Also investigate how the potential distribution of this species changes with the addition of new records over the decades (decade effect). Assuming that (1: hypothesis of the effect of the decade) the addition of new occurrence records over time increases the potential size of the species distribution; and (2: Wallacean distance hypothesis) over the years, the new points added are increasingly distant from the research centers. Considering the geographic knowledge gap of Histiotu velatus, our objective is to report a new record of this species and estimate its potential distribution in South America through ENMs. For this, we compiled records of occurrence of species, selected from 1900 to 2015. We used 19 bioclimatic variables available in the WorldCLim database to estimate the potential distribution of the species and we used three modeling algorithms: Maximum Entropy (MXT) Random Forest (RDF) and Support Vector Machine (SVM). We selected the main bat research centers in Brazil, using the Lattes platform for the Wallacean distance hypothesis, using the Euclidean distance calculation. To test the hypothesis of the decade effect, we used beta regression analysis, taking conservative and non-conservative approaches. The results showed that the predicted area expanded and retracted over the decades, with an improvement in the accuracy of the models with the addition of new data. Most of the records are located in the southeastern region of Brazil, but the algorithms predicted areas in countries where there were no records. Only the conservatism approach has had a positive relationship over the decades. The distance from new points does not increase over the years of research centers.

Species distribution database and data treatment

We compiled occurrence records of H. velatus available from SpeciesLink (http://www.splink.org.br/index?lang=pt) and GBIF (https://www.gbif.org/). We supplemented our geographical distribution database with records available in scientific articles using the following search code in the Web of Science platform: “bat*” OR “species list” OR “Histiotus velatus” OR “H. velatus”. We selected only the occurrence records since 1900 because the original data were incompatible with the range of the environmental dataset. Furthermore, we excluded the following records: (1) undated occurrence records; (2) records without coordinates; and (3) outside the Neotropical region. Therefore, to investigate the effect of new occurrences over the years, we split the data into eight portions: (1) 1900 to 1950; (2) 1900 to 1960; (3) 1900 to 1970; (4) 1900 to 1980; (5) 1900 to 1990; (6) 1900 to 2000; (7) 1900 to 2010; and (8) 1900 to 2020 with the addition of the new occurrence record localized in Goiânia city.

Environmental variables and Ecological Niche Models (ENMs)

We used 19 bioclimatic variables (resolution of 9.4 x 9.4 km) for the entire Neotropical realm, available in the WorldClim database (http://www.worldclim.org/). These variables are derived from monthly temperature and precipitation values sampled throughout 1960-1990. Also, these data are often used in ecological modeling techniques to estimate the potential distribution of species (e.g. Lee et al. 2012; Lisón and Calvo 2013; Sattler et al. 2007). To reduce multicollinearity in our dataset, we performed a Principal Component Analysis (PCA; Legendre and Legendre 2012) and used the eigenvalues as environmental variables. Then, we selected only the axes that represent an explanation equal to or greater than 95% (De Marco and Nóbrega 2018).

We fit models using three algorithms: Maximum Entropy (MXT; Phillips et al. 2017; Phillips et al. 2004), Random Forest (RDF; Prasad et al. 2006) and Support Vector Machine (SVM; Guo et al. 2005). RDF and SVM algorithms require species’ absence data, but we this data was not found for H. velatus in the literature. Therefore, we created 50 pseudo-absences based on an environmental envelope to allocate pseudo-absences only in places considered unsuitable for the occurrence of H. velauts (Engler et al. 2004). In the case of MXT, models are fitted by differentiating between occurrence records and a 10,000 background points randomly sampled throughout the study area.

We evaluated ENMs using a geographical partition (Muscarella et al. 2014, Roberts et al. 2017). We divided the study area as a checkerboard, which splits the occurrence data in two datasets, and selected each dataset alternately to fit and evaluate. This step allows to evaluate model predictive capacity, as the geographical partition reduces the spatial correlation between datasets used to fit and evaluate the models. Then, we measure model predictive capacity by its value for True Skill Statistics (TSS), True Positive Rate (TPR) and True Negative Rate (TNR). This procedure is considered appropriate in studies on geographic distributions of species (Allouche et al. 2006).  

We converted the suitability models into presence and absence maps using a threshold at which the sum of the sensitivity and specificity is highest (Allouche et al. 2006). Then, we produced assembled maps using the sum of the binary maps derived from the three algorithms. We used the ENMTML package in software R for all modeling procedures (Andrade et al. 2020; https://github.com/andrefaa/ENM_TheMetaLand).

Research centers data

Brazil is the second country with the highest bat richness; however, all of its biomes have a lack of information on the occurrence of species distribution (Bernard et al. 2011). We selected the main research centers that are developing or have developed surveys about bats in Brazil. For this, we conducted a search by topic in the Lattes platform (http://lattes.cnpq.br/) using the keyword “Quiroptera” (in Portuguese). We chose only those researchers that fall in one of the CNPq’s Productivity Researchers categories: 1A, 1B, 1C, 1D and 2. Furthermore, we established as criteria: (1) research projects about bats; (2) published articles about bats; and (3) academic guidance in bats studies. Included researchers present at least two of these three criteria. In situations in which researchers participated in more than one research center during their career, we choose the location where those professionals spent more time working with bats. We used the Google Earth Pro software to consult the geographic coordinates of the research centers.

Statistical analyzes

To test the decade effect hypothesis, we performed beta regression analysis (Ferrari and Cribari-Neto 2004) between the number of records over the decades and the proportion of predicted areas, assuming conservatism and non-conservatism approaches. The conservatism approach considers only the areas predicted by all three algorithms, whereas the non-conservatism approach considers all the areas predicted by any algorithm. We chose the beta regression analysis because our response variable is restricted to a range of 0 to 1. We performed this analysis in the betareg package in software R (Cribari-Neto and Zeileis 2010).

For the Wallacean distance hypothesis, we calculated the Euclidean distance between each occurrence record to the closest research center using the raster package in the R software (Hijmans et al. 2020). In addition, to reduce a possible forced relationship caused by the excessive number of records, we performed a Weighted Linear Regression considering the total of distances calculated for each year as the weight. Then, we related the maximum distance obtained per unit of time to its respective year. We used the highest values observed per year to find out if further away areas from research centers are sampled over time. We also performed the analysis in the R software, using the lm function of the stats package (R Core Team 2020).