Some bats are here: Reducing the Wallacean shortfall of bats in the amazon

Abstract The Amazon rainforest has approximately 23% of its sampled area dedicated to bats, making it one of the least sampled and most diverse regions for bats in Brazil. The lack of sampling results in a lack of knowledge regarding the accurate geographical distribution of bat species. This lack is referred to as the Wallacean shortfall, which should be addressed with primary data obtained from in situ collections. However, the use of Species Distribution Models (SDMs) can help alleviate this gap. The states of Pará and Acre are located in the Brazilian Amazon. So, our objective is to decrease the Wallacean shortfall concerning Amazonian bat species. To achieve this, we provide (i) a list of bat species sampled in the states of Pará and Acre in the last 5 years (2017 to 2022); (ii) the potential distribution of species considered as new occurrences for the region; and (iii) the potential distribution of species classified as Data Deficient (DD) and Near Threatened (NT) according to the IUCN classification. With 96 nights of collection and 129,600 m2h of mist netting, we obtained 75 bat species, with an estimated total of 94.78 species. Additionally, 21 species were considered as range extensions. The Brazilian Amazon region has a vast geographic expanse and few established research centers, resulting in a limited sampling of bats and other biological groups. Furthermore, we draw attention to the significant number of bat species with expanded geographical distributions, with 21 out of the 75 sampled species. This should be a reminder that primary biogeographic data is still necessary for the neotropical region.


| INTRODUC TI ON
The gap in knowledge about the geographical distribution of species, referred to as the Wallacean shortfall, is one of the challenges encountered in implementing conservation strategies effectively (Diniz-Filho et al., 2013;Nóbrega & De Marco Jr, 2011;Sousa-Baena et al., 2014).This knowledge gap about species distribution reflects the scarcity or complete absence of primary biogeographical data.
Despite research involving the collection of biogeographical data having been conducted for more than two centuries (Figueiró, 2015), many areas still remain under-sampled or unsampled altogether (Lobo et al., 2018;Lomolino, 2004), and this is relatively common in the Neotropical region.
In the Amazon rainforest, considered the largest and most diverse expanse of tropical forest on the planet (López-Baucells et al., 2019;Smith et al., 2023), approximately 23% of its area is sampled for bats, housing one in every ten known bat species (Bernard et al., 2012), in contrast to the 85% in the Atlantic Forest, the Brazilian biome best-sampled for bats (Aguiar et al., 2020;Bernard et al., 2012).However, we observe rapid land use and cover changes driven by deforestation and wildfires (Gomez et al., 2015;Silva et al., 2016).This swift change in land use and land cover, coupled with the lack of knowledge about species distribution, results in ineffective conservation strategies since many of the areas classified as mega-diverse, both in terms of species and ecosystem services, tend to overlap with deforested areas (Brasileiro et al., 2022;Delgado-Jaramillo et al., 2020).
Approximately 16.05% of the Amazon biome is no longer suitable for the ecosystem services bats provide (Brasileiro et al., 2022).
This loss reflects 17% of the total area of the biome that has been deforested and converted into pastures and short-cycle crops (Projeto MapBiomas, 2023).
This degradation of natural areas has a negative impact both on the climate as the forest is now considered a source of carbon emissions (Gatti et al., 2021), and on bats (Estrada & Coates-Estrada, 2002;Palheta et al., 2020;Vieira et al., 2021), leading to local extinctions in certain areas of the anthropocene (Hutson et al., 2001;Voigt & Kingston, 2016) and long-term effects as in the Brazilian cerrado, for example (dos Santos et al., 2016).Bats are considered keystone species in neotropical environments (Kunz & Fenton, 2005), playing crucial roles as seed dispersers (Kasso & Balakrishnan, 2013;Suripto, 2021), essential agents in reforestation of degraded environments (Muscarella & Fleming, 2007), and in pollination (Baqi et al., 2022;Buxton et al., 2022;Maruyama et al., 2022), even for plants of high economic, social, and ecological value (Fleming & Muchhala, 2008).Moreover, bats help control insect populations, including pests in agriculture and small vertebrates (Kunz et al., 2011;Ramírez-Fráncel et al., 2022).Bats are estimated to consume around 3200 caterpillars per hectare per night, amounting to a cost savings of US$ 390.6 million per harvest in Brazil (Aguiar et al., 2021).
Currently, there are 1474 species of bats recognized worldwide (Simmons & Cirranello, 2023), with 184 species occurring in Brazil (Garbino et al., 2022;Lopes et al., 2023;Zortéa et al., 2023), making them the second most diverse order of mammals (Simmons & Cirranello, 2023).This great taxonomic diversity is accompanied by a diversity of feeding habits, including insect consumption, nectar, fruits, seeds, amphibians, fish, small mammals, and even blood (Kunz et al., 2011;Schnitzler & Kalko, 2001).Of these 1470 bat species, 1332 species have been assessed by the International Union for Conservation of Nature (IUCN).They are listed in some of the 11 categories in the Red List of Threatened Species (IUCN Red List).Approximately 17.27% (230) of the species are classified as vulnerable (or extinct-9 species), 6.83% (91) as vulnerable, and 17.72% (236) as data deficient (DD).It is worth noting that one of the criteria for classifying a species as DD is a lack of knowledge about its geographical distribution.According to the IUCN, DD species are those that, although well-studied and have data on natural history, biology, and ecology, lack information about abundance (Prestonian shortfall) and geographical distribution (Wallacean shortfall).
Both shortfalls should be addressed by increasing efforts to collect primary data.However, fieldwork in remote areas is timeconsuming and costly, with costs varying depending on the region where the work is conducted (Aguiar et al., 2020;Balmford & Gaston, 1999).Furthermore, determining where to allocate efforts for primary data collection remains a key challenge (Aguiar et al., 2020).Nevertheless, the occurrence of a species in a particular area indicates its environmental limits related to abiotic factors and interactions within communities (Pellissier et al., 2010).In this context, species distribution models (SDMs) can address the Wallacean shortfall (Platts et al., 2010;Raxworthy et al., 2003;Razgour et al., 2011).Our goal is to reduce the Wallacean gap among Amazonian bat species.For this, we present (i) a species list of bats sampled in the Pará and Acre states in the last 5 years (2017 to 2022), (ii) the potential distribution of the species considered as new occurrences for the region, (iii) the potential distribution of the DD and NT (IUCN classification) species, and (iv) the comparison of species richness in three sample classes (urban, semi-urban, and natural).

| Bats sampling
The bats (Figure 1) were collected in the states of Pará and Acre (Figure 2), reaching 107 nights of mist net in 59 points (Table S1).
Additionally, we include nine colonies where bats were registered.

| Data analysis
We calculated the sampling effort following the method proposed by Straube and Bianconi (2002).Species richness was estimated by the Jackknife procedure (Heltshe & Forrester, 1983)

in the program
EstimateS version 8.0 (Colwell, 2005) with 1000 randomizations.Estimated and observed species richness was compared using the 95% confidence intervals.Species richness in the three classes (urban, semi-urban, and natural) was compared using the intersection among confidence intervals of estimated species richness.We consider only the locations sampled with the mist net to perform the jackknife.

F I G U R E 1 (Continued)
F I G U R E 2 Bat collection points and municipalities.
We created potential distribution models, SDMs, for species classified as new occurrences in the region.The distribution data were sourced from Map of Life (https:// mol.org/ ), which compiles biogeographic information and distribution maps for 1784 taxa (Marsh et al., 2022).We also developed SDMs for Data Deficient (DD) and Near Threatened (NT) species, following the classification of the Red List of Threatened Species (IUCN, 2024).These models were created with and without spatial restrictions, following the recommendations of Pimenta et al. (2022), using occurrence points from the entire Neotropical region for bat species included in the study.To minimize overfitting in the models, we filtered the occurrences for each species, avoiding duplicate occurrences and spatial autocorrelation.This procedure involves performing a Moran's correlogram (based on the linear distance between points) and identify and remove occurrences with significant autocorrelation, including the duplicated points.The number of unique occurrences of each species is presented in Table S2.

| Environmental variables
We used 19 bioclimatic variables (resolution of 9.The obtained data are part of the group of monthly climate variables sampled between 1970 and 2000 from WorldClim version 2.1 (Fick & Hijmans, 2017).These data are frequently used for species distribution modeling (SDM) to assess the potential distribution of species (Lee & Chen, 2012).To reduce multicollinearity in our dataset, we performed a principal component analysis (PCA) (Legendre & 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 & Nóbrega, 2018), using these axes as model variables.

| Algorithms
For the creation of SDMs, four algorithms were used: Maxent (MXE) (Phillips et al., 2017), Random Forest (RDF) (Prasad et al., 2006), Support Vector Machine (SVM) (Guo et al., 2005), and Bayesian Gaussian (GAU) (Golding and Purse, 2016).An ensemble combining the final suitability maps was generated by the four algorithms to minimize model uncertainties (Araújo and New, 2007;Diniz-Filho et al., 2009).To mitigate model uncertainties, an ensemble approach was adopted as the final model (Pimenta et al., 2022;Velazco et al., 2019).This ensemble model consists of the average suitability across models for which the Jaccard threshold values (Pimenta et al., 2022) were greater than the average thresholds for each species (Velazco et al., 2019).The Jaccard threshold was selected to minimize omission and commission errors in the models (Pimenta et al., 2022).
Additionally, we applied spatial restrictions to the models to minimize overprediction in distribution models (Mendes et al., 2020;Pimenta et al., 2022).To do this, we created a binary occurrence map, where suitability values greater than the Jaccard threshold indicated species presence, and then partitioned it into pixels with species occurrence and pixels without species occurrence.
Subsequently, only pixels where the species was predicted and had species records or pixels where the species was predicted and were near pixels with predictions and occurrence points were retained in the species' potential distribution map (Pimenta et al., 2022).For the partitioning of the binary map, we considered two methods: (i) Species with more than 30 occurrence points: partitioning the map using the chessboard method (Andrade et al., 2020); (ii) Species with fewer than 30 points: randomly selecting a percentage of points for modeling and another for evaluation, with 70% of the points selected for the model and 30% for evaluation (Pimenta et al., 2022).Since spatial restriction generates more conservative maps, limiting the occurrence areas to locations near or with species occurrence, we conducted a second modeling without spatial restriction.Thus, we have a more restrictive and conservative model (a model with spatial restriction) and a less conservative model containing areas with environmental suitability without considering whether the species occurs or not.All procedures were performed using the enmtml function implemented in the ENMTL package (Andrade et al., 2020) for the R environment (R Core Team, 2010).

| Model evaluation
The evaluation was performed using Receiver Operating Characteristic (ROC) curves, and the efficiency of each model was assessed using the True Skill Statistic (TSS) test, which has been widely advocated as an appropriate discrimination metric that is independent of prevalence (Allouche et al., 2006;Shabani et al., 2018).TSS is an intuitive method for measuring the performance of Species Distribution Models (SDMs), which calculates sensitivity (true positive rate, TPR) and specificity (true negative rate, TNR) values, where predictions are expressed as presence-absence maps.However, TSS values can be misleading when the number of true negatives assigns higher values to species with lower prevalence (Lawson et al., 2014).To avoid these deficiencies, we propose focusing the evaluation metrics on three components of the confusion matrix: true positives, false positives, and false negatives, neglecting true negatives that could inflate the data.In other words, we aim to maximize true positives while minimizing false positives and false negatives relative to true positives (Leroy et al., 2018).

| DISCUSS ION
The high diversity of Phyllostomidae bats, a relatively common pattern in studies conducted with mist nets in the Neotropical region, can be explained by the very use of the capture method, a methodology known to be selective.This may explain the abundance of the family in our database (Pedro & Taddei, 1997;Sipinski & Reis, 1995).

F I G U R E 4
Occurrences included in the species distribution modeling.The procedures of modeling are described in the Appendix S1 and the evaluation of models are in the Table S3.
We observed different environments, ranging from urban and rural areas to natural habitats, but despite the high diversity of bats in our study, it would still be possible to add new species to the list, especially if we use other sampling methodologies, such as acoustic monitoring.This way we could capture other families that are not normally captured in mist nets.
Supporting this, we captured six species, five from the Molossidae family (Eumops glaucinus, Eumops perotis, Molossus coibensis, Molossus currentium, and Nyctinomops laticaudatus) and one from the Natalidae (Natalus macrourus) that were sampled only through active capture when we located their shelters and manually captured them.Additionally, one species, Thyroptera discifera, was recorded from a single individual donated to the laboratory and was not captured with mist nets or active search.
The species richness estimator predicts these additions to the species list (we recorded 75 and estimated the occurrence of nearly 95 species) based solely on mist net data.Additionally, 28% (21 species) of the captured species are considered range extensions, supporting that work with primary biogeographic data are still necessary today, especially in the Brazilian Amazon region.
The Brazilian Amazon region has a vast geographic expanse and few consolidated research centers (Aguiar et al., 2020;Delgado-Jaramillo et al., 2020).This pattern is observed not only for bats but also for other biological groups (Delgado-Jaramillo et al., 2020;Dias-Silva et al., 2021).We observe major research centers concentrated in the southwestern and coastal regions of Brazil, with few research groups in the interior of Brazil, including the Amazon Biome and the Caatinga (Brito et al., 2009;Delgado-Jaramillo et al., 2020;Lewinsohn & Prado, 2002).
Studies investigating priority areas for the conservation of biological groups tend to rank the Caatinga region as less critical for conservation (Dias-Silva et al., 2021).However, this result is not due to its actual low importance but rather because of the failure of species distribution models (SDMs) to predict occurrences in these areas.This failure results from a need for primary biogeographic data (Delgado-Jaramillo et al., 2020;Silva et al., 2018).This absence can be observed in the distribution maps of data points used for these SDMs.This gap is particularly evident in the SDM of Myotis levis, where the species has three occurrence points for Pará (our study area), with one reported by us and many points concentrated in the South American region.This concentration tends to generate an SDM that does not include the study area, even though there are (few) species records in the region.Some other less obvious examples of the lack of primary biogeographic data are observed for the species Natalus macrourus, Diphylla ecaudata, Anoura geoffroyi, Pygoderma bilabiatum, Platyrrhinus infuscus, and Dermanura anderseni.Despite the SDMs showing suitable areas for the species in the region, these areas are limited to small "islands" near the collected species.These gaps in the geographic distribution of species are, in many cases, the result of a lack of primary biogeographic data.This highlights the need for investments in collecting primary data for emerging groups in regions with little sampling.The lack of primary biogeographical data and uncertainties regarding the geographical distribution of species have been observed in other bat species, such as Tonatia maresi (Aguiar et al., 2015), Histiotus velatus (Da Silva et al., 2021), and more recently in Tadarida brasiliensis brasiliensis (do Amaral et al., 2023).
Furthermore, changes in the species' distribution areas over the years due to the addition of new occurrence points to models have also been noted (Da Silva et al., 2021), thus reinforcing the need for investments in obtaining primary biogeographical data (Aguiar et al., 2015;Da Silva et al., 2021).
For the Amazon region, in particular, collecting biogeographic data is very costly because transportation is mainly done by river or on poorly maintained roads that become impassable during specific periods of the year.In the region analyzed in this study, along the Trans-Amazonian Highway, there is a period of 4-5 months (from December to April/May) when the roads are practically impassable, including the main highway, BR 230.This means sampling is concentrated during the dry season or in areas closer to the research center where we operate (Aguiar et al., 2020;Dias-Silva et al., 2021).
We want to emphasize two points regarding the reduction of the Wallacean gap.The first point pertains to bat species with expanded geographic distributions, with 21 out of the 75 sampled species falling into this category.This result highlights that the areas where we focused our studies, especially those in Pará, are areas with poorly known bat fauna.The study area we explored is highly diverse and potentially harbors species that are new to the literature (Linnaean gap), as we have observed the recent description (after 2010) of six bat species (Aguiar et al., 2020).Furthermore, the number of sampling points in the studied region is low compared to the central portion of the Atlantic Forest (Aguiar et al., 2020), underscoring the need for inventories and studies that generate primary data.
The other point to be noted, related to the Wallacean shortfall, concerns the environmental similarity between the sampled points in our study and the areas with the highest sampling efforts in Brazil.
Areas with low bat sampling, especially those in the Amazon biome, have low environmental similarity to Brazil's more heavily sampled areas (Aguiar et al., 2020).Therefore, areas considered as sampling gaps may be areas of occurrence for new species, both in terms of species already described but with geographic distributions that do not include Brazil and species new to science.This argument can be supported by the expanded distribution of Platyrrhinus guianensis Velazco andLim, 2014 (Lopes et al., 2023), Artibeus amplus Handley, 1987 (Zortéa et al., 2023), andChoeroniscus godmani (Thomas, 1903) (Garbino et al., 2022).
All these species are considered new occurrences for Brazil, with P. guianensis and A. amplus having a single occurrence point for Brazil, located in the Amazon region near our study area in Pará (Lopes et al., 2023;Zortéa et al., 2023).Additionally, the species P. guianensis was described in 2014, reinforcing the idea that the region sampled in our study is potentially a diverse area and a potential source of new species unknown to science (Linnaean shortfall).
The environmental difference between the areas considered as sampling gaps for bats in Brazil and well-sampled locations (Aguiar et al., 2020), measured based on the bioclimatic conditions used in SDM (Bendjeddou et al., 2022;Da Silva et al., 2021;Da Silva et al., 2022;Delgado-Jaramillo et al., 2020;Pimenta et al., 2022), could explain why the distribution models presented here do not predict species occurrence in the region, even though the species is captured on-site.Indeed, the failure of our models to predict occurrence may be a combination of limited sampling in the region, along with occurrence points that are concentrated in wellsampled areas with different environmental conditions than our study area.
Occurrence points centered in environmentally similar areas, contrasting with a few points with low environmental similarity, can lead to models with over-prediction of the distribution area, mainly when geographic restriction is applied (Pimenta et al., 2022;Soberón & Nakamura, 2009).Consequently, models may suggest that areas with few occurrence points (and low environmental similarity compared to the more extensive set of points, in the case of Brazilian bats) have a low probability of species occurrence and are considered unsuitable for the species (Chang et al., 2020;Peterson et al., 2008;Soberón & Nakamura, 2009).
However, the need for more suitability might be due to the environmental difference between the areas where occurrence points are concentrated and where the species has an extended range.This scenario is observed in some of the presented occurrence extension models, where many points are located in environmentally dissimilar areas from the new occurrence point (Aguiar et al., 2020).
Consequently, besides having a large sampling gap for bats, the Amazon region might have its diversity overestimated by species distribution models (SDMs).This sampling gap has a direct impact on species conservation prioritization efforts since some methods are derived from SDMs (e.g., Fielding & Bell, 1997;Meggs et al., 2004;Silva et al., 2018), potentially leading to the creation of inefficient reserves for biodiversity conservation (Brasil et al., 2021;Dias-Silva et al., 2021).
We hope the SDMs can serve as a foundation or be directly used to guide conservation efforts, especially for DD and NT species.Additionally, 21 species classified as new occurrences should be seen as a reminder that primary biogeographic data, includ-

CO N FLI C T O F I NTE R E S T S TATE M E NT
There is no conflict of interest.
The region has a tropical climate of type Am, according to the Köppen climate classification, with an average temperature of 26°C and an average annual rainfall of 1914 mm.Bats were sampled using mist nets (10 nets measuring 9 x 2.5 m at each point) open at sunset and remaining for 6 h, inspected every half hour.The captured bats were packed in 100% cotton fabric bags.In the field, the individuals were weighed, underwent morphometric F I G U R E 1 Images of species collected in the states of Pará and Acre.The name of each species is already described in the corresponding image.
measurements, and were also identified by gender, age, and reproductive status.Subsequently, the individuals that were first-time captures or were not identified to the species level in the field were taken to the Laboratory of Ecology of Altamira (LABECO) of the Federal University of Pará (UFPA), Altamira campus, euthanized by cervical dislocation and the morphometric data (total length of the foot, ear, tragus, forearm, and weight) were measured.Subsequently, the bats were fixed with 10% formaldehyde and stored in glass containers with 70% alcohol in the ChiroXingu Bat Collection: Nucleus of Studies in Ecology and Conservation of Chiroptera, located at UFPA, Altamira-PA campus, Brazilian Amazon.The ChiroXingu research group collected the bats from 2017 to 2022 under license 57,294 issued by the Biodiversity Authorization and Information System-SISBIO.

F
4 x 9.4 km) for the neotropical region, obtained from the WorldClim database (http:// www.world clim.org/ ).These variables include: Mean annual temperature; Monthly mean diurnal temperature range; Isothermality; Temperature seasonality; Maximum temperature of the warmest month; Minimum temperature of the coldest month; Annual temperature range; Mean temperature of the wettest quarter; Mean temperature of the warmest quarter; Mean temperature of the coldest quarter; Annual precipitation; Precipitation of the wettest month; Precipitation of the driest month; Precipitation seasonality; Precipitation of the driest quarter; Precipitation of the wettest quarter; Precipitation of the warmest quarter; Precipitation of the coldest quarter.
This test slightly restricts the occurrence area, leading to a less inclusive map, considering errors of omission in species distribution (false negatives) and commission (false positives), with values ranging from −1 to +1 (Sensitivity + Specificity) to indicate the predictive ability of the models.Models with TSS values close to +1 reflect good predictive ability; models with TSS values between 0.2 and 0.6 are considered fair to moderate; and models with TSS values close to 0 or negative indicate low capability.

F
I G U R E 3 Observed and estimated bat species richness sampled between 2017 and 2022.The bars represent the 95% confidence interval.The SDMs showed significant variation in terms of distribution areas, with the areas of SDMs with geographic restrictions tending to be smaller than those without restrictions.Some models exhibited limited potential distribution areas for the sampled Amazon region (Figures S4-S9), with three SDMs showing restricted or nonoccurrence areas for the sampled Amazon region (Figures S8, S15, S17, and S18).
ing mist-netting and other sampling methods, are still necessary for the Neotropical region.Investments in such research in the Amazon and Caatinga biomes are crucial, mainly when directed toward small emerging research groups located outside the state capitals.AUTH O R CO NTR I B UTI O N S Thiago Bernardi Vieira: Conceptualization (lead); data curation (lead); formal analysis (lead); funding acquisition (lead); investigation (lead); methodology (lead); project administration (lead); software (lead); supervision (lead); validation (lead); visualization (lead); writing -original draft (lead); writing -review and editing (lead).Rafaela Jemely Rodrigues Alexandre: Data curation (equal); methodology (equal); writing -review and editing (equal).Simone Almeida Pena: Data curation (equal); methodology (equal); writing -review and editing (equal).Letícia Lima Correia: Data curation (equal); investigation (equal); methodology (equal); writing -review and editing (equal).Ariane de Sousa Brasil: Investigation (equal); methodology (equal); writing -review and editing (equal).Ludmilla Moura de Souza Aguiar: Formal analysis (equal); methodology (equal); validation (equal); writing -review and editing (equal).Paulo De Marco: Formal analysis (equal); software (equal); writing -review and editing (equal).Albert David Ditchfield: Formal analysis (equal); methodology (equal); validation (equal); writing -review and editing (equal).ACK N OWLED G M ENTS This research benefited from resources from Vale SA's environmental compensation administered by the Centro Nacional de Pesquisa e Conservação de Cavernas (Cecav/ICMBio) and services to the Brazilian Society for the Study of Chiropterans-SBEQ, under the administration of the Instituto Brasileiro de Desenvolvimento e Sustentabilidade-IABS.FU N D I N G I N FO R M ATI O N This research benefited from resources from Vale SA's environmental compensation administered by the Centro Nacional de Pesquisa e Conservação de Cavernas (Cecav/ICMBio) and services to the Brazilian Society for the Study of Chiropterans, SBEQ, as part of the DD Program, The Species More Unknown in Brazil, and with resources from the Termo de Compromisso de Compensação Espeleológica, TCCE VALE 1/2018, Edital Ferruginosas 01/2021, under the administration of the Instituto Brasileiro de Desenvolvimento e Sustentabilidade, IABS.National Council for Scientific and Technological Development, CNPQ.Amazon Foundation to Support Studies and Research-FAPESPA.Foundation Coordination for the Improvement of Higher Education Personnel-CAPES.