Tropical amphibians face a severe decline crisis with ca. 35% of species being currently threatened in the Neotropics. We selected 16 endangered-hylid species and used species records to model their potential geographical distribution for the continental Neotropics. We found that there is a strong influence of slope in hylid geographical distribution that interacts synergistically with maximum rainfall and temperature changes over the year. We identified some intersecting areas of species overprediction along southern Neotropics, which could be important for future biological surveys searching for undescribed microendemic hylid species. Nine of the 16 studied hylids have small geographic ranges with only 25% of its potential distribution being currently protected in the Neotropics. The remaining seven species are still in need of additional conservation areas to ensure the protection of at least 25% of its original distribution range in Mesoamerica. Most Neotropical endangered hylids have only the periphery of their distribution protected with its core distribution outside protected areas. These species may be especially threatened because they now occur in small, isolated subpopulations due to habitat fragmentation and loss. We suggest that conservation efforts for Neotropical hylids should be focused on restricted-range species and in the establishment of additional conservation area networks in Mesoamerica. Remaining habitats for threatened hylids need to be managed as a coordinate network including site-scale and landscape-scale actions to buffer the extinction-driven process caused by inbreeding, genetic drift, and demographic stochasticity.
Introduction
Neotropical anurans are a key component of biodiversity because they are an integral part of terrestrial and aquatic ecosystems linking these environments and playing important roles in species interaction networks, as they feed upon plants and algae, prey upon small animals, and serve as food for larger predators [1]. The Neotropics harbor ca. 3046 amphibian species (2065 in South America and 685 in Mesoamerica; [2]) and 35% of anuran species are current threatened with extinction, being classified by The World Conservation Union (IUCN) as “critically endangered”, “endangered” or “vulnerable”. This percentage increases up to 41% if we add species considered to be “near threatened” [3] without taking into account rare species classified as “data deficient”. Furthermore, relative to other animal groups, an outstandingly high proportion of amphibians are in higher threat categories [4, 5]. Amphibian populations are also declining worldwide and such high threats at the population and species level is causing growing concern [678–9].
The leading factors that threaten amphibians and determine their population declines are habitat fragmentation and loss, which affect amphibians through population isolation, inbreeding, edge effects, and the disconnection between aquatic and terrestrial environment (also known as habitat split) which are key systems for amphibian reproduction [2, 6, 8, 10]. Amphibians are also threatened by climate shifts and increasing ultraviolet-B radiation [7, 11], introduction of alien species [12], and fungal diseases [13]. The later is particularly important in the Neotropics given that Chytridiomycosis infection, caused by the fungus Batrachochytrium dendrobatidis, has been responsible for decline of many populations even in undisturbed environments in this particular region [7, 13].
In the face of such a drastic scenario of population decline and species extinctions, the necessity of high-quality accurate data on amphibian geographic distribution from which to derive reliable science-based studies is quite obvious. However, our knowledge about biodiversity remains inadequate and plagued by the so-called Wallacean shortfall [14, 15]. This refers to the fact that for the majority of taxonomic groups geographical distributions are poorly understood and contain many gaps. This is especially problematic in the Neotropical region, in which species records are fairly sparse and highly uneven [16, 17]. For Neotropical frog species, in particular, few data on geographical distribution is linked to their huge diversity, associated to the existing of highly specialized species that occur in very specific microhabitats. The low number of taxonomists relative to the number of species to be studied strengthens even further the lack of availability on frog distribution across this realm. To a certain extent, the lack of field records may be overcome by summing expected distributions of species obtained through ecological niche modeling [18]. Species distribution models attempt to provide detailed predictions of distributions by relating presence of species to environmental predictors, providing researchers with novel tools to explore questions in ecology, evolution, and conservation [18]. Ecological niche modeling while relating species locality records and environmental coverage data also provides informative biogeographical data for poorly known tropical landscapes [19]
A wide range of methods has been used for predicting species potential geographic distributions [20], but despite their frequent use, the number of occurrence records available for individual species from which to generate predictions is often limited. Records are even scarcer for rare species that are difficult to sample and limit the availability of locality records. This, in turn, affects the performance of species distribution models, given that they seem to depend on sample size [20]. Due to the difficulty to obtain rigorous records of species absences, presence-only data are effective for modeling species distributions. This kind of data is the raw material of maximum entropy machine-learning methods, which were designed to predict species distributions under current environmental conditions, and have demonstrated to be one of the highest performing methods when ranked against other approaches [18].
Methods for predicting species potential distribution across different geographical scales have been applied also in conservation planning exercises (e.g. [2122–23]) and invasive species ecology (e.g. [24–25]). The results of these studies, coupled with high threat levels imposed to amphibians, clearly highlight the need for creating effective strategies to maximize conservation efforts for these vertebrates and call for an urgent evaluation of existing ones [26]. To date, natural protected areas seem still to be the best option for safeguarding species across multiple spatial scales as the in situ conservation of viable populations in natural ecosystems is widely recognized as a fundamental requirement for the maintenance of biodiversity [27–28]. However, to attain such a thing we need to know how much biodiversity is currently protected and where new protected areas should be established to move toward complete coverage [29]. We call this approach a Gap Analysis, defined as a planning approach based on assessment of the comprehensiveness of existing protected-area networks and identification of gaps in their coverage (see [27]). Several gap analyses at regional and continental scales revealed that coverage of biodiversity by existing networks of protected areas is actually inadequate (e.g. [23, 30]. Nevertheless, no study so far has addressed the effectiveness of the Neotropical network of protected areas in representing threatened amphibians (but see [31]), although a comprehensive set of areas for the conservation of threatened anurans has been recently proposed for the entire region (see [3]).
In this study we focused our efforts in Neotropical threatened hylids (Amphibia: Hylidae) because they are the largest anuran family in this realm having 587 threatened species in continental Neotropics [5], and they also hold the best individual species records for this region. Our objective was, therefore, twofold: (1) we aimed, by modeling species ecological niches, to predict endangered-hylid potential geographic distributions across the continental Neotropical region and their relation with topographic and climatic variables; (2) we evaluated the effectives of the network of protected areas in representing these threatened species along the continental Neotropics (an optimistic estimate), and along Mesoamerica (a conservative approach).
Methods
Scope of study
We centered our analyses in the continental Neotropics (Mesoamerica and South America) which are composed by 17 countries (Belize, Bolivia, Brazil, Colombia, Costa Rica, Ecuador, El Salvador, French Guiana, Guatemala, Guyana, Honduras, Mexico, Nicaragua, Panama, Peru, Suriname and Venezuela) spanning a total area of 16.133.914 Km2 (Fig. 1). On the one hand, the Neotropics encompass six megadiversity countries and more than 10,000 vertebrate species [32], harboring more than a half of the World's amphibians [2]. It holds the largest remaining wilderness areas in the World [33], and includes most of the tropical ecosystems still offering significant options for successful broad-scale conservation action. On the other hand, it also supports about 462.409.877 people with a mean rate of population growth reaching 1.48% [34]. This entails a huge human footprint on natural resources altering patterns of biodiversity and ecosystem services within this region [35].
Species occurrence data
In order to derive species distribution models we selected a priori endangered-hylid species (sensu [5]) given that species with restricted ecological niches have smaller geographic ranges (such as endemics) providing more robust and precise niche distribution models [36373839–40]. We started our study with a dataset of species geographical records obtained from HerpNet ( http://www.herpnet.org/), CONABIO ( http://www.conabio.gob.mx/remib/doctos/remib_esp.html), WWF ( http://www.worldwildlife.org/wildfinder), the Global Amphibian Assessment ( http://www.globalamphibians.org), and Species Link ( http://splink.cria.org.br). We choose 16 endangered-hylid species (sensu [5]), being six of genus Plectrohyla (which have 41 endangered species in the Neotropics), one of genus Hylomantis (which have 8 endangered species in the Neotropics), two of genus Isthmohyla (which have 14 endangered species in the Neotropics), one of genus Ptychohyla (which have 13 endangered species in the Neotropics), two of genus Duellmanohyla (which have 8 endangered species in the Neotropics), one of genus Charadrahyla (which have 5 endangered species in the neotropics), one of genus Bromeliohyla (which have 2 endangered species in the Neotropics), and two of genus Agalychnis (which have 6 endangered species in the Neotropics). All these species had at least 19 independent locality records. This produced a dataset of 551 individual records with a mean number of records per species equal to 32.4, ranging from 19 to 58 (Appendix 1).
A typical problem in potential distribution modeling is that species geographical data are often presence only, rather than presence-absence, resulting in a lack of information about species that have been searched in the field, but not found. One way to mitigate this limitation is to use species records to model expected geographical distribution in the study region [41]. The geographical distribution of species are most accurately predicted in multi-dimensional environmental space using ecological niche modeling on the basis of climatic and topographic variables [42]. These variables, in turn, have a potential influence on the distribution of amphibians across the Neotropics [43]. We assumed that each species has a unique distribution within an environmental space determined by its genetic constitution and its physiological requirements [44]. Species ecological niche distribution is also constrained by ecological interactions (sensu realized niche [45]). Hence, the challenge of identifying distributional areas for species requires two conditions to be met: favorable abiotic conditions and favorable biotic factors. As highlighted by Papes and Gaubert [46], a third condition – the geographical accessibility (i.e. landscape configuration, dispersal abilities of species), both historical and actual, are also determinants of the actual presence of species (see also [47]).
Ecological niche distribution modeling
We predicted the geographical distribution for the 16 endangered-hylid species based on ecological niche models generated by MaxEnt software version 3.2.1 [42, 48]. MaxEnt estimates species distributions based on presence-only occurrence data by finding the distribution of maximum entropy, subject to the constraint that the expected value of each environmental variable under this estimated distribution should match its empirical average [48]. The obtained model reveals the relative probability of a species distribution over all grid cells in the defined geographical space, in which a high probability-value associated to a particular grid cell indicates the likehood of this cell having suitable environmental conditions for the modeled species [18].
We obtained 19 environmental variables from the WorldClim database ( http://www.worldclim.org/), which were interpolated from global climate datasets at a resolution of 0.01° or 1 km2 approximately [49]. We also used additional spatial layers of topography, slope and topoindex from 0.01° U.S. Geological Survey's Hydro-1K ( http://edc.usgs.gov/products/elevation/gtopo30/gtopo30.html).
All this totaled 22 layers of topographical and environmental variables (Table 1). All these layers were clipped to an area circumscribed between 32.72 N to −33.74 S and 118.40 E to −34.79 W, which included the countries of Belize, Bolivia, Brazil, Colombia, Costa Rica, Ecuador, El Salvador, French Guiana, Guatemala, Guyana, Honduras, Mexico, Suriname, Nicaragua, Peru, Panama and Venezuela.
Table 1.
Codes for 22 environmental and topographic variables layers used to model amphibian's distribution.
We run MaxEnt under the “auto-features” mode as suggested by Phillips and Dudik [42]. The use of default settings is reasonable given that its use has been validated in studies over a wide range of species, environmental conditions, individual species records, and in cases in which sample selection bias occurred (see [42]). We configured the machine-learning algorithm to use 75% of species records for training data set and 25% for testing the model. We also selected the logistic output format because it is robust to unknown prevalence, being also easier to interpret as the estimated species probability of presence given the constraints imposed by environmental variables [42]. In this case, grid cells with a small logistic value are predicted to be unsuitable or only marginal suitable for the studied species given their assumed ecological niche [42]. We reclassify each species map using the 10 percentile training presence of the logistic threshold of the distribution model. MaxEnt determined the heuristic estimate of relative contributions of each climatic and topographic variable in each species distribution model and we performed a Principal Component Analysis (PCA) to reduce dimensionality and obtain a smaller number of species groups based on the percentage of contribution delivered by each variable, using Statistica 6.0 software [50].
Ecological niche modeling cannot include aspects such as biogeography or species natural history, ignoring if some species may have failed to disperse due to geographical barriers or were excluded from an area due to resource competition, for instance [42]. We selected, therefore, only those models with AUC values above 0.75 in the training data (as suggested by Elith [18]) and those in which the test data curve (in the ROC sensitivity–specificity plot – see [48]) overcame the random-prediction curve. Based on this, we assumed that those models were robust enough to predict species presences included in our sampling data. As an example, an AUC = 0.75 means that in places where a species is present, in 75% of cases the predicted value will be higher than where the species has not been recorded. Moreover, when evaluating AUC as the correct ranking of random suitable sites versus random unsuitable sites, a model with AUC = 0.75 ranks the suitability of the site correctly in 75% of the cases (see [>20]).
Table 2.
Number of registers, AUC values of ecological niche geographic distribution models and the heuristic estimate of relative contributions for most important variables for 16 endangered hylids in the Neotropics. See methods for further details.
Current protected areas and their effectiveness in species conservation
As a final goal, we assessed the conservation status of potential distributions for the 16 studied species. We calculated the proportion of species potential distribution currently covered by the Neotropical protected-area network for all studied species using data available from the World Database of Protected Areas [51] at a resolution of 0.5° or 3025 km2 approximately. Although the IUCN recognizes six categories of protected areas, we focused our analyses to categories I to IV, i.e. those which are managed primarily for biodiversity conservation [52]. We performed calculations in ArcGIS 3.2a [53] in which we masked out the areas outside of designated reserves, which allowed for evaluation of the extent of species potential geographic range which is under protection, and that in which no protection exists. Here, we considered as protected only those grid cells having ≥ 25% of their surface filled by natural reserves (see [54]). In conservation studies, analysis of range-map data at inappropriate resolutions may lead to optimistic estimates of species representation in reserves [55]. Given that only Hylomantis lemur is reported to be marginally distributed outside Mesoamerica (in the Darién region, just across the border to Colombia), we also assessed the conservation status of species potential distributions under more conservative models, in which we used only predictions made within the limits of Mesoamerica, and in which species probability of occurrence was between 90−100%.
Results
Relative contribution of variables to species distribution models
The most important variables contributing to 52% of species distribution models were slope (Mean=29.4%, SD=16.4), precipitation of wettest month (bio13; Mean=12.3%, SD=15.7) and temperature seasonality (bio4; Mean=10.6%, SD=7) (Table 2). Based on the percent contribution of each of the 22 variables to each species distribution models we identified two species groups according to the two first factors of the PCA, which explained 69.5% of variance (Table 2). The first group is composed by Duellmanohyla uranochroa, Isthmohyla rivularis, Isthmohyla tica, H. lemur, Plectrohyla glandulosa, Plectrohyla pentheter (Fig. 2A) and Ptychohyla leonhardschultzei (Fig. 2B); whereas the second harbors the species Charadrahyla chaneque, Duellmanohyla ignicolor, Plectrohyla cyclada, Plectrohyla guatemalensis and Plectrohyla sagorum (Table 2).
Table 3.
Predicted geographic range distribution attained by the application of niche-based models to endangered hylid species in the Neotropics and only in Mesoamerica. Protected range and percentage of protection were calculated by overlapping spatial locations of Neotropical protected areas (IUCN I–IV). Predicted range distributions and their percentage of protection, in Mesoamerica, are more conservative given that only grid cells having 90−100% probability of species occurrence were considered. See methods for further details.
Species potential distribution models
Among evaluated hylids, 62.5% of species had small potential geographic distributions with range values being under the mean predicted range (Fig. 3A, Table 3): P. pentheter, I. tica, B. dendroscarta, Agalychnis annae, I. rivularis, D. uranochroa, P. leonhardschultzei, H. lemur, P. glandulosa and P. cyclada. Most endangered hylids have relatively small geographic ranges based on their potential distribution (mean 352,650 km2; minimum: 102,625 km2, maximum: 1,140,806 km2), encompassing 3% or less of the Neotropics (Table 3, Appendix 2). When potential distributions were restricted to grid cells in Mesoamerica, the results were similar, although predicted ranges were even smaller, as expected (Fig. 3B, Fig. 4).
Effectiveness of the Neotropical network of protected areas
Most cells with similar environmental conditions have ca. 35% of its total area covered by protected areas in the Neotropics (Fig. 3C). This means about 4235 km2 of area covered in each of these cells, ranging from 0 to 12,100 km2. When potential distributions were restricted to Mesoamerica, most cells presented only 10% (about 1210 km2) of their area protected by natural reserves (Fig. 3D).
Within the 557 cells having $ge;25% of its surface protected, all studied species had at least 13% of their potential niche distribution represented. We found that ten species have more than 25% of their potential range current protected, but six are still in need of additional area to be protected in at least a quarter of its potential distribution range (Table 3). Mean proportion of geographic range protection was ca. 29% (ranging from 13 to 42%) and nine species were under this value. The most protected species was D. uranochroa, with 42.34% of its range included in protected areas, whereas less protected were P. sagorum and P. pentheter, having 17.57% and 13.31%, respectively, of their potential distribution located inside reserves (Table 3). Most species had only the edge of their geographic range included in protected areas, but only few species had the core of its distribution protected by natural reserves (see Fig. 5).
Fig. 3.
(A) Number of grids per area protected in the Neotropical region, (B) number of grids per area protected in Mesoamerica region, (C) number of species per geographic range class (measured in Km2) in the Neotropical region, and (D) number of species per geographic range class in Mesoamerica region.
When conservative models were evaluated (i.e. those in which only grid cells having a 90−100% probability of species occurrence in Mesoamerica), results were somewhat different. We find that eleven species are in need of additional cells to be protected in at least 25% of its potential distribution in Mesoamerica. Moreover, the species I. rivularis had no part of its range included in protected areas. Other species, such as P. leonhardschultzei, P. pentheter, P. cyclada, B. dendroscarta and P. glandulosa had less than 10% of its potential geographic distribution protected in this region. Conversely, four species (D. uranochroa, I. tica, H. lemur and A. annae) were more protected under this conservative scenario.
Discussion
This is one of the few studies applying niche-based models to predict potential geographic distributions of endangered hylids in the continental Neotropics. It is also the first attempt to evaluate the effectiveness of the Neotropical network of protected areas in representing and safeguarding hylids. Our results demonstrate that the extent of occurrence of ecological niche of some Neotropical endangered hylids may be much larger than the current species distribution reported by international conservation agencies [56], albeit the proportion of their geographic range currently under protection is still low for most species, especially if their potential distributions are restricted to Mesoamerica.
Fig. 5.
Total (summed) potential geographic distribution of the 16 endangered-hylid species evaluated (shown in red) and the network of Neotropical protected areas (shown in green).
For lack of better alternatives, range maps and estimates of species geographic ranges based on niche-modeling techniques have become the baseline data for many broad-scale analyses in ecology and conservation biogeography [15, 57]. In this study we found that climate and topography exert a great deal of influence on threatened hylids' distribution. Such influence is not as simple as reported by literature (see [43]). It seems that there is a strong influence of slope (more than elevation) that interacts synergistically with rainfall and temperature to determine species geographic distribution. Hence, the relation between hylid species occurrence with climatic variables is not as simple given that the utmost variables determining species potential distributions in this study were maximum precipitation and temperature change over a year. Taking that into consideration at a microenvironment scale, some important variables influencing amphibian ensembles are canopy cover, understory density, leaf litter cover and temperature [10, 58]. This gives us an insight about how drastic could be climate change effecting threatened Neotropical hylids distribution at different spatial scales.
It is also known that extent of occurrence maps obtained by niche-based models can overestimate species current distribution and geographic range sizes, biasing broad-scale ecological patterns and their correlates [57]. Following current distribution maps of the Global Amphibian Assessment [56], all 16 studied species have geographic distributions historically restricted to Mesoamerica. Nevertheless, all potential distribution models seem to present a certain degree of over prediction in South America (Appendix 2). This does not mean that not all studied species necessarily occur at overpredicted areas. The environmental conditions of a predicted ecological niche could be represented in multiple areas along a geographical space [45]. However, species do not use all suitable ecological niches available along the geographical space, since it is constrained by species behavior, dispersal ability, and inter and intra-specific interactions that take place at local and landscape scales [18, 59]. This is the main reason why we have built more conservative species distribution models, restricted to Mesoamerica. In that case, the probability of occurrence of a given hylid species is indeed high and, therefore, the degree of geographic range overestimation may be low – reflecting actual species distributions and some particular areas needing more detailed surveys in order to confirm the occurrence of species. In fact, when modeling species actual distributions (which are based on real species occurrence data [47]) over-predicted areas could indicate the occurrence of some phylogenetically close-related hylids which are expected to have similar ecological niches. Overlapping areas of overprediction in South America could be themselves extremely important for the discovery of unknown distributional areas and undescribed species (see [19]), which, in turn, could be as threatened as the modeled ones due to their microendemicity patterns.
We suggest the use of MaxEnt (instead of other presence-only methods [18, 48, 60]) to assess the effectiveness of protected areas in representing endangered species because: (1) this software constrains predicted species ranges reducing and avoiding commission errors (i.e. when a model predicts the presence of a given species in particular areas, although it is known that this species is not present there [48, 61]). Commission errors (or false positive rate) could lead to erroneous conservation decisions focusing financial investments and management efforts in non-priority areas; (2) Although MaxEnt generates high omission errors or false negative rate (i.e. when a model predicts the absence of a species in particular areas, though it is known that this species is indeed present there [48, 61]), such errors are preferable when models are conceived for conservation purposes [62]. Loiselle et al. [62], for instance, demonstrated that using distribution models that minimize false positives (such as MaxEnt's models) for well known taxa, priority areas highlighted for conservation matched up those previously selected by experts in biogeography, ecology and taxonomy.
Implications for conservation
When predicting species distributions for the entire Neotropics, we found that six hylids (P. pentheter, P. sagorum, C. chaneque, D. ignicolor, Agalychnis moreletii and P. leonhardschultzei) are still in need of additional conservation areas to ensure the protection of 25% of its potential distribution range. Most important however, was the finding that P. pentheter while holding the smallest potential distribution range (102,625 Km2), also have the smaller percentage of its range (13.3%) included in protected areas. Restricted-range species, such as P. pentheter, are worthwhile given that they usually tend to be endemic. Several global conservation assessments highlight endemic species as a worthwhile conservation goal, e.g. the Global 200 ecoregions [63], and the Biodiversity Hotspots [32]. Some studies also pointed out that endemic species also provide a useful guideline for identifying conservation priorities at a global or regional scale [9, 64]. We suggest, therefore, that Neotropical hylids with restricted ranges should receive marked attention of conservationists and policy makers, especially if they are threatened of extinction, like P. pentheter.
Under more conservative models that predicted species geographic range within Mesoamerica, the number of species needing additional areas for the protection of at least a quarter of its potential geographic range increased up to ten. We found that most Neotropical endangered hylids have only the periphery of their distribution protected, and this aspect is critical given that human population growth is much higher around protected area edges than in other rural areas [65]. When predicted distributions of species were restricted to Mesoamerica, mean percent range protected decreased from ca. 29% to ca. 23%. For the species I. rivularis, in particular, range protection fell from 37.6% to 0%. Species like that have most of their protected range located in South America, but as mentioned before, to date we have no data on the occurrence of these species at sites predicted by our models. Many species may be actually threatened because they now occur in small and isolated subpopulations due to habitat fragmentation. Whereas the sites where they survive need to be managed as a coordinated network, the lack of protection of species core distribution usually implies in protecting populations threatened by several ecological and genetic processes like inbreeding, genetic drift, and demographic stochasticity. In the longer term, site-scale actions for effective protection of these species will likely need to be supported by broad-scale approaches, such as the restoration of connectivity. Recently, Loyola et al. [3] proposed priority sets of Neotropical regions that should be sufficiently covered in a reserve system to protected threatened anurans with distinct reproductive modes. Most of their proposed areas for the conservation of species requiring aquatic habitats for their reproduction are found in Mesoamerica. The results of our study, while being attained at a finer spatial scale, corroborate and push even further the need of effective natural protected areas in this region if endangered anurans that require aquatic habitats – which are the majority of species with reported population declines (see [26, 66]) – are meant to be protected.
Niche-based distribution modeling is an innovative analytical approach to evaluate the effectiveness of protected areas, especially in regions lacking comprehensive databases of species distribution. Combination of niche-based distribution modeling and reserve selection algorithms is also a promising approach [67–68]. It works as an effective tool that should be applied in systematic conservation planning to identify and interconnect priority regions, particularly those already covered by natural protected areas [69]. Moreover, it is an efficient tool for identifying gaps in actual reserve systems, especially when it highlights regions that surround protected areas and, therefore, complement proposed conservation plans [6970–71]. Although amphibians and reptiles are not commonly used as biodiversity surrogates in systematic conservation planning [22], recently, niche-based distribution models combined with reserve selection techniques were used to pinpoint conservation priorities in India [22] and Mexico [72]. These authors generated models to different taxa to find overall congruences among different taxonomic groups. Such congruence is obviously attractive given that it indicates that priorities identified for a particular species subset would be effective for non-target ones. In a recent essay, Bode et al. [73] found that funding allocations were less sensitive to choice of taxon assessed than to variation in economic costs of land acquisition and species threat. These results strengthen confidence in decisions guided by single taxonomic groups [73].
Finally, among the leading factors that threaten amphibians, habitat loss, habitat fragmentation, and habitat split are the most important and, perhaps, the major causes of species extinction in general [2, 67–8]. All these factors are thought to be minimized within a network of natural protected areas, which remains as the cornerstone of conservation strategies. Loucks et al. [28] have demonstrated that, globally, species endemism, species richness, and to a lesser extent threatened species explained better the global pattern of protected area coverage. Our results, by mapping threatened species potential geographic distribution, revealed that we need more protected areas in Mesoamerica contributing to other studies that have highlighted this for other taxonomic groups such as amphibians and reptiles [3, 8, 23, 74], and carnivores [54, 75]. Given the rapid ongoing transformation of habitats worldwide, proactive attitudes are imperative and uncertainty cannot be used as a pretext for not performing researches or not implementing conservation actions [44]. Besides the inherent uncertainties associated with field data, geographical databases and niche-modeling algorithms; niche-based distribution models have a major potential use in ecology, biogeography, conservation biology and policy that should be better explored. Gaps in geographic range protection presented here helps to pinpoint were conservation assessments should be focused to ensure the persistence of endangered hylids in the Neotropical region.
Acknowledgments
We thank two anonymous reviewers for their comments on an earlier version of this manuscript. C. González-Salazar helped with the environmental layers to MaxEnt models. F. Cassemiro provided us some species individual records. RDL is funded by CNPq (grant n° 140267/2005-0) and JNU-C is funded by DGAPA-UNAM postdoctoral fellow.
