Modeling the potential distribution to present and future of the poorly known species Xenohyla eugenioi Caramaschi, 1998 (Anura: Hylidae) with findings about its distribution, natural history, and conservation

. The distribution of amphibians is conditioned by historical factors and ecological drivers. Thus, Ecological Niche Models are important tools to provide information on the potential distribution of the species and determine where they will be requirements in future. Mainly, rare species or with restricted areas of occurrence, because this information are essential to understanding of their living area, microhabitat use, or natural history, serving as a basis for preservation actions. Xenohyla eugenioi is a poorly known species, restricted to transition areas of Caatinga and Atlantic Forest from Minas Gerais to Sergipe. This species exhibits little biological information available and few specimens housed in museums, which difficulty to plan conservation strategies. Here our aim was modeling the current and future distribution and discussing about conservation of Xenohyla eugenioi. We searched for occurrence records through literature and scientific collections data. For the future (2071-2100), we used ensemble models from three algorithms (CTA, GLM and ANN) to two shared socioeconomic pathways (SSP370 and SSP585 scenarios). This species has approximately 650 km in straight line between the extreme localities, occurring mainly in bromeliads near to water bodies. The Ensemble method indicates the most suitable areas of occurrence were over ecotonal range between Caatinga and Atlantic Forest and our projections have showed suitable conditions to highlands (up to 1.000 m). However, in the future, is expected total erosion of the X. eugenioi populations, due climatic changes, which reinforce the caution to conservation of this poorly known species and necessity of studies about its ecology, natural history and distribution. Moreover, we hope that this work can contribute to the discovery of new records, characterizing the narrower niche space than this species may actually inhabit.


INTRODUCTION
Understanding the processes and mechanisms that influence species diversity patterns is a widely studied and discussed issue in macroecology (Ricklefs & Schluter, 1993).The distribution of species is conditioned by historical factors, such as the dynamics of geographical barriers (Hoo-rn & Wesselingh, 2011).Furthermore, ecological drivers such as the set of environmental conditions provide a favorable habitat for the establishment of species (Soberón, 2007).For example, studies about resources used by anurans have shown how the biotic (e.g., calling sites and prey availability) and abiotic (e.g., temperature and humidity) factors bias their occurrence (Bertoluci & Rodrigues, 2002;Xavier et al., 2015;Neves et al., 2020;Ceron et al., 2022).Understanding the distribution patterns of species provides an assessment of the risks to which it is subjected and is an important tool for determining priority areas for conservation (Guisan et al., 2013).
Around 25% of amphibian species do not have any records within protected areas worldwide (Nori et al., 2015;Oliveira et al., 2017).In addition, many species have an inaccurate distribution and/or contain huge areas not sampled (Bini et al., 2006).These facts make it difficult to dimension the real distribution of species, especially for those rare or with restricted areas of occurrence (Pearson et al., 2007).Moreover, endemic species generally exhibit restricted relationships to specific conditions (e.g., climate and biotic interactions), and many of them are poorly known in relation to their living area, microhabitat use, or natural history (Neves et al., 2018).Therefore, it is essential to study the distribution of these species to serve as a basis for preservation actions.
Ecological Niche Models (ENM) are important tools to provide information on the potential distribution of the species according to suitable environmental conditions used by them (Elith & Leathwick, 2009).Besides, the use of the current environmental information about ecological niches may determine models where the requirements of the species niche in the future may be satisfied (Wiens et al., 2009) and predict range shifts under climate changes (Fordham et al., 2012).Thus, the ENMs have been applied to promote better conservation strategies (e.g., Araújo et al., 2011;Chen et al., 2017), mainly, for poorly known species (Pearson et al., 2007;Siqueira et al., 2009).
There are only eight localities known to X. eugenioi with some wide gaps between them, which may represent a lack of sampling effort or the presence of geographical barriers.The biggest gap is between the municipality of Ipirá, in Bahia State, and the municipality of Pedra Mole, in Sergipe State (Almeida et al., 2015), with approximately 280 kilometers of discontinuity.In addition, habitat loss due to agriculture and cattle raising are potential threats to populations of this species (Almeida et al., 2015).There is little information available on biological data about the species (Caramaschi, 1998;Almeida et al., 2015;Marciano-Jr. et al., 2021), with few specimens housed in museums, which reflect the difficulty to plan conservation strategies for this species.
For this reason and to better understand the distribution of this species, this study aims: (1) to predict a current and future distribution to X. eugenioi using ENMs; and (2) to discuss the conservation and effectiveness of the PAs in cover the potential distribution of the species.

Species occurrences
We used eight occurrence points of Xenohyla eugenioi to fit the models (Table 1).We used seven points available in the current bibliography (Caramaschi, 1998;Napoli & Brandão, 2004;Feio et al., 2006;Almeida et al., 2015;Orrico et al., 2020) and one record from three spec-Table 1. Known localities of Xenohyla eugenioi in "Agreste" region (Caatinga and Atlantic Forest transition area), Brazil, with respectively municipality and locality, state, latitude and longitude in decimal degrees, elevation above sea level and references of the records.States: Sergipe (SE), Bahia (BA), and Minas Gerais (MG).imens housed in the amphibian collection of Universidade Federal do Rio de Janeiro (ZUFRJ): ZUFRJ168486, ZUFRJ168487 and ZUFRJ168488.

Environmental data
We used bioclimatic variables to model the current and future geographic range of X. eugenioi.The climatic data were obtained from CHELSA v2 dataset (Karger et al., 2017), which represents a set of 19 bioclimatic variables based on monthly precipitation and temperature values averaged between 1981-2010 (current time) and 2071-2100 (future time).We extracted bioclimatic layers at 30 arc-sec of spatial resolution and applied a variable selection procedure.More specifically, we passed the pairwise correlation matrix among the bioclimatic layers through a UPGMA clustering analysis to identify which climatic variables exhibit lower levels of multicollinearity (Dormann et al., 2013).We ended up with five climatic variables, namely: Annual Mean Temperature (BIO01), Isothermality (BIO03), Temperature Seasonality (BIO04), Annual Precipitation (BIO12), and Precipitation Seasonality (BIO15).For future projections, we use the CMIP6 climatologies from three general circulation models (GCMs) (GFDL-ESM4, IPSL-CM6A-LR, and MRI-ESM2-0) provided by CHELSA.We selected these models due to their better ability to simulate temperature and precipitation variables we use (Firpo et al., 2022).
For the future, we used two scenarios of shared socioeconomic pathways (SSP): the SSP370 scenario represents high challenges to mitigation and adaptation and SSP585 represents high challenges to mitigation and low challenges to adaptation (IPCC, 2022).

Species distribution models
For the elaboration of Ecological Niche Models (ENMs), we implemented Ensemble of Small Models (ESM, Lomba et al., 2010;Breiner et al., 2015) that is developed for rare species (Lomba et al., 2010).The ESM strategy builds models with predictor subsets, and then, these subsets are evaluated according to a model performance metric obtained through cross-validation (see below).Finally, all subsetted models are averaged into a weighted ensemble model (Breiner et al., 2018).We used three modelling algorithms in our ESM: artificial neural network (ANN, Manel et al., 1999), generalized linear models (GLM, McCullagh & Nelder, 1989), and classification and regression tree (CTA, Breiman et al., 1984).These algorithms were found to be best performing when using presence-only data (Breiner et al., 2015(Breiner et al., , 2018)).Pseudo-absences were randomly generated within the study area and matched the same number of real presences for X. eugenioi.We built the models and predictions 100 times for each algorithm, selecting each time a different 75% random sample of the true occurrences (Lomba et al., 2010;Breiner et al., 2015).Model accuracy was always verified against the remaining 25% using True Skill Statistics (TSS, Allouche et al., 2006).TSS range from −1 to 1, where values above zero indicate a better than random prediction.All analyses were performed in R language (version 3.5.2,R Core Team, 2019) with raster, rgdal, and ncdf4 packages (Bivand et al., 2023;Hijmans, 2023;Pierce, 2023).
We applied the minimal predicted area method (MPA, Engler et al., 2004) to establish a threshold value to convert habitat suitability measures (varying between 0 and 1) into binary maps (0 = absence, 1 = presence).MPA predicts a potential area that is adequately small but still includes the maximum number of true species occurrences.To define the probability threshold, we extracted the value of habitat suitability for each observed occurrence in the current projection map and computed the value corresponding to the percentile 90%.Thereby, the binary distributions encompassed 90% of the real occurrences.After that, we overlap the potential area found in our models with the existent network of Protected Areas of the Brazilian Ministry of Environment and calculated the potential protected distribution of the species.

Extent of occurrence records
Xenohyla eugenioi is known from eight localities (Table 1) occurring near water bodies between the north of the Jequitinhonha River in Minas Gerais state and south Vaza-Barris River in Sergipe state, also taking place in the interior of the state of Bahia (Fig. 1).The records have approximately 650 km between the extreme localities in the south (municipality of Salto da Divisa) and in the north (municipality of Pedra Mole).This species is restricted to the northern Atlantic Forest and the region of its ecotone with the Caatinga and no record is found in Protected Areas, only in "Areas of Permanent Protection" (APPs) under Brazilian Federal Law 12,651/2012 (see below in discussion session; Toledo et al., 2010).

Potential distribution and modeling
The TSS values for X. eugenioi were relatively high (mean ± SD = 0.9844 ± 0.023) indicating good model fits.TSS value for Ensemble model reached higher than the others algorithms (mean = ANN: 0.9957, CTA: 0.9493, GLM: 0.9961, and Ensemble: 0.9962) and thus the following results are based on the Ensemble models and their projections.
The Ensemble binarized projection (threshold = 917.85)indicates the most probable areas of occurrence were over an ecotonal range between Caatinga and Atlantic Forest which extends of the Minas Gerais to Pernambuco states through of extent area of approximately 6.005 km² (Fig. 1).Our Ensemble projection have shown suitable conditions for areas with high altitudes (up to 1.000 m), and the Central-South region of Bahia.Moreover, we found environmental suitability to Per-nambuco state, the only region found above São Francisco River and far from any known record of X. eugenioi (Fig. 1).
Of the total potential area resulting from our models, about 480 km² (7.99%) are sheltered by some Protected Area (PA).These PAs that can house the species are Cha- For the future (2100), the projections showed large contractions in the suitable areas for the species (Fig. 2).Climatic suitability areas passed its range from 0-1 at present to 0-0.378 in both future scenarios (SSP370, and SSP585).Applying the threshold, there will not have environmental conditions for the survival of X. eugenioi according to all GCMs and in both scenario.Thus, it is expected the extinction of the species, due to dramatic climate changes.

DISCUSSION
We presented the first ecological and distribution study of Xenohyla eugenioi integrating literature data and specimens housed in museums to predict the potential distribution currently and in the future, beyond the extinction risk due to climate change.Here, our current projection for X. eugenioi suggests that environmental traits such as temperature and precipitation can explain the occurrence of the species in the ecotonal regions of the Atlantic Forest and Caatinga.Moreover, suitable areas for its occurrence must withdraw in the future (2100) and the species may become extinct as a consequence, according to all projections generated.
The Ensemble projection indicates the species has narrow distribution to ecotonal areas and highlands from branch of the National Park Chapada Diamantina in the Figure 1.Geographic distribution of known records of Xenohyla eugenioi, and total extent area from Ensemble binarized projection (Potential distribution) (6.005 km²).The number of the records follow localities in Table 1.Inset map: South America with states: Pernambuco (PE), Bahia (BA), Alagoas (AL), and Minas Gerais (MG) states.Andrade, H. et al.: Modeling the distribution of Xenohyla eugenioi Pap. Avulsos Zool., 2024;v.64: e202464002 4/8 Central-South region of Bahia state (Figs. 1 and 2).The Chapada Diamantina, the northern part of Espinhaço Mountain Complex, is characterized by the occurrence of endemic taxa with a high diversity of anurans (Garda et al., 2017;Guedes et al., 2020).Thus, we believe to be possible the occurrence of X. eugenioi in this area, because its relationships with high altitudinal and ecotonal areas, as demonstrated through the true occurrence of the specimens (Table 1) and the result of our models (Fig. 1).Furthermore, we suggest expeditions with focal samples where the Ensemble projection shows high environmental susceptibility to its occurrence, which can enable new records and, consequently, more information about their populations and threats.
Nowadays, all records of X. eugenioi are found inside terrestrial bromeliads or perches in shrubs of up to 1.5 m height on edge of streams in areas of permanent protection (APPs) (Caramaschi, 1998;Almeida et al., 2015;Marciano-Jr. et al., 2021).Although approximately 8% of its potential distribution is within Protected Areas, this protection does not currently encompass any record.This information brings up the caution to species conservation because its occurrence is associated to areas with high anthropic pressure, due cattle raising and agriculture (Metzger, 2010;Almeida et al., 2015).Mainly, because the populations of X. eugenioi found here (Fig. 1) are in regions with only 5% to 20% from original landscape (Ribeiro et al., 2011).
In addition to the known impacts on the area of occurrence of the species, the future projections exhibited here did not recover suitable climatic areas for the species in 2100 for all GCMs used, increasing the pressure for the extinction of X. eugenioi.This result match closely to previous estimates indicating loss of climatically suitable areas to anurans from Atlantic Forest (Toledo et al., 2010;Vasconcelos et al., 2018).Thus, we consider fundamental the protection of the land and water over its distribution as suggested by Silvano & Pimenta (2014), mainly, due to the actual pressure of the habitat loss on their populations.Then, we suggest integrated strategies as maintenance of large conservation corridors and restoration of gallery forest (see Tabarelli et al., 2005), which are action can guarantee connectivity among populations and quality of the environment (Toledo et al., 2010).
Ecological research related to distribution, life history, ecological traits, population size, and trends are necessary to understand and assess the real threat situation of the species (Silvano & Pimenta, 2004).This way, the lack of information on biological aspects difficult the understanding of the ecological relationship between X. eugenioi and other species.While is well-documented X. truncata life cycle as a bromelicolous species (sensu Peixoto, 1995) that lives in the Neoregelia cruenta and reproduces in pounds nearby by indirect development (Izecksohn, 1998), the knowledge about X. eugenioi life cycle is still unclear.All literature and our records have found X. eugenioi in terrestrial bromeliads, but the lack of observation of reproduction and juvenile specimens prevents the recognition as a bromelicolous or bromeligenous species (sensu Peixoto, 1995).Fieldwork also made it possible to observe the presence of X. eugenioi in N. cruenta (BBC pers.obs.) and Aechmea sp.(Marciano-Jr. et al., 2021), but it is still soon to confirm as a restriction like it is with X. truncata (Izecksohn, 1959(Izecksohn, , 1998)).This possible strict relationship can reveal a coadaptation process, while understanding the life cycle may elucidate which factors are necessary to X. eugenioi lives in bromeliads.Thus, we hope future ecological questions of X. eugenioi can be measured and answered, which are important to species conservation and management.

Figure 2 .
Figure 2. Ensemble suitability areas projections (range from 0 to 1) for Xenohyla eugenioi based on three climatic models and climate layers, considering an intermediate and high anthropogenic CO₂ emission (SSP370 and SSP585).