Introduction

As the extent of undisturbed natural ecosystems continues to decline (Morales-Hidalgo et al. 2015), anthropogenically modified and semi-natural landscapes steadily replace them (Seto et al. 2011). Wildlife populations in such landscapes are forced to find resources in “novel ecosystems” that may be more or less suitable than what they replaced (Kennedy et al. 2018; Martínez-Hesterkamp et al. 2018). The novelty of such ecosystems may be the result of new anthropogenic disturbances (i.e., agriculture, urbanization, pollution, fragmentation, etc.) or the alteration of natural disturbance regimes (i.e., fire suppression, flood mitigation, wind breaks, etc.). While unmodified natural areas are likely to have the highest habitat suitability for many wildlife taxa, they may represent only a fraction of a species’ distribution (e.g., Müller et al. 2017). Because many animal populations are forced to adapt to novel ecosystems, conservation objectives must adapt with them. In fact, many species of conservation concern co-occur in areas of high human population density (Luck 2007). For example, cities in Australia contained more threatened species per unit area than non-urban sites; this pattern was especially notable for animals relative to plants (Ives et al. 2016). Urban wildlife populations tend to rely on patches of vegetation amidst a more modified matrix, but different types of green space vary in their utility as wildlife habitat (Gallo et al. 2017). The conservation value of semi-natural areas (i.e., parks, cemeteries, and other urban green spaces) remains an open question, and it is unclear whether fragmented patches of vegetation in a modified matrix can sustainably support an ecological community. Semi-natural areas vary considerably in size, vegetative characteristics, water supply, and degree of landscape modification (Nielsen et al. 2014). Therefore, identifying the internal and external factors determining their utility is crucial for maximizing their conservation potential.

Golf courses represent a relatively large and common semi-natural landscape. As of 2021, > 38,000 golf courses were distributed throughout 206 countries (R&A 2021). Nearly 42% of golf courses globally occur within the United States (16,156 golf courses), making it the number one golfing country in the world by a wide margin (R&A 2021). Given their size and abundance, golf courses alter the environment for both humans and wildlife.

Golf course construction and maintenance result in several negative impacts to local environments (Briassoulis 2010; Palmer 2004; Wheeler and Nauright 2006), including soil composition changes, wetland loss, and groundwater contamination (Doytchev 2019; Winter et al. 2003). Additionally, native vegetation is often replaced with non-native grasses to match the industry standard (McCarty 2018). The large amounts of fertilizers, herbicides, insecticides, and fungicides required to maintain these turf grasses may contaminate nearby bodies of water (Bock and Easton 2020; Grande et al. 2019; Kunimatsu et al. 1999; Sudo et al. 2002), and the irrigation requirements can severely deplete groundwater reserves and natural reservoirs (Platt 1994). Accidental exposure to pesticides has contributed to high-mortality events in waterfowl (Littrell 1986; Stone and Knoch 1982; Zinkl et al. 1978), passerines, and bats (Stansley et al. 2001). However, golf courses can also have conservation value (Hodgkison et al. 2007; Tanner and Gange 2005), supporting populations of threatened or endangered birds (Rodewald et al. 2005; Smith et al. 2005; Terman 1997), amphibians (Montieth and Paton 2006), and mammals (Ditgen et al. 2007), and sometimes exhibiting demographic performance metrics comparable to protected wildlife areas (Winchell and Gibbs 2016). Golf courses may also maintain habitat connectivity in human-dominated environments. Particularly if they retain patches of native habitat, golf courses may provide a refuge for animals dispersing through otherwise developed landscapes.

Discordant conclusions about the effects of golf courses on wildlife suggest that external factors influence their ecological value. Because golf course features tend to converge on a similar design (often referred to as the Augusta National Syndrome; Millington and Wilson 2016), we hypothesize that the surrounding landscape strongly influences the use of golf courses. Where the surrounding matrix is dominated by natural cover, golf courses may have lower benefit to resident wildlife. However, in highly urbanized areas, golf courses should function as habitat refugia and support biodiversity in much the same way as other urban green spaces (Petrosillo et al. 2019).

Compared to more readily observable taxa (i.e. birds, insects, and amphibians), bats are understudied on golf courses (Petrosillo et al. 2019), but they may be especially attracted to golf courses as foraging sites due to the high densities of arthropod prey within the highly heterogeneous habitat types within out-of-play areas (Dale et al. 2020; Mata et al. 2017; Saarikivi et al. 2010; Tanner and Gange 2005). Because bats are frequently reported to make use of comparable urban green spaces (Suarez-Rubio et al. 2018), bats may live on or frequently visit golf courses to feed or roost (Burgin and Wotherspoon 2009; Gitzen et al. 2001). For example, bats occurring in urban areas of London, UK were observed feeding within golf courses; however, light pollution from the surrounding development restricted their use to the course interior (Fure 2006). Australian bats increased golf course use over time in an anthropogenically modified landscape (Burgin and Wotherspoon 2009). North American bats, including the endangered Florida bonneted bat, Eumops floridanus (Webb et al. 2021), have also been documented on golf courses (Bazelman 2016; Wallrichs 2019). Although these accounts exist, many are single observational reports rather than quantitative studies (but see Wallrichs 2019). Where robust studies occurred, authors speculated that the pseudo-natural features of the course provided an escape from a less hospitable surrounding landscape (i.e. urban refugia effect; Burgin and Wotherspoon 2009). These previously mentioned studies call for further investigation into the conservation value of golf courses and the novel landscapes in which they are situated. As novel landscapes gradually replace natural ones, understanding how bats use them is likely to become increasingly important for conservation.

We tested our hypothesis by documenting the activity of North American bats on golf courses situated across a gradient of landscape matrices. In this study, we investigated the relationship between surrounding landscape characteristics and bat activity (species-specific and community-level) and foraging rates (community-level) on golf courses. We monitored bat activity levels on golf courses that varied in their surrounding landscape features across New Jersey, USA for two consecutive years using standardized passive acoustic monitoring procedures. In agreement with the urban refugia effect observed in other studies, we predicted that generalist, disturbance tolerant bat species would dominate golf courses surrounded by anthropogenically altered landscape matrices. Additionally, we predicted that golf courses situated in modified landscapes would experience greater total bat activity and foraging rates, while more natural landscapes would correlate with lower golf course use.

Materials and methods

Study area

We conducted our study on 11 golf courses distributed across 10 counties throughout New Jersey, USA (Fig. 1). Sites ranged in size from 10.2 ha to 100.1 ha and were located a minimum of 9.8 km apart. Because we were interested in examining the relationship between the surrounding landscape and golf course use by bats, we deliberately chose sites that represented a gradient of land use types. Landscapes surrounding golf courses ranged from predominantly agricultural areas (e.g., Eastlyn Golf Course; hereafter, “Eastlyn”), predominately urban areas (e.g., Overpeck Golf Course; hereafter, “Overpeck”), to predominantly forested areas (e.g., High Point Golf Club; hereafter, “High Point”).

Fig. 1
figure 1

Golf courses in New Jersey, USA surveyed for bats during June through August 2019–2020. Black dots indicate course name and location. Sea Oaks Country Club closed permanently following the 2019 season and was replaced in 2020 by the nearby Atlantis Golf Course

Full size image

Survey protocol

We surveyed golf courses for bats between June and August in 2019–2020. To develop our survey protocol, we combined guidance presented in two widely-accepted bat monitoring approaches: North American Bat Monitoring Program (NABat; Loeb et al. 2015) and the US Fish & Wildlife Service’s Range wide Indiana bat and northern long-eared bat summer survey guidelines (U.S. Fish and Wildlife Service 2023), ensuring that surveys were replicated both spatially throughout the golf course and temporally across years (de Torrez et al. 2017). At each study site, we deployed two Pettersson D500X (Pettersson Elektronik AB, Sweden) direct recording full-spectrum acoustic detectors with omni-directional microphones mounted at least 3 m above the ground and placed at least 10 m away from vegetation (Loeb et al. 2015). Microphones were aimed towards the open space above either bodies of water or the fairway adjacent to a forest edge. Microphones deployed simultaneously were positioned at least 200 m apart from one another. We scheduled detectors to passively record bat calls from 15 min prior to sunset until 15 min after sunrise over four total nights (eight detector-nights per course per year) when little/no precipitation was predicted in the forecast and winds were below 10 km/hr. To reduce the number of nontarget noise recordings, we programmed detectors with a 500 kHz sampling rate and a medium trigger sensitivity (for detailed system settings, see Online Resource 1). Once triggered, detectors recorded sound for 3 s and stored recordings as .WAV files for subsequent analysis.

Species classification

We analyzed all detector recordings using SonoBat version 4 and the northeastern North America regional library (Szewczak and Szewczak 2017). We considered a bat call to be any sound produced by a bat, primarily echolocation pulses intended for navigation and foraging. To eliminate extraneous noise files and reduce the number of files that needed manual verification, we used SonoBat’s file-scrubbing utility set to the default medium call-quality threshold.

All files that remained after scrubbing were processed using SonoBat’s call attributer and compared against the northeastern regional library of bat calls to classify them by species when possible. Measurements to calculate these classifications had an acceptable call quality threshold of 0.80, a species decision threshold of 0.90, and a maximum of 32 echolocation pulses considered per file. For recordings that were below the quality threshold of automatic classification (0.90) but were still clearly visible on the sonogram, a general label was automatically attributed to the file (either “high frequency unknown” or “low frequency unknown”). Similarly, if there were fewer than four clear echolocation pulses in a file, it was labeled generally because a confident species identification cannot be made with so few calls. After files were automatically classified, all calls were manually vetted to verify species classification. We manually identified recordings conservatively to minimize false-positive species detections, adhering to a rubric of known call characteristics for each species. Calls assigned to a species not known to occur in New Jersey (e.g., Corynorhinus rafinesquii; IUCN 2008) were manually reclassified as unknown. Because of the highly overlapping call characteristics between Myotis lucifugus and M. sodalis, and because there were so few recordings attributed to M. sodalis, we combined classifications for M. lucifugus and M. sodalis into a “Luso” category (Szewczak and Harris 2013) and treated it as a single species for all subsequent analyses. Calls with a general label (“high frequency unknown” or “low frequency unknown”) were included while calculating total bat activity, but not for species-specific activity measures. We defined nightly bat activity as the count of all bat calls (both species-specific and the sum of all species) detected at a single course each night.

Foraging activity

We quantified foraging rate at each golf course, defined as the count of all terminal buzzes detected at a single course each night. Bats tend to change their echolocation behavior as they approach a food item (Ratcliffe et al. 2013), where evidence of hunting behavior is characterized by a slight drop in the characteristic frequency, increased pulse repetition rate, and increased bandwidth and shorter call duration of each call in the “terminal buzz” (Schnitzler and Kalko 2001). Therefore, we used SonoBat’s sonogram viewing window to visually examine every call file for indicators of feeding activity. Buzzes are readily visible on the SonoBat viewing window and can be heard as a distinct cadence when the audio file is played at 10x reduced speed. Because a terminal buzz is accompanied by changes in call characteristics, many sound files containing a terminal buzz did not meet the criteria for reliable species classification. Therefore, we we did not attempt to model differences in foraging activity among species.

Relationship between the surrounding landscape and use of golf courses by bats

We used remotely sensed data and GIS to characterize the surrounding landscape variables likely correlated to the use of golf courses by bats. All spatial data manipulations were performed using ArcMap version 10.8.1 (ESRI 2020). We first projected the polygon shapefiles of study golf courses obtained from the New Jersey Department of Environmental Protection (NJDEP), Division of Science, Research and Technology digital repository (Online Resource 2; NJDEP 2001). The bat species in our study tend to travel approximately 2 km from their roost each night to access foraging areas (Brigham 1991; Crampton and Barclay 1998; Elmore et al. 2005; Sparks et al. 2005; Walters et al. 2007); therefore, we created a 2-km buffer around the perimeter of each golf course to represent the approximate roost-to-forage distance of local bat species. Thus, all bats roosting in our 2-km buffer could reasonably be expected to consider the golf course as a potential roosting or foraging habitat.

We also downloaded and projected land use data for the state of New Jersey from the NJDEP (Online Resource 2; NJDEP 2019). This dataset classified the state of New Jersey into six land use types: urban, agriculture, forest, water, wetland, and barren land (i.e., bare rock and sand). However, these general land use types may contain more specific land uses hypothesized to be more relevant to bats. Thus, we used the Reclassify tool in ArcMap to reclassify some of these land use types into more specific categories. For example, the label “urban” was used to describe a wide range of land types, including industrial areas, cemeteries, residential areas, and roads. Although most bats tend to avoid highly urban areas (Russo and Ancillotto 2015), they may actively select habitat near small roads and suburban areas (Threlfall et al. 2012). Accordingly, we used these fine-scale labels to reclassify low-density residential areas as a new category, “suburban.” Some bat species may also prefer to forage over open fields of grass or other low-lying vegetation (Barclay 1985; Patriquin and Barclay 2003); however, that type of habitat was not represented by the existing six land-use types. We reclassified the features representing “Cemetery”, “Cemetery on Wetland”, “Recreational Land”, “Athletic Fields (Schools): Community Recreation Areas”, “Managed Wetland in Built-up Maintained Recreation Area”, “Old Field (< 25% Brush Covered)”, “Phragmites Dominated Old Field (2002)” as the new land use label “open fields”. Ultimately, the final eight land use types considered in our study were: (1) urban; (2) suburban; (3) agricultural; (4) open fields (non-agricultural); (5) forest (greater than 10% canopy closure); (6) water; (7) wetland; and (8) barren land (beaches, quarries, and bare rock). We overlaid and clipped the NJDEP land use polygon shapefile to each 2-km buffer and converted the resulting polygons to rasters with a 10-m resolution.

Bats also make extensive use of streams (Bergeson et al. 2013; Kniowski and Gehrt 2014), which function as corridors for travel, insect-rich foraging habitat, and a source of drinking water (Pauli et al. 2017). Therefore, we quantified the total length of streams with in the 2-km buffer surrounding each site using the NJ National Hydrography Dataset (NHD) Waterbody and Stream Network (Online Resource 2; NJDEP 2010). This dataset is a subset of the larger NHD dataset that is specific to New Jersey, and we used the most recent 2011 NHDFlowline feature class. We further refined this dataset by considering only streams that are within 100 m of a forested landscape, which retained the relevant spatial data required for analysis while also eliminating legacy land use characterization within the shapefile (i.e., former tidal land that is currently inundated).

From the spatial data gathered, we quantified all landscape characteristics using Fragstats (McGarigal et al. 2012) with the exception of total length of forested streams in the surrounding landscape (calculated directly in ArcGIS). Because forests represent especially important resources for bats (Lacki et al. 2007), we also used Fragstats to determine both the density of forest edge in the surrounding landscape and the largest patch index (LPI) of forest, which is used as a measure of dominance over a landscape. The LPI approaches 0 as the largest patch of forest becomes increasingly small, whereas a forest LPI of 100 would indicate that the largest patch of forest is the same size as the study area.

Statistics and modeling

Analyses and data manipulations were done in R studio version 2022.12.0 (RStudio Team 2020). We first checked for correlations among environmental variables using Pearson’s r (Lee Rodgers and Nicewander 1988), eliminating all but one variable involved in a correlation above 0.70. Forest edge density, percent forested land, and forest largest patch index were highly correlated; therefore, we eliminated forest edge density and percent forested land. We used the R package “lme4” (Douglas Bates et al. 2020) to construct sets of generalized linear mixed effects models (GLMMs)to test the relationship between the surrounding landscape and golf course use by bats. Models included the count of verified total bat recordings per night (as a measure of total nightly bat activity), recordings of each individual species per night (as a measure of species-specific nightly bat activity), and terminal buzzes per night (as a measure of foraging rate) as the response variables. This resulted in eight datasets: total nightly bat activity, total foraging rate, and six subsets of species-specific nightly bat activity. We sought to examine community-level and species-specific habitat associations with the percent urban land, percent open field, percent agricultural land, percent suburban land, forest largest patch index, and length of streams on our response variables. We also included detector location to indicate whether or not the survey location was adjacent to water, as some bat species tend to prey upon aquatic insects (Clare et al. 2011; Maslo et al. 2022; O’Rourke et al. 2021). Finally, to account for intra-course differences in bat activity, as well as repeated sampling sites, we considered golf course and year as random effects. In 2020, New Jersey golf courses were closed until early May due to restrictions relating to COVID-19. Because the closures did not overlap with our survey period, we did not expect this to influence our results.

We constructed 25 a priori GLMMs using a negative binomial error distribution (Online Resource 3) to investigate competing hypotheses regarding the effects of landscape composition on golf course use by bats. Global models contained all fixed effects, which were scaled by dividing by the standard deviation of each effect and centered. We ranked candidate models using AICc (Online Resources 4–11) and averaged those within ΔAICc < 2 (Burnham and Anderson 1998) using the MuMIn package in R (Barton 2020) for each of the eight data sets. The residuals of top models were tested for zero inflation, correct distribution, dispersion, and outliers using the DHARMa package (Florian Hartig and Lohse 2021). None of the top models showed significant zero inflation or deviations from expected residuals. We considered variables within the top models to be significant predictors of the dependent variable if the 95% confidence interval of the model estimate did not contain zero.

Results

Community composition of bats on golf courses

After accounting for detector malfunction or battery failure, we secured 132 detector nights across the 2-yr period. We recorded 33,272 total bat echolocation call sequences (course averages ranging from 46.1 to 792.6 calls per detector night; Table 1), with approximately 73% of our recordings occurring in 2020. The most recordings at a site for a single season occurred at Atlantis Golf Course (hereafter, “Atlantis”) in 2020 (N = 5,548), whereas the fewest recordings for a single season occurred at Rutgers Golf Course (hereafter, “Rutgers”) in 2020 (N = 375). We consistently recorded high nightly activity and foraging rates at Harbor Pines Golf Club (hereafter, “Harbor Pines”), Eastlyn, and Overpeck in both years. Avalon Golf Club (hereafter, “Avalon”) and High Point had consistently low total nightly activity, but nightly foraging rate was highly variable between years at these sites.

Table 1 Nightly average bat activity per site per year (standard error in parentheses). When possible, recordings were identified to a species-level classification. Number of terminal buzzes per site is a measure of foraging activity
Full size table

Eptesicus fuscus was the most common bat species observed in our study, representing ~ 77% of all identifiable recordings. In contrast, we recorded only two recordings each of Myotis leibii and M. septentrionalis. Therefore, these two species were eliminated from subsequent analyses. Despite having > 500 recordings, we also excluded Nycticeius humeralis from our species-specific models because New Jersey is bisected by the northern edge of its geographic range (BCI 2021) and so it was not present at all sites; therefore, N. humeralis use of golf courses in our study is confounded with range limits. N. humeralis recordings were still included in our measure of total nightly bat activity.

We found E. fuscus, Lasiurus borealis, and Lasionycteris noctivagans at all sites in both 2019 and 2020. While “Luso” calls were predominantly rare, we detected this species complex consistently at Harbor Pines and High Point in both 2019 and 2020. Similarly, Perimyotis subflavus was consistently observed only at Overpeck and, to a lesser extent, High Point and Eastlyn. The number of bat species present ranged from 4 to 8, with a mean of 6 bat species per course. The most species were recorded at Atlantis, Eastlyn, and Overpeck and the fewest species were recorded at Rutgers and Cream Ridge Golf Course (hereafter, “Cream Ridge”), which each only contained the four most common species. The same four common species (E. fuscus, L. borealis, L. cinereus, and L. noctivagans) were observed at all sites in both years except Avalon in 2019 (where L. cinereus was absent).

Influence of surrounding Landscape on Bat Activity

Top models describing total bat activity demonstrated reductions in total bat activity with increasing percentages of open field (-0.64, 95% CI: -1.07, -0.21), agriculture (-0.63, 95% CI: -1.10, -0.17), and forest LPI (-0.34, 95% CI: -0.66, -0.03) within the surrounding landscape (Fig. 2). Total bat activity was lower at microphones placed near open water compared to those that were placed at forest edges (-0.71, 95% CI: -1.18, -0.24).

Fig. 2
figure 2

Model averaged estimates (dots) and 95% confidence intervals (error bars) for variables predicting total nightly bat activity on golf courses. Black dots and error bars indicate significant effects

Full size image

Increases in percentage of surrounding urban land were associated with greater on-course activity of L. borealis (0.94, 95% CI: 0.19, 1.73) and P. subflavus (3.77, 95% CI: 1.71, 5.85; Fig. 3). While not statistically significant, the activity of E. fuscus (0.35, 95% CI: -0.14, 0.84) and L. noctivagans (0.32, 95% CI: -0.24, 0.88) also slightly increased along with increased urban percent. Notably, the majority of P. subflavus recordings occurred at Overpeck, the most urban course (28.5% urban land cover) in the study. Meanwhile, increasing extent of surrounding suburban land was significantly associated only with L. borealis, serving to reduce on-course activity (-0.59, 95% CI: -1.17, -0.01). Increasing percent of open fields was significantly related to reduced activity on courses for E. fuscus (-0.69, 95% CI: -1.21, -0.17), L. borealis (-0.88, 95% CI: -1.55, -0.20), and L. noctivagans (-0.66, 95% CI: -1.21, -0.11). Higher agricultural coverage had a similar negative association with the activity of E. fuscus (-0.68, 95% CI: -1.23, -0.13), L. noctivagans (-0.65, 95% CI: -1.26, -0.04), and the Luso complex (-1.66, 95% CI: -3.16, -0.15); however, P. subflavus activity significantly increased (3.24, 95% CI: 1.13, 5.35) on courses surrounded by higher amounts of agriculture.

The length of forested streams in the surrounding landscape appeared in the top models for four species but were significant predictors only of L. borealis (-0.86, 95% CI: -1.30, -0.41) and L. cinereus (-0.78, 95% CI: -1.41, -0.14) activity (Fig. 3). Similarly, greater forest LPI appeared in the top models for several species, but was only associated with increased activity for Luso (1.09, 95% CI: 0.02, 2.16) and P. subflavus (3.93, 95% CI: 1.52, 6.34).

Fig. 3
figure 3

Model averaged estimates and 95% confidence intervals for predictors influencing species-specific bat activity on golf courses: (A)E. fuscus, (B)L. borealis, (C)L. cinereus, (D)L. noctivagans, (E)M. lucifugus/sodalis, and (F)P. subflavus. Points represent the scaled and centered estimated values, and black indicates statistical significance

Full size image

Bat foraging activity on Golf Courses

We identified 1460 terminal buzzes over the two seasons. Harbor Pines, Eastlyn, and Overpeck had consistently high nightly foraging rates. Atlantis was only surveyed in 2020, but it had similarly high foraging rates. Nightly foraging rate increased with greater extent of surrounding suburban land (0.28, 95% CI: 0.01, 0.56) and decreased with agriculture (-0.63, 95% CI: -1.04, -0.22), open field (-0.48, 95% CI: -0.87, -0.09), and length of forested streams (-0.30, 95% CI: -0.59, -0.02) in the surrounding landscape (Fig. 4).

Fig. 4
figure 4

Model average estimates and 95% confidence intervals of the predictors for bat foraging rate. Black points represent the scaled and centered estimated values

Full size image

Discussion

Relationship between the surrounding Landscape and Use of Golf Courses by bats

Previous studies have focused on the value of golf courses in an urban setting, but to our knowledge none have quantitatively compared bat activity on golf courses across a gradient of landscape mosaics. In the present study, bat activity on golf courses was significantly associated with the surrounding landscape features. We consistently observed less bat activity on golf courses when the surrounding landscape contained more open spaces. Almost all models demonstrated decreased bat activity and foraging rate on golf courses when the surrounding landscape contained more open fields and/or agricultural land (the sole exception being P. subflavus). Vegetation in fields and agricultural areas have little canopy cover, reduced vertical stratification, and uniform ground cover; much like the expansive turf characteristic of golf courses. Because these landscape features are structurally similar to habitats typically found on golf courses, bats may divide their time among open green spaces and golf courses in proportion to their abundance in the landscape. We posit that golf courses, similar to fields and agricultural areas, represent open green space; therefore, bats may use golf courses more often when they have fewer alternatives in the surrounding landscape.

In agreement with our predictions, developed land within the surrounding landscape was associated with greater overall bat activity and foraging rate on golf courses. Although the strength and statistical significance of this relationship varied among bat species, high percentages of developed land were always associated with increased bat activity on golf courses when present in our models (apart from L. borealis, which was negatively associated with suburban landscapes). Golf courses in urban areas appear to serve as habitat refugia for bats and other species (Burgin and Wotherspoon 2009; Saarikivi et al. 2010; Threlfall et al. 2016; Wurth et al. 2020), and therefore may support biodiversity in highly modified landscapes (Petrosillo et al. 2019).

The two rarest bats in our analyses were the multispecies complex “Luso” (M. lucifugus and M. sodalis) and P. subflavus. Both Luso and P. subflavus populations declined substantially following the introduction of Pseudogymnoascus destructans (Pettit and O’Keefe 2017), the pathogen causing white-nose syndrome. M. sodalis is listed as endangered in the United States under the U.S. Endangered Species Act of 1973, as amended (16 USC 1531), and P. subflavus is being considered for endangered status (USFWS 2017). These species are also listed as either threatened or endangered in Canada (COSEWIC 2013). Thus, it is unlikely that their scarcity on golf courses is solely related to the surrounding landscape. Nevertheless, these species were primarily documented on golf courses adjacent to large tracts of forest; unsurprising, given their habitat preferences (Ethier and Fahrig 2011; Farrow and Broders 2011). While P. subflavus can be flexible in its habitat uses (Ellis et al. 2002; Ethier and Fahrig 2011; Loeb and O’Keefe 2006), these species are thought to prefer forested areas, and they tend to avoid landscapes with more open space (Bergeson et al. 2013; Farrow and Broders 2011; Nelson and Gillam 2017; Sparks et al. 2005). Because golf courses themselves are a series of connected open spaces, we may not expect to find high Luso or P. subflavus activity on golf courses in most landscape contexts. Further, golf courses that fragment or replace undisturbed patches of forest may displace populations of Luso and P. subflavus. When these species were present, it may have been because they were using golf courses primarily as a corridor as they commuted to and from their foraging ground in the forest. Further detailed habitat studies of P. subflavus in the northeastern United States are clearly warranted.

Implications

Golf courses have been shown to support wildlife and, therefore, may play an important role in wildlife conservation in human-altered landscapes. In the United States, the majority of urban areas are growing faster than their human populations might predict (Bounoua et al. 2018), and these developed areas may interfere with valuable ecosystem services. Therefore, golf courses (and similar anthropogenic green spaces) may demand more conservation attention over time to help mitigate the effects of habitat loss and fragmentation.

Our results indicate that the conservation value of modified green spaces, such as golf courses, depends on landscape context. We demonstrate that the surrounding matrix is significantly associated with bat activity on golf courses. Bat home-ranges may be less likely to overlap with golf courses when suitable bat habitat exists elsewhere in the nearby landscape. Supporting this conclusion, we observed lower bat activity on golf courses when there are more open landscapes in the surrounding area. To a lesser extent, we also observed higher bat activity on golf courses in more developed landscapes. Indeed, a well-placed golf course may provide a stable patch of native trees for bats dispersing throughout dynamic urban landscapes.

Most bats in our study were detected on all golf courses surveyed, though their activity levels differed by site. However, some species (and notably, those of most conservation concern) were only present at certain sites. M. lucifugus, M. sodalis, and P. subflavus are listed by the International Union for Conservation of Nature (IUCN) as endangered, near-threatened, and vulnerable, respectively (Arroyo-Cabrales and Ospina-Garces 2016; Solari 2018, 2021). Determining factors that increase or decrease their use of green spaces (e.g., golf courses) is becoming increasingly relevant as the global human footprint expands. Depending on the landscape context, golf course construction may either fragment or connect bat habitat. In a densely forested ecosystem, golf courses and their associated infrastructure (parking lots, club houses, and roads) may become a barrier for species that specialize in forest interiors. Remnant fragments of forest may not adequately support species that require large patches of forest in their home range. Our study suggests that P. subflavus and M. lucifugus/sodalis require landscapes with continuous patches of intact forest, which is supported by studies of roosting and foraging habitat preferences for these species (Bergeson et al. 2013; Farrow and Broders 2011; Gardner and Cook 2002; Perry et al. 2007; Watrous et al. 2006). Therefore, minimizing the extent to which golf courses fragment forests may be an important goal for bat management in an increasingly fragmented landscape.

Although our study did not address the on-course variability of bat roosting and foraging habitat, proper management of golf courses that serve as habitat refugia could further increase their utility for bats. Tree-roosting bats make use of foliage, crevices, cavities, and exfoliating bark of dead trees (Drake et al. 2020). To improve roosting habitat quality for Myotis species, it is recommended that golf courses retain standing dead trees as often as possible, provided that they do not pose a risk to human health or property (Lacki and Schwierjohann 2001). Indeed, the structure and distribution of trees and forests are likely to be important for both roosting and foraging habitat (Froidevaux et al. 2016; Yates and Muzika 2006). Specifically, Myotis species tend to prefer forest management practices that avoid creating large canopy gaps, such as thinning and small regenerative cuts that promote structural heterogeneity (Divoll et al. 2022). Improving the connectivity of forest patches within the golf course by allowing natural regeneration in out-of-play areas of the course may increase habitat value for forest-dwelling bats. Additionally, bats may congregate over bodies of water while foraging for arthropod prey (Korine et al. 2016). Our study indicated that the amount of water in the surrounding landscape was an important predictor of bat foraging activity on golf courses. When streams are lacking in the surrounding landscape, bats may be using water bodies on golf courses as both foraging grounds and a source of drinking water. Water sources are present in urban, natural, and agricultural settings, but the quality of water may differ significantly (Baker 2006; Tong and Chen 2002). Maintaining acceptable water quality on golf courses could drastically improve the value of the wildlife habitat that it provides. Reducing run-off of pollutants like fertilizer, road salt, and pesticides should be a high priority of any golf course seeking to participate in conservation action.

Because golf courses are so numerous, allocating equal conservation effort to all courses would be impractical and inefficient. From a landscape perspective, building golf courses in highly developed landscapes can add valuable habitat refugia for bats. Similarly, fragmenting intact forested landscapes with a golf course should be avoided, where possible. Finally, maintaining bat-friendly habitat features on the course can increase suitability for resident bat populations.

Limitations

Because our study does not compare bat activity on golf courses to reference sites, we are unable to draw conclusions regarding habitat quality of golf courses for bats. Additionally, bat species may be active on golf courses in proportion to their presence in the surrounding landscape rather than the features of the landscape itself. The use of reference sites in the surrounding landscape would also provide insights on the relative rarity of each species in the area. We also did not account for all internal differences among golf courses. Golf-course scale factors (i.e., tree composition and pesticide use) may influence which bat species visit each course. Further testing using controlled experiments may further elucidate the causes of differing bat activity on golf courses.

Conclusion

We demonstrate here that the landscape context surrounding a golf course may be used to predict bat activity on the course in this state. While they are not likely to effectively replace natural habitat for many species, golf courses can provide foraging and roosting opportunities for bats, particularly in heavily developed landscapes. They can also serve as travel corridors between patches of undeveloped habitats.

Our findings suggest that golf courses situated in landscapes with fewer open spaces and greater development may provide important habitat for many bat species. Habitat management practices (i.e., retention of suitable roost trees, maintenance of forest connectivity, and construction of bat roosting boxes) on such golf courses are expected to yield greater bat conservation benefits than similar practices on courses in more natural landscapes. We recommend that the surrounding 2-km landscape be used to assess golf courses (and likely other man-made green spaces like parks, cemeteries, and courtyards) for their potential to provide bat habitat in the eastern United States. Other parts of the world with different bat communities may need to adjust the radius of analysis to better match the habits of local species.