INTRODUCTION

The climate of the Earth has been characterized by multiple glacial-interglacial cycles occurring over millions of years (Paillard, 1998). Much attention has been paid to the transition from the cold climate of the Last Glacial Maximum (LGM; ca. 25–18 ka) to the warmer interglacial climate of the Holocene (HOL; ca. 10–0 ka) because this period includes events in human history and it lies within a time interval that can be easily radiocarbon dated. Isotopic evidence indicates that during the LGM, climate in the Caribbean was more arid with temperatures about 5°–8° C colder than today (Curtis et al., 2001). In contrast, mid-Holocene (ca. 6 ka) temperatures were about 1°–2° C warmer than today and habitats changed from xerophytic to mesophytic after the last glaciation (Pregill and Olson, 1981; Foley et al., 1994; Huang et al., 2008). Sediment cores from mainland and insular Caribbean localities indicate major shifts from extensive dry scrubland to more moisture-tolerant vegetation following climate change at the Pleistocene-Holocene transition (PHT, ca. 11–9 ka; Leyden, 1984, 1985; Higuera-Gundy et al., 1999). Faunal changes are also associated with the PHT; for example, some birds such as the Bahama mockingbird (Mimus gundlachi) and the burrowing owl (Athene cunicularia) are known to have experienced range contractions and extirpation in the Bahamas with the loss of xeric environments (Pregill and Olson, 1981).

Several hypotheses have been proposed to explain the effect that the PHT had on the biotic diversity in the Caribbean (Pregill and Olson, 1981). Fossil evidence indicates that about 80% of the land mammals in the Caribbean became extinct sometime after the LGM (MacPhee, 2009; Dávalos and Turvey, 2012). Many of these taxa (e.g., sloths and insectivores) disappeared shortly after the last glaciation (Pregill and Olson, 1981; Steadman et al., 2005). In contrast, only about 18% of Caribbean bats became extinct during the same period (Pregill and Olson, 1981; Morgan, 2001; Dávalos and Turvey, 2012). Changes in island area due to rising sea level and concomitant changes in cave availability due to flooding and loss of roosting habitats after the PHT have been proposed to explain Caribbean bat extinctions (Morgan, 2001; Dávalos and Russell, 2012). However, recent evidence from radiocarbon chronology has indicated that many bat fossils in the Caribbean are younger than previously thought, and that many species of bat survived the PHT by 5–7 ky (Soto-Centeno and Steadman, 2015). Furthermore, evidence from time-scaled ecological niche models (ENM) covering the LGM to the present have revealed that climate-based distributions for the bats Monophyllus redmani, Macrotus waterhousii, and Pteronotus parnellii remained largely stable over time (Soto-Centeno and Steadman, 2015). This evidence from fossils and ENMs suggests that some species of bats were able to maintain viable populations until the late Holocene despite habitat loss due to rising sea level after the PHT.

Ecological niche models combine species localities and environmental data to estimate the predicted distributions of organisms based on available suitable habitat (Franklin, 2009). ENMs have been used to study biogeographic, ecological, and evolutionary questions (Carnaval et al., 2009; Soto-Centeno et al., 2013) and are an exceptional tool for understanding how climate may influence the distribution of a species. ENMs can use current and past climate datasets to hindcast species distributions across time (Elith and Leathwick, 2009). These time-scaled ENMs, in combination with fossil data, provide a strong framework for testing hypotheses of climate-based extinction or extirpation (Soto-Centeno and Steadman, 2015).

As with all mammals, the distributions of bats are dependent on many variables of which climate is just one factor (Peterson et al., 2011; Anderson, 2013). For many bat species, the availability of suitable roost sites may be a significant factor that influences population distributions and, ultimately, the geographic range of species (Peterson et al., 2011; Anderson, 2013). For example, cave-roosting bat species may be limited by the availability of suitable caves in which to roost (Rodríguez-Durán, 2009). Hot caves, where temperatures range from 28°–40° C and humidity exceeds 90%, are preferred roosts of some bat species of the Neotropics, but their development depends upon the underlying geology of a region (Ladle et al., 2012). Neotropical hot caves are known from northeastern Brazil, the Greater Antilles, Mexico, and Venezuela (Rodríguez-Durán, 2009; Ladle et al., 2012).

The Caribbean is rich in calcium carbonate (CaCO₃) and calcium-magnesium carbonate (CaMg[CO₃]) deposits (Mylroie and Mylroie, 2007). These deposits, commonly known as karst or limestone, have a high prevalence of caves that are formed by dissolution of minerals by water moving through the deposits (Mylroie and Mylroie, 2007; Rodríguez-Durán, 2009). Although caves are common geological features worldwide, only some have the geomorphology necessary to become hot caves (Ladle et al., 2012). Fluviokarst caves—caves formed by rivers or precipitation that percolates through limestone—provide the most suitable conditions for hot cave formation (Rodríguez-Durán, 2009). In the Caribbean region, fluviokarst caves are found only in the Greater Antilles (Rodríguez-Durán, 2009).

The microclimatic conditions of hot caves are thought to be caused mainly by the radiating body heat from large colonies of bats and heat/humidity associated with decomposing guano (Rodríguez-Durán, 2009; Ladle et al., 2012). Species of Caribbean bats in the families Mormoopidae, Natalidae, and some members of Phyllostomidae primarily roost in hot caves, where the range in temperatures promotes the segregation of species (Rodríguez-Durán and Soto-Centeno, 2003; Rodríguez-Durán, 2009). Thus, given their specific microclimatic conditions, hot caves in the Caribbean play a significant role as roosts for many species of bats. Indeed, some bats such as Chilonatalus micropus, Erophylla bombifrons, Monophyllus redmani, and Pteronotus quadridens have been posited to roost only in hot caves (Rivera-Marchand and Rodríguez-Durán, 2001; Rodríguez-Durán and Soto-Centeno, 2003; Tejedor et al., 2005a).

In contrast to Greater Antillean fluviokarst caves, karst in the Bahamas is predominantly eolianite (Brooke, 2001). This type of karst results from the lithification of sediment that is deposited by wind processes in shallow banks, such as those in the Bahamas (Brooke, 2001). The flooding of carbonate platforms during the middle to late Pleistocene sea level high stands shaped large expanses of eolianite karst in the Bahamas (Brooke, 2001). Weathering of exposed eolianite by glacial-interglacial sea level changes aided in the formation of numerous small caves but not the large, complex caves commonly found in fluviokarst. Consequently, even though caves are common throughout the Bahamas, these caves usually lack the geomorphology and distinctive features of Greater Antillean hot caves.

We studied three species of bats in the family Mormoopidae that are codistributed in the Caribbean (fig. 1A). The ghost-faced bat (Mormoops blainvillei Leach, 1821; fig. 1B), Parnell's mustached bat (Pteronotus parnellii Gray, 1843; fig. 1C), and sooty mustached bat (Pteronotus quadridens Gundlach, 1840; fig. 1D) are small- to medium-sized aerial insectivores that prefer to roost in hot caves (Rodríguez-Durán and Soto-Centeno, 2003; Mancina et al., 2007; Ladle et al., 2012). Given the recent evidence from radiocarbon-dated fossils and ENMs, we hypothesized that climate-based distributions of mormoopid bats will show stability in the Caribbean similar to other species of bats. However, Morgan (2001) suggested that rising sea level following the PHT damaged the structure of many caves, rendering them unable to maintain a warm microclimate. If so, PHT change in climate could have affected the distribution of mormoopid bats in the Caribbean because of the lack of suitable hot cave roosts (Morgan, 2001). We tested this hypothesis in two ways: first, we developed a series of buffers around karst outcrops to estimate the proportion of bat sampling localities within karst as well as outside of karst; second, we created ENMs of the intersection of climate and karst to examine the role of karst (i.e., where hot caves are likely to exist) in the distributions of mormoopid bats.

FIGURE 1.

(A) Map of the Caribbean region showing locations of major islands and archipelagos discussed in the text. Portraits of the mormoopid species that are the focus of this study: (B) Mormoops blainvillei (C) Pteronotus parnellii, and (D), Pteronotus quadridens. All photographs by J.A.S.-C.

MATERIALS AND METHODS

Species Localities and Climate Data

Bat-sampling locality records with decimal latitude and longitude coordinates were obtained from the Global Biodiversity Information Facility (GBIF;  www.gbif.org) and from the literature. We plotted all points in ArcGIS v10.2 (ESRI, Redlands, CA) to assess the quality of the spatial data and we corrected localities that were clearly erroneously georeferenced (e.g., lying in the ocean). Localities georeferenced from the literature were recorded at a resolution of 0.0001 decimal degrees, while points obtained from GBIF were kept at their original resolution. Information on fossil localities was obtained from the literature and from collections held at Florida Museum of Natural History, and localities were subsequently georeferenced. Via these methods, we obtained a total of 95 current and 12 fossil localities for Mormoops blainvillei, 109 current and 13 fossil localities for Pteronotus parnellii, and 75 current and 10 fossil localities for P. quadridens (figs. 2–3; appendix table A1). A recent search after the submission of this article produced two additional locality records: Pteronotus quadridens, Mona, Puerto Rico (recent; Lancaster, 2010), and Mormoops blainvillei, Exuma, Bahamas (fossil; Koopman, 1951). Careful examination of the climate models revealed that the addition of these localities did not have a significant impact on or change our interpretation of the climate models. To maintain the integrity of the dataset used for producing the models, we did not include these points in our database.

We used climate data from 19 WorldClim variables at 30 sec (ca. 1 km²) spatial resolution for current, mid-Holocene (HOL, ca. 6 ka), and Last Glacial Maximum (LGM, ca. 21 ka) climate (Hijmans et al., 2005). Both paleoclimate conditions were obtained as statistically downscaled estimates of past conditions from the ESCHAM3 and CCSM models available in WorldClim (see  www.woldclim.org for downscaling details). Each variable was clipped to the extent of the Caribbean using ArcGIS v10.2. We performed a Pearson correlation analysis in ENMTools v1.4.3 (Warren et al., 2010) on the 19 climate variables to avoid issues with collinearity and chose seven uncorrelated variables to represent regional temperature and precipitation (correlation coefficient >0.8; Peterson, 2011). The seven climate variables chosen were annual mean temperature (Bio1), mean diurnal temperature range (Bio2), isothermality (Bio3), temperature annual range (Bio7), annual precipitation (Bio12), precipitation of driest month (Bio14), and precipitation of warmest quarter (Bio18; see Hijmans et al., 2005).

We extracted climate information for our species localities using ArcGIS v10.2. A high volume of background climate points has the potential to overfit models to the locality data because of the high abundance of absence points surrounding presence localities (Elith et al., 2011). To correct for this, we ran models by randomly sampling 3000 background localities within the Caribbean region. We used a 2 km buffer around species locality points and removed background points located within this buffer to prevent bias in our background climate dataset.

FIGURE 2.

Current localities (black dots) used to develop and evaluate ecological niche models of current climate conditions for Mormoops blainvillei, Pteronotus parnellii and Pteronotus quadridens in the Caribbean. See figure 1 for geographic reference.

FIGURE 3.

Fossil localities (black dots) used to develop and evaluate hindcasted ecological niche models for Mormoops blainvillei, Pteronotus parnellii, and Pteronotus quadridens in the Caribbean. See figure 1 for geographic reference.

Ecological Niche Modeling

We developed ecological niche models (ENM) in Maxent v3.3, which uses a maximum entropy algorithm to predict the relationship between a species locality and the environmental and spatial characteristics of those localities (Elith et al., 2011). Previous studies suggest that the use of default parameters in Maxent for every species is not appropriate (Anderson and Gonzalez, 2011). Therefore, for each species of bat we explored multiple values of regularization, which controls the ability of a model to be more restricted or relaxed. We tested which regularization value showed better predictive ability for each of the species datasets using two threshold rules: minimum training presence threshold (MTP) and the 10th percentile training presence threshold (10PTP), estimated within Maxent. The MTP is the lowest threshold value that is able to accurately predict all training locality points (Radosavljevic and Anderson, 2014). We used MTP because previous studies suggest that relaxed thresholds provide useful estimates when hindcasting distributions (Waltari and Guralnick, 2009). The 10PTP is a more restrictive threshold that attempts to exclude only 10% of the training locality points (Radosavljevic and Anderson, 2014). We compared the number of correctly predicted localities (i.e., true positive fraction) against the theoretical expectation of correctly predicted localities for each threshold to test for model overfitting. Using this method we estimated regularization values, which were kept constant when developing hindcasted model projections (table 1).

Final models were created using 100 bootstrap replicates and projected to current, HOL, and LGM climate in the Caribbean. Bootstrap replicates consisted of sampling with replacement of each training dataset (i.e., 75% of the presence localities for each species). Because bootstrap iterations are generated as pseudoreplicates from the original dataset, we took the average of each threshold in each replicate to determine a MTP and a 10PTP threshold value for each model. We used the logistic output created by Maxent, which represents a continuous model, and applied thresholds in ArcGIS v10.2 to convert it into a binary (i.e., presenceabsence) model. The resulting binary distributions for the three time periods were summed and corrected for the current island area to develop stability maps for each species. Predicted areas of stability show suitable areas where each species is predicted to occur across the current, HOL, and LGM time periods.

The Importance of Karst

It is difficult to distill all factors that rule the distribution of a species into one abiotic variable such as presence of karst because species geographic distributions are governed by complex interactions among many abiotic and biotic variables (Hutchinson, 1957; Holt, 2009; Peterson et al., 2011). Nonetheless, we wanted to assess whether the presence of karst outcrops, and thus the potential availability of hot caves, is a useful predictor in the distribution of Caribbean mormoopid bats. Data on presence and absence of karst were obtained from the KROW database (Hollingsworth, 2006), which to our knowledge is the only readily available digital map of karst that includes the Caribbean. This database includes areas of fluviokarst formations in the Greater Antilles, but not areas of eolianite karst that occur in shallow banks, such as those in the Bahamas. Developing an ENM using presence and absence of karst alone would not answer our question effectively because bats fly and a preliminary assessment showed that many localities from the three species of bat studied lie outside karst areas across the Caribbean. As a result, the model considered areas inside and outside karst areas as equally suitable. Instead, we estimated the proportion of localities of each species that lie within and outside karst areas as well as their distance from karst by creating a series of buffers at 5 km intervals. This approach is justified because many species of bats are known to routinely fly long distances from their roosts while foraging. In the Neotropics, movement between day roosts and foraging areas has been reported to range from 2.5 km in Carollia perspiscillata (Phyllostomidae; Bernard and Fenton, 2003) to 18 km in Tadarida brasiliensis (Molossidae; Horn and Kunz, 2008) and even 30 km in Leptonycteris yerbabuenae (Horner et al., 1998). On Cuba and Puerto Rico, Pteronotus quadridens has been reported to disperse up to 9 km to their feeding grounds (Rodríguez-Durán, 1984) and are also able to return to their roost on a single night from distances up to 30 km (Silva-Taboada, 1979). It is likely that Mormoops blainvillei and Pteronotus parnellii are also capable of long-distance flight (e.g., Bateman and Vaughan, 1974). Thus, karst might be important (i.e., for providing cave roost sites) even to bats captured in nonkarst areas if foraging areas are located many kilometers away from roost sites.

Table 1.

Model statistics estimated for current (CUR), mid-Holocene (HOL), and Last Glacial Maximum (LGM) climate using a minimum training presence threshold (MTP) for Mormoops blainvillei, Pteronotus parnellii, and Pteronotus quadridens in the Caribbean. True-positive fraction (TPF) estimates were based on climate-only and climate-plus-karst models (i.e., karst-clim). Karst-climate models are based on the intersection of climate and occurrence of karst in the Caribbean. Statistics: Regularization value (β), area under the curve (AUC), and true-positive fraction (TPF).

To determine how karst affected each species' climate-based distribution, we used the karst layer to estimate the intersection of karst and climate. This indicated where both climatic and habitat suitability cooccurred in the Caribbean at each time scale. The relaxed nature of the MTP threshold used in the climate models combined with karst provided conservative estimates suitable for hindcasting distributions while reducing the probability of overfitting. We projected the KROW database from World Eckert IV into World Geodetic System 1984 (WGS 84) datum to match our climate data and adjusted irregularities of the karst map to match the boundaries of the Caribbean in ArcGIS v10.2.

Evaluating Model Performance

Model performance was assessed using the area under the receiving operating characteristic curve (AUC) statistic from Maxent. AUC is a threshold independent measure of performance where a value of 1 indicates perfect model performance and a value of 0.5 indicates a model performing no better than random (Peterson et al., 2011). Values greater than 0.7 indicate good model performance (Swets, 1988; Peterson et al., 2011). AUC values do not take into account whether a model is overfit (Radosavljevic and Anderson, 2014). Therefore, we also examined the true-positive fraction using the MTP and 10PTP thresholds as an additional measure of performance. True-positive fractions for the karst-climate models were estimated using only localities that fell within karst to avoid bias in the estimates of correctly predicted localities.

One goal of this study was to determine whether recent climate change had an effect in the distributions of each species. Therefore, for the climate-only models we estimated two indices: relative range size (current ≥ past RRS = c/m - 1 or current ≤ past RRS = -1×[m/c - 1]) and overlap index (OI = o/m; where m = current climate conditions, c = past climate conditions, and o = amount of overlap between current and paleo-ENMs; Hijmans and Graham, 2006). We used RRS to estimate the magnitude and directionality of change in distributions, where an RRS value of zero indicates no change in distributions across time. Conversely, a positive RRS means that the past range size was larger than the current one and a negative value indicates that the past range size was smaller than the current one (Hijmans and Graham, 2006). Overlap index was used to understand the proportion of overlap between the present distribution and past distributions (Hijmans and Graham, 2006).

RESULTS

The Effect of Climate on Caribbean Mormoopids

Climate-based ENM projections showed relative range sizes (RRS) that indicate range contraction from the LGM to the HOL followed by a range expansion from HOL to the present in two of the three species of bats (table 3; figs. 4–6). A comparison of the overall RRS from LGM to the present reveals a trend of range contraction in Mormoops blainvillei and Pteronotus parnellii, which exemplifies the loss of habitat (i.e., land area) due to sea level rise after the end of the Pleistocene (table 3; figs. 4–5). Despite these observed changes, the amount of overlap (OI) in distribution across time (a measure of geographic range stability through time) was at least 81% in Mormoops blainvillei and 73% in Pteronotus parnellii, indicating relative stability despite loss of habitat. In contrast, the overall trend in P. quadridens showed a range expansion more or less continuously from LGM to the present (table 3; fig. 6). The overlap through time was 47% in P. quadridens (table 3), but in this case the low value reflects that new areas of suitable habitat became available after the LGM. We represent the areas of overlap as stability maps that show the intersection of predicted distributions across the three time periods for each species of bat (fig. 7).

Table 2.

Alternative climate-only model statistics using 10th percentile training presence threshold (10PTP) for three species of mormoopid bat on the West Indies. Models using 10PTP also showed a range expansion from mid-Holocene to the present (see table 1). Overlap indexes in the 10PTP models were lower than when using a minimum training presence threshold (MTP), indicating that 10PTP is more restrictive. Statistics: Regularization value (β), area under the curve (AUC), overlap index (OI), relative range size (RRS), true-positive fraction (TPF).

We also estimated RRS corrected for the present-day island area to examine the amount and directionality of change for land areas that have not been affected by sea level rise. These estimates show a trend of range expansion from the LGM to the present for all three bat species. The magnitude of this change for Mormoops blainvillei and Pteronotus parnellii was <0.32, while the range expansion in P. quadridens was >0.5 (table 3).

Karst and ENMs of the Intersection of Karst and Climate

We generated a series of buffers at 5 km intervals to obtain the proportion of localities of each species that lie outside of fluviokarst as well as their distance from karst for each species of bat (fig. 8). These data are summarized in table 4. Only 50%–70% of modern sampling localities for the three focal species actually occur in karst areas (table 4). However, >90% of known capture localities for these taxa occur within 15 km of karst (table 4). Only 1% of modern localities for Pteronotus quadridens and 4% of capture localities for P. parnellii and Mormoops blainvillei occur farther than 25 km from karst.

For Mormoops blainvillei, 96% of current localities and 83% of fossil localities occur within 15 km of karst outcrops. Pteronotus parnellii is known from somewhat more localities, but the pattern remains the same: 93% of modern localities and 85% fossil localities occur within 15 km of karst (table 4). A total of 93% of modern localities and 70% of fossil localities for P. quadridens are within 15 km of karst (table 4). For all three species, the totals for fossil localities are affected by exclusion of localities from Bahamas and Cayman Islands, because eolianite karst is not included in the KROW database. To be conservative, we treated eolianite karst in these areas as nonkarst for this analysis because its geomorphology is different than that of fluviokarst.

Table 3.

Overlap index (OI) and relative range size (RRS) estimates for Mormoops blainvillei, Pteronotus parnellii, and Pteronotus quadridens in the Caribbean. Time of the predicted distributions is indicated as current (CUR), Holocene (HOL), and Last Glacial Maximum (LGM). Threshold values were determined using the minimum training presence threshold (MTP). RRS for LGM distributions corrected for current island area are indicated with an asterisk (*). Negative RRS values indicate that the more recent range is bigger than the past (i.e., range expansion). Conversely, positive RRS values indicate that the more recent range is smaller than the past (i.e., range contraction).

To estimate the importance of karst, and how PHT change in climate and the resulting loss of suitable hot cave roosts could have affected the distribution of mormoopid bats in the Caribbean, we constrained climate-based distributions to the consistently exposed areas of karst on the Greater Antilles. Similar to the ENMs produced using climate data alone, the karst-climate ENMs using the MTP threshold showed good overall performance in all species for localities occurring within karst (table 1). True-positive fractions for Mormoops blainvillei, Pteronotus parnellii, and P. quadridens for the current karst-climate models were 98%, 100%, and 100%, respectively. Past projections to the HOL and LGM of the karst-climate ENMs had a truepositive fraction of 80% for Mormoops blainvillei and 100% for Pteronotus parnellii and P. quadridens (table 1). As noted previously, in contrast to the Greater Antilles, the Bahamas are part of two shallow banks and eolianite karst in these islands has been consistently eroded by glacial-interglacial changes in sea level. As a result, karst-climate ENMs were unable predict the fossil localities in the Bahamas (fig. 9).

DISCUSSION

Bats are widely distributed across the Caribbean, and for many islands they are the only extant native mammals (Morgan, 2001; Rodríguez-Durán and Kunz, 2001; MacPhee, 2009). Examining the effects of recent climatic changes and the availability of suitable roosting habitat may illuminate our understanding of causes contributing to extirpation of populations of insular bats. In this regard, estimating the intersection of karst with climate-based models can provide insight into the importance of karst and whether the loss of suitable hot caves due to climatic changes and associated sea level changes could explain the extirpation of bat populations in the Caribbean.

FIGURE 4.

Current (A), Holocene (B, ca. 6 ka), and Last Glacial Maximum (C, ca. 21 ka) ecological niche model projections for Mormoops blainvillei in the Caribbean. Models were developed using Maxent v3.3.3 and projected as binary distributions in ArcGIS v10.2. Island boundaries (black outlines) are based on current land areas. See figure 1 for geographic reference.

FIGURE 5.

Current (A), Holocene (B, ca. 6 ka), and Last Glacial Maximum (C, ca. 21 ka) ecological niche model projections for Pteronotus parnellii in the Caribbean. Models were developed using Maxent v3.3.3 and projected as binary distributions in ArcGIS v10.2. Island boundaries (black outlines) are based on current land areas. See figure 1 for geographic reference.

FIGURE 6.

Current (A), Holocene (B, ca. 6 ka), and Last Glacial Maximum (C, ca. 21 ka) ecological niche model projections for Pteronotus quadridens in the Caribbean. Models were developed using M axent v3.3.3 and projected as binary distributions in ArcGIS v10.2. Island boundaries (black outlines) are based on current land areas. See figure 1 for geographic reference.

FIGURE 7.

Predicted habitat suitability in the Caribbean for Mormoops blainvillei, Pteronotus parnellii, and Pteronotus quadridens represented as stability maps of the intersection of ecological niche model climate projections at three time periods: current, Holocene (HOL, ca. 6 ka), and Last Glacial Maximum (LGM, ca. 21 ka). Blue shading indicates areas where each species could have maintained viable populations across time. See figure 1 for geographic reference.

Table 4.

Proportion of present and fossil localities occurring within karst (i.e., karst buffer = 0 km) and at increasing 5 km distances away from karst (i.e., karst buffer = 5, 10, 15, 20, 25, and >25 km).

Previous studies have addressed the usefulness of ENMs to estimate the effect of recent climate change on species distributions and population structure (Waltari et al., 2007; Carnaval et al., 2009; Waltari and Guralnick, 2009). In this study, we developed time-scaled ENMs from the current to the LGM climate and validated them using fossil information for three species of Caribbean bats. Based on previous studies of other Caribbean bats, we hypothesized that climate-based distributions of mormoopid bats would show stability across time. ENMs developed for current climate were highly accurate in all three species of bat with true-positive fractions ranging from 99%–100%. ENM projections to the HOL and LGM for Mormoops blainvillei and Pteronotus parnellii also performed well and showed true positive fractions of 91% and 92%, respectively. This suggests that our model projections accurately represent the known distribution in the present and past (as evidenced by fossils) of these two mormoopid bats in the Caribbean. RRS for Mormoops blainvillei and Pteronotus parnellii suggest a range contraction from LGM to the present. This supports the hypothesis that loss of land area had a negative effect on populations of bat in the Caribbean (Dávalos and Russell, 2012). However, when corrected for island area, changes in RRS are small for land areas that were not affected by sea level change over time, such as the Greater Antilles, and in all species indicate a range expansion from past to present (table 1). At least 73% of the distribution estimates in Mormoops blainvillei and Pteronotus parnellii represent areas of stability and distribution overlap, indicating that a large portion of the land area available over time has been part of the suitable climatic habitat for these species (fig. 7). These results support our hypothesis that climatebased distributions of mormoopid bats show stability in the Caribbean, and also support the hypothesis that extirpation of these bats from localities in the Bahamas was unlikely driven by PHT climate change alone (Soto-Centeno and Steadman, 2015).

In the case of Pteronotus quadridens, we observed hindcasted models with a low level of accuracy and true-positive fractions of 80% for HOL and 60% for LGM. RRS estimates show little change from the LGM to the present. In contrast, there was a trend of range expansion from LGM to the present in RRS estimates corrected for island size, which suggests that estimates of the LGM distribution of P. quadridens for areas not affected by sea level change were much smaller than for the present. This is also shown by our estimates of stability and distribution overlap, where at most 47% of the available suitable climatic habitat was shared with LGM distribution, whereas 80% was shared between CUR and HOL distributions. Fossil evidence indicates that P. quadridens was once extant on north Andros, Great Abaco, and New Providence in the Bahamas—all areas where it is no longer found (Morgan, 2001; Dávalos and Turvey, 2012). However, the age of these Bahamian fossils is uncertain because they have not been radiocarbon dated. Most bat fossils from extirpated populations in the Bahamas are not older than late Holocene (i.e., <4 ky; Soto-Centeno and Steadman, 2015). If P. quadridens fossils were the same age, this pattern may be evidence that this species was able to colonize the Bahamas in the Holocene when climatic conditions were favorable to sustain viable populations, but then became extinct shortly after during the same time period.

Extant mormoopid bats in the Caribbean are known for their nearly exclusive use of hot caves (Silva-Taboada, 1979; Rodríguez-Durán and Soto-Centeno, 2003; Mancina et al., 2007). Because of this tight association, Morgan (2001) hypothesized that the disappearance of mormoopids in the Bahamas was due to the loss of large cave systems and their associated hot microclimate following sea level rise at the end of the Pleistocene. We tested this hypothesis by examining the importance of karst to the distribution of mormoopid bats in the Caribbean. Although karst may seem to be a poor predictor for mormoopid distributions in the Caribbean (i.e., at most 68% of modern localities occur within karst areas), our karst buffer analysis show that >90% of all modern localities lie either in karst areas or within 15 km of the karst edge (fig. 8). Furthermore, the current and hindcasted karst-climate ENMs performed well in each species when corrected for karst area (table 1; fig. 9). This makes sense given that bats are volant animals, and that many bat species have large home ranges and are known to travel long distances from their roost to foraging sites (Rodríguez-Durán, 1984; Horner et al., 1998; Frick et al., 2007; Zeale et al., 2012). In the Greater Antilles, a region known for extensive fluviokarst deposits, it seems clear that proximity to karst is an important factor driving distribution of mormoopid bats.

FIGURE 8.

Map of karst in the Caribbean and localities of Pteronotus parnellii. Fossils are shown in red dots, present localities in green dots, and stippling represents areas of karst. Outlines show buffers around karst at 5 km intervals from 0–25 km (black to light gray shading). Insets represent Hispaniola and Puerto Rico for fossils and present localities to show detail.

The Bahamas present a different problem. Roughly 22 glacial-interglacial cycles that altered sea level and island size in the Caribbean have been documented in the past 2.5 my (Cuffey and Marshall, 2000; Lambeck and Chappell, 2001). Rising and falling sea level during these cycles repeatedly inundated the shallow platforms of the Bahamian banks and likely precluded the formation or long-term existence of hot caves in this archipelago. The KROW database includes Caribbean karst formations that have not been affected by erosion caused by sea level change—i.e., it does not include eolianite karst deposits such as those characteristic of the Bahamas (Hollingsworth, 2006). Therefore, our evaluation involving karst using the KROW database could not be extended to the Bahamas.

FIGURE 9.

Ecological niche model of the intersection of karst and climate in the Caribbean for Pteronotus parnellii. Gray shading indicates areas where karst is available and climate is suitable for P. parnellii. Insets represent (A) the Bahamas and (B) Hispaniola and Puerto Rico to show detail. Locality distributions represent fossils (red dots) and current (green dots) samples.

It is important to point out that caves do exist in the Bahamas even though eolianite karst does not have the geomorphological features that allow hot caves to form as in the Greater Antilles (Brooke, 2001; Mylroie and Mylroie, 2007; Ladle et al., 2012). The colonization of the Caribbean by mormoopids (ca. 3 ma; Dávalos, 2009) predates the last 22 glacial-interglacial cycles and our ENM results suggest that a large proportion of the available suitable climatic habitat remained stable at least since the LGM. Furthermore, recently radiocarbon-dated fossils show that the extirpation of Pteronotus parnellii in the Bahamas occurred in the late Holocene (ca. 3.4 ka; Soto-Centeno and Steadman, 2015). Contrary to Morgan's (2001) hypothesis, this evidence suggests that mormoopid bats could and did survive glacial-interglacial cycles across long time scales in the Caribbean, and that climate change and associated sea level changes might not have contributed directly to extirpation of populations. Although mostly associated with hot cave environments, Mormoops blainvillei and Pteronotus parnellii have been reported to occasionally roost in cool cave conditions (Silva-Taboada, 1979; Gannon et al., 2005). Some bat species typically associated with hot caves (e.g., Nyctiellus lepidus and Pteronotus parnellii) are known to change their roosting behavior by forming tight clusters in bell holes or spreading out on ceilings depending on the temperature and humidity of the cave (Gannon et al., 2005; Tejedor et al., 2005b). In the absence of large hot caves, we suggest that mormoopid bats in the Bahamas were able to use alternative cave roosts, perhaps behaviorally creating suitable roosting conditions by forming tight clusters in bell holes or using small chambers within the eolianite caves characteristic of the Bahamian region.

Our climate-only, karst buffer, and karst-climate analyses suggest that karst is a good predictor for the distribution of mormoopid bats in the Caribbean based on their present distribution. However, evaluating the importance of karst for Mormoops blainvillei, Pteronotus parnellii, and P. quadridens based on climatic and habitat suitability alone cannot account for changes in microclimate within roost sites. Further research is necessary to fully understand the use and importance of hot caves for Caribbean bat populations, and to investigate more thoroughly the possibility that some or all species may use alternative roosts under some circumstances. In this regard, studies of physiology and roosting behavior, as well as more detailed analyses of fossil sites including radiocarbon dating, may shed additional light on the causes of recent change and extirpation of mormoopid bats in the Caribbean islands.