–1995 Cundinamarca, Yacopí, Guadalito, Cabo Verde 1, 2
5.5553°N, 74.2844°W 1,540 ICN-42926
11–2015 Boyacá, Otanche, La Cunchalita1, 2 5.6553°N, 74.1867°W 1,331 GECOH 1111–4
08–2014 Santander, Florián, La Vueltiada1 5.8138°N, 73.9817°W 1,746 This study (UIS-A 5505–6)
12–2015 Santander, La Belleza, Buena Vista 5.8456°N, 73.9628°W 1,969 Daniel Mejía, pers. obs.
01–2013 Santander, Gámbita, Bogotacito1 6.0146°N, 73.2157°W 2,400 ICN-12744
09–2014 Santander, Charalá, Virolín, El Palmar 6.0604°N, 73,2144°W 1,906 This study (not collected)
09–2014 Santander, Charalá, Virolín, Cerro El Rayo 6.0612°N, 73.1807°W 2,137 ICN-08551
09–2014 Santander, Charalá, Virolín, Costilla de Fara 6.0781°N, 73.2300°W 2,108 UIS-A-03584
06–2013 Santander, Charalá, Virolín, Costilla de Fara 6.0783°N, 73,1964°W 1,785 ICN-05482
09–2013 Santander, Charalá, Virolín1
6.0954°N, 73.2006°W 1,807 UIS-A-00108
09–2013 Santander, Charalá, Virolín, El Reloj1 6.0983°N, 73.2187°W 1,744 ICN-16101; Type Locality
04–2015 Santander, Charalá, Virolín 6.0989°N, 73.1743°W 1,780 ICN-04588
04–2015 Santander, Charalá, Virolín 6.1651°N, 73,2463°W 1,946 ICN-04256
07–1985 Santander, Ocamonte1, 2 6.3411°N, 73.1048°W 1,670 IAvH-Am-1132
07–1996 Santander, Confines, Km 11.2 road Oiba to Socorro1 6.3558°N, 73.2567°W 1,650 ICN-52860
Unknown Santander, Socorro 1, 2
6.4658°N, 73.2430°W 1,578 Brown et al. (2011)
12–2013 Santander, San Vicente de Chucurí, Pamplona1 6.7048°N, 73.4361°W 1,700 ICN-26984
08–2014 Santander, San Vicente de Chucurí, La Colorada1 6.7966°N, 73.4785°W 1,450 Meza-Joya et al. (2015)
07–2014 Santander, San Vicente de Chucurí, Cerro de las Tetas 6.8453°N, 73.3812°W 1,803 This study (not collected)
06–2013 Santander, San Vicente de Chucurí, El Centro1 6.8669°N, 73.3793°W 1,543 This study (UIS-A-3755)

We eliminated variables with low contribution and and Anderson (2014). We used recommended default
permutation importance scores (< 5%) in the full model. values for convergence threshold (10–5), maximum
We retained environmental variables with the highest number of iterations (500), maximum number of
jack-knife scores. We deleted variables that were highly background points (104), and default prevalence of the
correlated with these kept variables (Pearson r > 0.70). species (0.5). Lastly, we selected the logistic output
This process resulted in three bioclimatic variables for format, which yields continuous values ranging from 0 to
subsequent models (Table 2). 1 that indicate the probability of suitable environmental
Because of the low number of independent occur- conditions for the species (see Phillips and Dudík 2008).
rences (12), we generated models using the cross-
validated approach, with the minimum training presence Performance of models.—We evaluated model
(equal to the lowest presence decision threshold) to performance using (1) area under the curve (AUC) of
distinguish suitable from unsuitable areas (Pearson et the receiver operating characteristic (ROC) measure
al. 2007). This threshold identifies pixels predicted provided by MaxEnt, (2) success rate in jack-knife
to be at least as suitable as those where the species tests using the pValue Compute program from Pearson
has been recorded (Pearson et al. 2007). This method et al. (2007), and (3) sample size-corrected Akaike
has previously been used with sample sizes as small information criteria (AICc; Akaike 1974; Burnham and
as five records (e.g., Pearson et al. 2007; Anderson Anderson 2002) using ENMTools 1.3 (Warren et al.
and Raza 2010; Kamino et al. 2012; Chunco et al. 2010). Once we selected the best model (Table 3), its
2013; Shcheglovitova and Anderson 2013). To avoid logistic output was transformed to a binary prediction
overparameterization, we used linear plus quadratic model for the suitable habitat of the species (i.e., a
(LQ) and hinge (H) features (Shcheglovitova and presence/absence map) by applying the minimum
Anderson 2013; van Proosdij et al. 2016). We assessed training presence threshold value (0.401) obtained by
three alternative regularization multiplier values (0.5 MaxEnt. Then, we evaluated the final binary model
and 2.0; default setting is 1.0) following Radosavljevic by visual examination based on our knowledge of

61
Herpetological Conservation and Biology

Figure 3. Distribution of Santander Poison Frog (Andinobates virolinensis) in the Cordillera Oriental of the Colombian Andes, Colombia.
(A) Location of the study area in north of South America. Rectangles represent the two methods used to define the study region for
calibrating distribution models of species. Method 1 is red. Method 2 (blue) defines a smaller region mainly in the Andes. Elevation
units are meters. (B) Regional map of historical (white circles) and new localities found during this study (green circles). Red polygon
indicates the range of the species sensu the International Union for the Conservation of Nature. (C) Spatially filtered localities used to
build the species distribution model (white circles). Locality details are in Table 1.

the natural history and geographic distribution of A. convex polygon enclosing all occurrence sites. We
virolinensis. This examination led us to the exclusion removed pixels identified as unsuitable habitats that
of a few small isolated areas located on the eastern were outside the documented altitudinal range of
slope of the Cordillera Oriental in Norte de Santander this species (i.e., 1,331–2,400 m; see Results), as
department, a region where no species of this genus is defined in the IUCN mapping protocols (https://www.
known to occur, and areas on the northwestern slope of conservationtraining.org [Accessed 16 March 2016]).
the Cordillera Central in Antioquia Department where Lastly, we calculated the area of the range polygon of
another species in the bombetes group (Andinobates the species as a proxy for the extent of occurrence of
opisthomelas) occurs (Acosta Galvis, A.R. 2017. Lista the species by summing the pixels within the final MCP.
de los anfibios de Colombia: Referencia en línea.
V.07.2017.0. Electronic database available at http:// Table 3. Performance statistics for two methods used to model
www.batrachia.com. [Accessed 28 December 2017]). the potential distribution of Santander Poison Frog (Andinobates
After our evaluation, we calculated the extent of virolinensis). An asterisk (*) denotes the combination of features
with highest performance values. AICc with a dash (--) are
occurrence based on pixels within the binary model. models with more parameters than occurrence points (i.e., > 12
parameters). The abbreviation Regul. = Regularization.
Minimum convex polygon.—We generated a
Success
minimum convex polygon (MCP) using Quantum GIS
Method Features Regul. AUC Rate AICc P
software (QGIS Development Team. 2016. Quantum
1 Linear plus 0.5 0.928 0.8 251.7 < 0.001
GIS Geographic Information System. Open Source quadratic
Geospatial Foundation Project. Version 2.8.2. Available
1 0.914 0.8 260.6 < 0.001
at http://qgis.osgeo.org [Accessed 22 January 2016]).
2 0.884 0.8 260.9 < 0.001
We used the Convex Hull function to create the smallest
Hinge 0.5 0.950 0.8 -- < 0.001
Table 2. Contribution percentage and permutation importance 1 0.954 0.8 -- < 0.001
of environmental variables used in distribution modeling of 2 0.955 1.0 257.6 < 0.001
Santander Poison Frog (Andinobates virolinensis). Variable codes
are defined in Appendix A. 2 Linear plus 0.5 0.948 0.6 242.1 < 0.001
quadratic
Environmental Contribution Permutation 1 0.936 0.8 247.8 < 0.001
Variable Code percentage importance
2 0.903 0.8 250.1 < 0.001
Precipitation of the BIO14 77.9 64.3
Driest Month Hinge 0.5 0.959 0.8 248.1 < 0.001

Mean Diurnal BIO2 17.1 19.7 1 0.960 0.8 246.1 < 0.001
Temperature Range 2 0.962* 1.0* 232.1* < 0.001

62
Ramos et al.—Distribution and conservation of Andinobates virolinensis.

Figure 4. MaxEnt models of the potential geographic distribution of Santander Poison Frog (Andinobates virolinensis). (A) Logistic
output and (B) binary model (orange) after applying the decision threshold and visual validation. White circles indicate unfiltered
presence records (see Table 1). Elevation units are meters. (C) Protected areas (blue) near or in the binary model (orange). (D) Protected
areas (blue) near or in the minimum convex polygon (orange). Protected areas are: 1 = Parque Nacional Natural Serranía de los Yariguíes,
2 = Santuario de Fauna y Flora Guanentá Alto Río Fonce, 3 = Reserva Forestal Protectora Sierra El Peligro, and 4 = Reserva Forestal
Protectora Cuchilla El Minero.

Results virolinensis from Boyacá Department in the municipality

of Otanche (GECOH 1111–4; Fig. 3A), as well as
Distribution data.—Our field surveys detected A. the northernmost locality for the species in northern
virolinensis in 13 localities: four were new and nine Serranía de los Yariguíes, Santander Department
were documented in herpetological collections. We (UIS-A-3755; municipality of San Vicente de Chucurí,
did not find the species in 11 of the surveyed localities vereda El Centro). Based on our detections and review
(Appendix B). We obtained additionally seven locality of historical records, we define the altitudinal range
records of which we were not aware at the outset of the for A. virolinensis as from 1,331 m (GECOH 1111–4)
study: two from published literature, four from scientific to about 2,400 m (ICN-12744). These data extend
collections, and one new locality from an unpublished the distribution of A. virolinensis along the western
observation (Table 1). Hitherto unpublished specimens slope of the Cordillera Oriental in Colombia; its range
in herpetological collections provided first records of A. now extends from northern Serranía de los Yariguíes

63
Herpetological Conservation and Biology

Table 4. Extent of occurrence (EOO) estimated for Santander Our model yielded very low habitat suitability scores
Poison Frog (Andinobates virolinensis) from the species distribution at high elevations (above about 2,400 m) in the Cordillera
model (SDM) and the minimum convex polygon (MCP).
Oriental and low elevations (below about 1,300 m)
Protected areas in Colombia are Parque Nacional Natural Serranía
de los Yariguíes (SY), Santuario de Fauna y Flora Guanentá Alto in the Chicamocha and Sogamoso canyons. Habitats
Río Fonce (GF), Reserva Forestal Protectora Nacional Cuchilla El beyond these unsuitable features were identified by the
Minero (CM), and Reserva Forestal Protectora Nacional Sierra El model as of low suitability and were clipped from the
Peligro (SP). final model based on the threshold value (Fig. 4B; see
EOO Methods). We estimated the total EOO for the species
EOO in protected
areas (km2) under to be 10,828 km2. The SDM showed that most of the
EOO protection predicted occurrence of this species (about 96.5% of
Method (km²) SY GF CM SP (%) EOO) is outside boundaries of protected areas (Fig. 4C;
SDM 10,828 311.8 30.6 18.7 15.3 3.47 Table 4).
MCP 6,973 446.9 2.1 - 11.9 6.61
Minimum convex polygon.—The EOO estimated
in Santander Department (UIS-A-3755) to northern with the MCP was 6,973 km2. The MCP (Fig. 4D)
Cundinamarca Department (ICN-12744; Figs. 3A and excludes large regions of Andean forest identified as
4). This represents an increase of at least 4,482 km2 over suitable habitat by the SDM (Fig. 4C). The MCP showed
the range reported by the IUCN (2,491 km2; Amézquita that a small portion of the predicted occurrence of this
and Rueda-Almonacid 2004). species (about 6.6% of EOO) is outside boundaries of
protected areas (Fig. 4D; Table 4).
Species distribution model.—Inclusion of
georeferenced localities generated models that better Discussion
represent the range limits of the species as currently
understood. Thus, we included these data in our The new records presented here improve our
distribution models. The extent of the study region knowledge of the geographical distribution of A.
affected model performance. Models calibrated mainly virolinensis, extend the range of the species, and
in the Andean region (Method 2) performed better than suggest areas for additional surveys where the species
models based on the broader region (Method 1). Method could be present. In Santander Department, the known
2 yielded a higher success rate than Method 1 (Table distribution includes the type locality and several
3), indicating low omission rates. Models with the additional locations through the western slope of the
hinge feature and high regularization multipliers (i.e., Cordillera Oriental. Single localities also occur in Boyacá
2.0) performed better and with less parameterization and Cundinamarca departments. The new locality
(Table 3). SDMs generated with Method 2 led to more record provided here from Boyacá Department fills the
statistically robust, less parameterized models and distribution gap between Santander and Cundinamarca
fewer predictions in environments where the species is departments. There is little or no information about
likely absent (Fig. 4A). The final model performed well A. virolinensis in several areas predicted to be suitable
with an AUC value of 0.957 ± 0.006 SD. The model by the models (e.g., southern Santander and northern
had a predictive success rate equal to 0.8 (P < 0.001), Boyacá and Cundinamarca). Factors not incorporated in
indicating low omission rates. These results indicate our models (e.g., land-cover, interspecies interactions,
that the model is informative for potential suitable diseases) may also limit the presence of A. virolinensis
habitat for A. virolinensis. in areas identified as suitable.
Predicted habitat suitability for A. virolinensis was According to the information available to date,
strongly associated with precipitation in the driest month we propose that the current IUCN status for A.
(contribution 77.9%) and modestly associated with virolinensis (i.e., Endangered, see Introduction) should
the mean diurnal temperature range and isothermality be downgraded to Vulnerable (VU), because its extent
(Table 2). Suitability increased sigmoidally with the pre- of occurrence (EOO) is > 5,000 km2 and < 20,000 km2
cipitation of the driest month, mean diurnal temperature (SDM = 10,828 km2, MCP = 6,973 km2). However, this
range, and isothermality. Suitability for A. virolinensis conservation status could change as new information is
was maximized around 50–60 mm of precipitation in the collected. The destruction of forests represents a major
driest month, 10–12° C for mean diurnal temperature threat for this species, as well as other species in this
range, and 83–93 for isothermality. The SDM predicted region of the Andes. This is especially true for Santander
suitable habitat for A. virolinensis on the western slope and Boyacá departments, where livestock grazing and
of the Cordillera Oriental in Santander, Boyacá, and intensive farming continue to increase (Sánchez-Cuervo
Cundinamarca departments (Fig. 4). et al. 2012). However, reliable data on how rapidly
forest destruction is occurring in the study area and how

64
 Ramos et al.—Distribution and conservation of Andinobates virolinensis.

this species responds to forest loss and degradation are However, some limitations are worth noting, including
lacking. The discovery of a population of A. virolinensis the limited occurrence data to build models using
on traditional sustainable agricultural systems at standard approaches and the absence of information
relatively high densities (42–73 adult individuals per ha; about changes in land use throughout the predicted range
Meza-Joya et al. 2015) suggests that it may be resilient of the species. Further surveys in unexplored or under-
to some types of land cover change, provided some key sampled areas, taking into consideration the detection
features of the native habitats are retained (e.g., large probabilities for the species, are needed. Such data
native trees with bromeliads). Estimations of detection would help generate more robust SDMs and improve
probabilities and local abundances are critical to assess understanding of the extent of occurrence and area of
occupancy and population trends for A. virolinensis. occupancy for this species.
Precipitation during the driest month was the most
important variable explaining the modeled range of A. Acknowledgments.—Collection and research permits
virolinensis (Table 2). Only two other predictors were in were granted by Autoridad Nacional de Licencias
our final model (mean diurnal range and isothermality) Ambientales (Resolución 0047-2015). Financial support
and these were much less influential. Areas with was provided by Conservation Leadership Programme
conditions outside favored range of precipitation during (Project # 02186714), Asociación Colombia Endémica,
the driest month (50–60 mm), and to a lesser extent, and Asociación Colombiana de Herpetología. We are
mean diurnal temperature range (10–12° C), likely very grateful to Daniel Mejía who kindly allowed us
represent unsuitable conditions for this species. High to use his unpublished locality records of Andinobates
elevation summits (above 2,400 m elevation) and low virolinensis and provided us with the data from Grupo
elevation canyons (below 1,300 m) may be outside the de Ecofisiología del Comportamiento y Herpetología
physiological tolerances and ecological requirements of of Universidad de Los Andes, Colombia (GECOH).
the species. The Chicamocha and Sogamoso canyons We thank Martha Ramírez for granting us access
reach depths of 400 m and host dry tropical forest and to specimens housed at herpetological collection of
semi-arid spiny tropical scrubland. High elevations Universidad Industrial de Santander (UIS). We also
(up to about 2,600 m) support forests and Páramo thank Claudia Infante, Jennifer Quintero, and Oscar
ecosystems characterized by low temperatures and Hernández, who collaborated during fieldwork. We
high solar radiation (Kattan et al. 2004; Navas 2006; also thank Daniel Mejía, Mauricio Torres, Ralph
Morales et al. 2007). Our inspections of the literature, Saporito, and Stefan Lötters for valuable suggestions
herpetological collections, and unpublished results from and comments on this paper.
surveys in and around these features have failed to
document A. virolinensis. Literature Cited
Our modeling suggests that national protected areas
currently provide limited refuge for A. virolinensis. Akaike, H. 1974. A new look at the statistical model
Through active participation of local communities, identification. IEEE Transactions on Automatic
economic development could be promoted with Control 19:716–723.
sustainable agriculture practices that would facilitate the Amézquita, A., and J.V. Rueda-Almonacid. 2004.
conservation of A. virolinensis (Meza-Joya et al. 2015). Andinobates virolinensis. IUCN Red List of
Such conservation efforts in agricultural landscapes, Threatened Species. Andinobates virolinensis.
however, should be considered as complementary to Version 2014.3. http://www.iucnredlist.org.
enforcing the protection and restoration of national Amézquita, A., R. Márquez, R. Medina, D. Mejía-Vargas,
protected areas. Further work is needed to reinforce T.R. Kahn, G. Suárez, and L. Mazariegos. 2013. A
habitat protection both outside and within the existing new species of Andean poison frog, Andinobates
network of protected areas in Colombia to ensure the (Anura: Dendrobatidae), from the northwestern
long-term survival of this and other organisms in this Andes of Colombia. Zootaxa 3620:63–178.
Andean region. Anderson, R.P., and A. Raza. 2010. The effect of the
Our SDM predictions are based on broad climatic extent of the study region on GIS models of species
variables, but other factors (e.g., biotic interactions, geographic distributions and estimates of niche
stochastic events) can shape distributions of amphibians evolution: preliminary tests with montane rodents
(Blank and Blaustein 2012). Further studies that link (genus Nephelomys) in Venezuela. Journal of
ecology, phylogeography, and behavior, are required Biogeography 37:1378–1393.
to develop integrated conservation strategies for A. Blank, L., and L. Blaustein. 2012. Using ecological
virolinensis. Our study presents a regional habitat niche modelling to predict the distributions of two
context for developing management plans and endangered amphibian species in aquatic breeding
conservation policies aimed at conserving this species. sites. Hydrobiologia 693:157–167.

65
Herpetological Conservation and Biology

Boria, R.A., L.E. Olson, S.M. Goodman, and R.P. surveys: a case study using the Oregon Spotted Frog
Anderson. 2014. Spatial filtering to reduce sampling (Rana pretiosa). Herpetological Conservation and
bias can improve the performance of ecological Biology 9:354–368.
niche models. Ecological Modelling 275:73–77. Guisan, A., O. Broennimann, R. Engler, M. Vust, N.G.
Brotons, L., W. Thuiller, M.B. Araújo, and A.H. Yoccoz, A. Lehmann, and N.E. Zimmermann. 2006.
Hirzel. 2004. Presence-absence versus presence- Using niche-based models to improve the sampling
only modelling methods for predicting bird habitat of rare species. Conservation biology 20:501–511.
suitability. Ecography 27:437–448. Hijmans, R.J., S.E. Cameron, J.L. Parra, P.G. Jones, and
Brown, J.L. 2014. SDMtoolbox: a python-based GIS A. Jarvis. 2005. Very high resolution interpolated
toolkit for landscape genetic, biogeographic and climate surfaces for global land areas. International
species distribution model analyses. Methods in journal of climatology 25: 1965–1978.
Ecology and Evolution 5:694–700. International Union for the Conservation of Nature
Brown, J.L., E. Twomey, A. Amézquita, M.B. De Souza, (IUCN Standards and Petitions Subcommittee).
J.P. Caldwell, S. Lötters, R. Von May, P.R. Melo- 2014. Guidelines for Using the IUCN Red List
Sampaio, D. Mejía-Vargas, P. Pérez-Peña, et al. 2011. Categories and Criteria. Version 11. http://www.
A taxonomic revision of the Neotropical poison frog iucnredlist.org.
genus Ranitomeya (Amphibia: Dendrobatidae). Kamino, L.H.Y., M.F. Siqueira, A. Sánchez-Tapia, and
Zootaxa 3083:1–120. J.R Stehmann. 2012. Reassessment of the extinction
Burnham, K. P., and D. R. Anderson. 2002. Model risk of endemic species in the Neotropics: how can
Selection and Multimodel Inference: A Practical modelling tools help us? Natureza and Conservação
Information-theoretic Approach. 2nd Edition. 10:191–198.
Springer-Verlag, Berlin, Germany. Kattan, G.H., P. Franco, V. Rojas, and G. Morales.
Chunco, A.J., S. Phimmachak, N. Sivongxay, and B.L. 2004. Biological diversification in a complex
Stuart. 2013. Predicting environmental suitability for region: a spatial analysis of faunistic diversity and
a rare and threatened species (Lao Newt, Laotriton biogeography of the Andes of Colombia. Journal of
laoensis) using validated species distribution models. Biogeography 31:1829–1839.
PLoS ONE 8:e59853. https://doi.org/10.1371/ Khafagi, O., E.E. Hatab, and K. Omar. 2012. Ecological
journal.pone.0059853. niche modelling as a tool for conservation planning:
Crump, M.L., and N.J. Scott Jr. 1994. Visual encounter suitable habitat for Hypericum sinaicum in South
surveys. Pp. 84–92 In Measuring and Monitoring Sinai, Egypt. Universal Journal of Environmental
Biological Diversity: Standard Methods for Research and Technology 2:515–524.
Amphibians. Heyer, W.R., M.A. Donnelly, R.W. Kumar, S., and T.J. Stohlgren. 2009. Maxent modeling
McDiarmid, L.A.C. Hayek, and M.S. Foster (Eds.). for predicting suitable habitat for threatened and
Smithsonian Institution Press, Washington D.C., endangered tree Canacomyrica monticola in
USA. New Caledonia. Journal of Ecology and Natural
Elith, J., and J. Leathwick. 2007. Predicting species Environment 1:094–098.
distributions from museum and herbarium records Mace, G.M., N.J. Collar, K.J. Gaston, C. Hilton-Taylor,
using multiresponse models fitted with multivariate H.R. Akcakaya, N. Leader-Williams, E.J. Milner-
adaptive regression splines. Diversity and Gulland, and S.N. Stuart. 2008. Quantification
Distributions 13:265–275. of extinction risk: IUCN's system for classifying
Foggi, B., D. Viciani, R.M. Baldini, A. Carta, and threatened species. Conservation Biology 22:1424–
T. Guidi. 2015. Conservation assessment of the 1442.
endemic plants of the Tuscan Archipelago, Italy. Meza-Joya, F.L., E. Ramos-Pallares, and C. Hernández-
Oryx 49:118–126. Jaimes. 2015. Use of an agroecosystem by the
Fourcade, Y., J.O. Engler, D. Rödder, and J. Secondi. threatened dart poison frog Andinobates virolinensis
2014. Mapping species distributions with MAXENT (Dendrobatidae). Herpetological Review 46:171–
using a geographically biased sample of presence 176.
data: a performance assessment of methods for Morales, M., J. Otero, T. Van Der Hammen, A. Torres,
correcting sampling bias. PLoS ONE 9:e97122. C. Cadena, C. Pedraza, N. Rodríguez, C. Franco, J.C.
https://doi.org/10.1371/journal.pone.0097122. Betancourth, E. Olaya, et al. 2007. Atlas de Páramos
Gaston, K.J., and R.A. Fuller. 2009. The sizes of species’ de Colombia. Instituto de Investigación de Recursos
geographic ranges. Journal of Applied Ecology Biológicos Alexander von Humboldt. Bogotá, D.C.,
46:1–9. Colombia.
Groff, L.A., S.B. Marks, and M.P. Hayes. 2014. Using Navas, C.A. 2006. Patterns of distribution of anurans
ecological niche models to direct rare amphibian in high Andean tropical elevations: insights

66
Ramos et al.—Distribution and conservation of Andinobates virolinensis.

from integrating biogeography and evolutionary Sánchez-Cuervo, A.M., T.M. Aide, M.L. Clark, and
physiology. Integrative and Comparative Biology A. Etter. 2012. Land cover change in Colombia:
46:82–91. surprising forest recovery trends between 2001 and
O’Donnell, M.S., and D.A. Ignizio. 2012. Bioclimatic 2010. PLoS ONE 7:e43943. https://doi.org/10.1371/
predictors for supporting ecological applications in journal.pone.0043943.
the conterminous United States. Data Series 691, Shcheglovitova, M., and R.P. Anderson. 2013.
U.S. Geological Survey, Reston, Virginia, USA. 10 Estimating optimal complexity for ecological niche
p. models: a jackknife approach for species with small
Pearson, R.G., C.J. Raxworthy, M. Nakamura, and sample sizes. Ecological Modelling 269:9–17.
A.T. Peterson. 2007. Predicting species distributions Stuart, S.N., M. Hoffmann, J.S. Chanson, N.A.
from small numbers of occurrence records: a test Cox, R.J. Berridge, P. Ramani, and B.E. Young.
case using cryptic geckos in Madagascar. Journal of 2008. Threatened Amphibians of the World. Lynx
Biogeography 34:102–117. Edicions, Barcelona, Spain, International Union for
Phillips, S.J., and M. Dudík. 2008. Modeling of species the Conservation of Nature, Gland, Switzerland,
distributions with Maxent: new extensions and a and Conservation International, Arlington, Virginia,
comprehensive evaluation. Ecography 31:161–175. USA.
Phillips, S.J., R.P. Anderson, and R.E. Schapire. 2006. Valderrama-Vernaza, M., V.H. Serrano-Cardozo, and
Maximum entropy modeling of species geographic M.P. Ramírez-Pinilla. 2010. Reproductive activity
distributions. Ecological Modelling 190:231–259. of the Andean frog Ranitomeya virolinensis (Anura:
Radosavljevic, A., and R.P. Anderson. 2014. Making Dendrobatidae). Copeia 2010:211–217.
better Maxent models of species distributions: van Proosdij, A.S.J., M.S.M. Sosef, J.J. Wieringa, and N.
complexity, overfitting and evaluation. Journal of Raes. 2016. Minimum required number of specimen
Biogeography 41:629–643. records to develop accurate species distribution
Raxworthy, C.J., E. Martinez-Meyer, N. Horning, R.A. models. Ecography 9:542–552.
Nussbaum, G.E. Schneider, M.A. Ortega-Huerta, Warren, D.L., R.E. Glor, and M. Turelli. 2010.
and A.T. Peterson. 2003. Predicting distributions of ENMTools: a toolbox for comparative studies of
known and unknown reptile species in Madagascar. environmental niche models. Ecography 33:607–
Nature 426:837–841. 611.
Ruiz-Carranza, P.M., and M.P. Ramírez-Pinilla. Warren, D.L., A.N. Wright, S.N. Seifert, and H.B.
1992. Una nueva especie de Minyobates (Anura: Shaffer. 2014. Incorporating model complexity and
Dendrobatidae) de Colombia. Lozania 61:1–16. spatial sampling bias into ecological niche models of
climate change risks faced by 90 California vertebrate
species of concern. Diversity and Distributions
20:334–343.

Eliana Ramos is a Biologist with a Master’s degree and she is largely interested in herpetology. Her main
research interests are the ecology, natural history, and conservation of Neotropical amphibians and reptiles.
(Photographed by Carlos A. Hernández).

Fabio Leonardo Meza-Joya is a Biologist with a Master’s degree who has been studying amphibians and
reptiles in Colombia since 2009. His research interests span a wide variety of topics, including morphol-
ogy, taxonomy, ecology, natural history, evolution, and conservation of amphibian and reptiles. (Photo-
graphed by Eliana Ramos).

Carlos Hernández-Jaimes is currently in a Master’s program in Biology at Universidad Industrial de

Santander in Colombia. His main research interests comprise the fields of molecular biology applied to
evolution and conservation studies. (Photographed by Oscar Hernández).

67
Herpetological Conservation and Biology

Appendix A. Environmental variables used in species distribution models. Data are from WorldClim (Hijmans et al. 2005). Units are in
brackets.

Variable Code Definition

Annual Mean Temperature BIO1 Annual mean temperature [° C].
Mean Diurnal Temperature Range BIO2 Mean of monthly (max temp - min temp) [° C].
Isothermality BIO3 (Mean Diurnal Range)/(Temperature Annual Range)×(100) [%].
Temperature Seasonality BIO4 (Standard deviation of mean monthly temperatures)×(100) [° C].
Max Temperature of Warmest Month BIO5 Maximum temperature value across all months within a given year [° C].
Min Temperature of Coldest Month BIO6 Minimum temperature value across all months within a given year [° C].
Temperature Annual Range BIO7 (Max Temperature of Warmest Month - Min Temperature of Coldest Month) [° C].
Mean Temperature of Wettest Quarter BIO8 Mean temperature during the wettest quarter [° C].
Mean Temperature of Driest Quarter BIO9 Mean temperature during the driest quarter [° C].
Mean Temperature of Warmest Quarter BIO10 Mean temperature during the warmest quarter [° C].
Mean Temperature of Coldest Quarter BIO11 Mean temperature during the coldest quarter [° C].
Annual Precipitation BIO12 Annual total precipitation [mm].
Precipitation of Wettest Month BIO13 Total precipitation during the wettest month [mm].
Precipitation of Driest Month BIO14 Total precipitation during the driest month [mm].
Precipitation Seasonality BIO15 Variation in monthly precipitation totals over the course of the year [%].
Precipitation of Wettest Quarter BIO16 Total precipitation during the wettest quarter [mm].
Precipitation of Driest Quarter BIO17 Total precipitation during the driest quarter [mm].
Precipitation of Warmest Quarter BIO18 Total precipitation during the warmest quarter [mm].
Precipitation of Coldest Quarter BIO19 Total precipitation during the coldest quarter [mm].

68
 Appendix B. Sampling localities where Santander Poison Frog (Andinobates virolinensis) has not been recorded. Data sources are from this study and from additional surveys by the authors (ERP,
FLMJ, and CHJ). Coordinates are in decimal degrees (WGS-84 datum). ‘Effort’ is survey effort in total sampling hours (number of surveys). Date (month and year) is for most recent survey carried
out in each locality. Localities are sorted chronologically from older to recent.

Date Municipality Department Locality Effort Coordinates Elevation (m) Source

01–2011 El Cerrito Santander Las Arreviatadas lagoon complex 72 (2) 6.8559°N, 72.5235°W 3,360 FLMJ
04–2011 Lebrija Santander Vereda La Girona 16 (1) 7.3234°N, 73.3375°W 202 CHJ
04–2011 Piedecuesta Santander Reserva Forestal El Rasgón 432 (8) 7.0438°N, 72.9808°W 2,345 ERP and FLMJ
05–2011 Betulia Santander Vereda Sogamoso, site Corintios 24 (1) 6.9966°N, 73.4119°W 658 FLMJ
05–2011 Zapatoca Santander Vereda San Javier, stream La Ramera 36 (1) 6.8219°N, 73.3347°W 2,145 FLMJ
01–2012 Bucaramanga Santander Barrio La Esperanza II, cañada La esperanza 56 (3) 7.1536°N, 73.1284°W 708 FLMJ and CHJ
03–2012 Floridablanca Santander Jardín Botánico Eloy Valenzuela 30 (3) 7.0673°N, 73.0872°W 915 CHJ
01–2013 Tona Santander Vereda El Quemado, Reserva Núcleo Arnania 245 (5) 7.2074°N, 73.0072°W 1,993 This study
05–2013 Floridablanca Santander Pico de La Judía, Reserva Los Maklenques 68 (2) 7.0874°N, 73.0464°W 1,678 This study
06–2013 Piedecuesta Santander Vereda Las Amarillas 64 (2) 6.9973°N, 72.9794°W 2,169 This study
06–2013 Girón Santander Vereda Sogamoso, site Linderos 120 (3) 7.1002°N, 73.3875°W 350 CHJ
08–2013 Floridablanca Santander Vereda Guarumales, farm El Carajo 104 (2) 7.1469°N, 73.0362°W 2,127 This study
08–2013 Matanza Santander Santa Cruz de la Colina, vereda Guayaquil 64 (2) 7.3722°N, 73.1024°W 1,495 FLMJ
10–2013 Los Santos Santander Mesa de Los Santos, Hacienda El Roble 134 (3) 6.8617°N, 73.0491°W 1,632 This study
10–2013 Guapotá Santander Vereda Las Flores, farm La Chocolatera 87 (2) 6.3905°N, 73.3262°W 1,102 This study

69
11–2013 Santa Bárbara Santander Vereda Esparta, farm Monterey 64 (2) 7.0203°N, 72.9027°W 2,263 FLMJ
04–2014 Barrancabermeja Santander Vereda Pozo de Nutria, Quinta Estrella 186 (4) 7.0350°N, 73.6416°W 102 ERP and FLMJ
06–2014 San Gil Santander Vereda San José, farm Villa Mandarina 32 (1) 6.5214°N, 73.1075°W 1,584 This study
08–2014 Barrancabermeja Santander Universidad de La Paz, Reserva Santa Lucía 196 (4) 7.0818°N, 73.7490°W 104 ERP
11–2014 Guaca Santander Vereda Quebradas 32 (1) 6.9146°N, 72.8711°W 2,928 This study
Ramos et al.—Distribution and conservation of Andinobates virolinensis.

04–2015 Páramo Santander Vereda Juan Curí 10 (2) 6.3683°N, 73.1718°W 1,414 CHJ
06–2015 Sabana de Torres Santander Reserva Natural El Cabildo Verde 88 (3) 7.3521°N, 73.4992°W 193 This study
07–2015 San Vicente de Chucurí Santander Vereda La Lizama, Bioparque El Arboretum 162 (3) 7.0767°N, 73.5383°W 257 This study
09–2015 Ríonegro Santander Llano de Palmas, vereda La Honda 72 (4) 7.2458°N, 73.2102°W 845 FLMJ and CHJ
09–2015 El Carmen del Chucurí Santander El Centenario, vereda Los Andes 98 (2) 6.6896°N, 73.6154°W 502 This study
10–2015 Charta Santander Vereda La Rinconada, stream El Juncal 16 (1) 7.2858°N, 72.9641°W 2,119 FLMJ
11–2015 Matanza Santander Vereda Vulcaré, stream Vulcaré 32 (1) 7.3407°N, 72.9932°W 1,597 FLMJ and CHJ
11–2015 Guaduas Cundinamarca Vereda Chipauta 24 (1) 5.0654°N, 74.5632°W 1,636 FLMJ
12–2015 El Colegio Cundinamarca El Triunfo, Hacienda Misiones 56 (2) 4.5441°N, 74.4375°W 1,508 FLMJ
12–2015 Albán Cundinamarca Vereda El Garbanzal, Hacienda Rancho Grande 12 (1) 4.8733°N, 74.4434°W 2,098 FLMJ
12–2015 La Vega Cundinamarca Vereda San Juan 24 (1) 4.9791°N, 74.3214°W 1,528 FLMJ
03–2016 Charalá Santander Reserva Nuestro Sueño 20 (2) 6.2912°N, 73.1904°W 1,745 CHJ