ANIMAL BEHAVIOUR, 1998, 56, 909–917 Article No. ar980858

Roost selection in the pipistrelle bat, Pipistrellus pipistrellus (Chiroptera: Vespertilionidae), in northeast Scotland

E. V. JENKINS*, T. LAINE†, S. E. MORGAN*, K. R. COLE* & J. R. SPEAKMAN* *Department of Zoology, University of Aberdeen †Department of Biology, University of Oulu

(Received 20 November 1997; initial acceptance 30 December 1997; final acceptance 27 March 1998; MS. number: 5707R)

ABSTRACT

Availability of suitable roost sites may limit bat distribution and abundance. We compared nine internal and 26 external features of 21 known roost buildings with those of 17 random buildings in northeast Scotland, U.K. (57)N) to assess whether pipistrelle bats, Pipistrellus pipistrellus (55-kHz phonotype) are selective in their use of roosts. Bats did not select roosts with specific structural attributes. Compared with random buildings, roosts were closer to a tree over 10 m tall and had a greater percentage cover (of trees over 10 m tall) within a radius of 50 m. Trees may provide benefits by providing shelter and thus ameliorating the microclimate of the roost. Cover may also provide protection from predators so that bats are able to emerge earlier, thus increasing foraging time. Bats in the present study emerged 11 min earlier from roosts with 29% cover than from roosts with 3% cover, potentially gaining as much as 10% of their daily energy requirements in this extra time. Roosts were also more likely to have linear vegetation elements leading away from them. These features may be important for navigation, foraging or predator avoidance. Compared with random buildings roosts were closer to and surrounded by a greater area of deciduous woodland within a radius of 0.5 km, had a greater area of coniferous woodland within a radius of 0.5 km, and were more likely to be found within 0.5 km of a major river. All these factors are likely to be beneficial for foraging. A logistic regression model indicated that percentage cover within 50 m of the building was the best predictor of the presence or absence of bats roosting in a building.  1998 The Association for the Study of Animal Behaviour

During the summer months, temperate zone bats spend recently, artificial structures, such as houses, have been up to 20 h of each day within a day roost. These roosts widely used and some species roost almost exclusively in provide protection from bad weather (Vaughan 1987) these sites (e.g. Pipistrellus pipistrellus: Thompson 1992; and from predators (Fenton 1983), and they also serve as Plecotus auritus: Entwistle et al. 1997). sites for social interaction with conspecifics (Morrison The specific type of roost site selected by different bat 1980). They are, therefore, one of the most important species may be determined by the morphology of the bat features of a bat’s environment, and the choices made by (Vaughan 1970), the temperature or humidity within the bats with respect to the type and location of roost sites are roost (Fenton & Rautenbach 1986; Churchill 1991; likely to have a major impact on their survival and fitness Entwistle et al. 1997), the proximity of the roost to (Vonhof & Barclay 1996). suitable foraging and drinking areas (Tuttle 1976; The majority of bat species roost inside natural struc- Speakman et al. 1991; Entwistle et al. 1997), or other tures such as caves (e.g. Tadarida brasiliensis: Davis et al. features of the landscape surrounding the roost (Wunder 1962), rock crevices (e.g. Antrozous pallidus: Vaughan & & Carey 1996). O’Shea 1976), tree holes (e.g. Nyctalus noctula: van Heerdt The risk of predation may also influence roost selection & Sluiter 1965) and spaces under bark (e.g. Myotis daub- by bats. The influence of predation on roost selection has entonii and M. bechsteinii: Cerveny´&Bu¨rger 1989). More been studied previously in tree-roosting bats (Vonhof & Barclay 1996), but not in bats using artificial structures. Correspondence: J. R. Speakman, Department of Zoology, University of Vonhof & Barclay (1996) found that tree-roosting bats in Aberdeen, Tillydrone Avenue, Aberdeen AB24 2TZ, U.K. (email: [email protected]). E. V. Jenkins is now at the Department of southern British Columbia (49)38*N) roosted in trees that Biology, University of York, PO Box 373, York YO10 S7W, U.K. T. were taller than a random sample of available trees, and Laine is at the Department of Biology, University of Oulu, Linnanmaa, they suggested that they did this to avoid terrestrial 90570 Oulu, Finland. predators such as weasels (Mustelidae). Aerial predators 0003–3472/98/100909+09 $30.00/0 909  1998 The Association for the Study of Animal Behaviour 910 ANIMAL BEHAVIOUR, 56, 4

such as owls, hawks and falcons are likely to be a greater conifer plantation. There were three small rivers called threat than terrestrial predators to bats that roost within the Bervie water, Cowie water and Carron water, and artificial structures. Therefore a different set of criteria many streams. may determine roost selection. Observations suggest that bats are most frequently attacked by raptors just after evening emergence and as Roost Sites they travel to and from the roost (Gillette & Kimbrough We conducted a survey of bats over three summers, in 1970; Fenton et al. 1994). Bats may reduce the risk of 1995, 1996 and 1997. Survey work involved placing predation by emerging later, when it is darker and they posters in local shops, in the villages and towns, request- are less likely to be detected by diurnal avian predators ing that people with known bat roosts in their houses (or (which do not have a high visual acuity in darkness; Fox other sites) contact us using forms which were also left in et al. 1976). Indeed, it has been suggested that nocturnal- the shops. Additionally, we sent out letters to 78 farms in ity in bats is largely due to the risk of predation from the region in 1995 and 30 in 1997 requesting information diurnal avian predators (Speakman 1995). However, peak on bats. This led to the identification of 28 summer day abundances of many aerial insects often occur before roosts used by P. pipistrellus (both 45-kHz and 55-kHz nightfall (Speakman 1995; Rydell et al. 1996), so bats phonotypes: Jones & van Parijs 1993) at 16 locations emerging earlier may increase their foraging times and (each location had one or more buildings occupied by food intake rates. The presence of trees near the roost may bats) within the study area. provide commuting bats with protection from aerial We visited each of these 28 day roosts on at least one predators during the early part of the evening (Limpens occasion (a maximum of three occasions) between 12 et al. 1989), and may allow bats to emerge earlier, thus June and 16 July 1997 to confirm that bats were present. benefiting from an extended and more profitable Counts began approximately 15 min before sunset. foraging period (Jones et al. 1995). Observers waited near the emergence holes and noted the We investigated roost selection in the common pipis- time at which the first bat emerged and the light level, as trelle, P. pipistrellus (a species that commonly roosts well as the species of the bats. Only four species of bats are within artificial structures) in northeast Scotland, U.K. found locally and, using heterodyne bat detectors (Batbox (57 N). Entwistle et al. (1997) have investigated roost ) III or QMC S25; Waters & Walsh 1994) we could selection in the long-eared bat, Plecotus auritus,inan easily determine the species of emerging bats from the adjacent area of northeast Scotland, by comparing the frequency of the echolocation calls. structural features and the surrounding habitat of known roost buildings with randomly selected available build- ings unoccupied by bats. We used a similar approach in Random Buildings the present study. Additionally, we investigated the influ- ence of trees near the building on roost selection and We selected a random sample of 17 houses within the upon emergence behaviour. study area, which were not occupied by bats (to be used as a control sample), by choosing the nearest house to a randomly generated grid reference (using random num- METHODS bers). Only houses where the householders were abso- Study Area lutely certain that there were no bats in their house were used. If the householders regularly saw bats near the The study area was located on the northeast coast of house (N=5), or if dead bats had been found on or near Scotland (56)45*N, 2)30*W) 20 km south of Aberdeen, and the property (N=2), we did not use those houses as extended 20 km from north to south and between 10 and controls and selected alternatives. 20 km east to west, covering a land area of approximately 300 km2. The principal topographic features were low, rolling hills (less than 300 m above sea level), the Features of the Buildings majority of which were farmed as either arable or pasture Features (listed in Table 1) of all 28 roosts and 17 land. The total human population of the area was control buildings were recorded. Since householders were approximately 20 000 (ca. 70/km2). The majority of these often unsure of the exact age of their house, we classed lived in two coastal towns (12 000 and 6000 individuals, the age of the property into one of four categories (see respectively). In addition, there were several small vil- Table 1). lages, each with fewer than 1000 residents, and 90 farms. Population densities varied between 0 and 2500 people 2 per km across the area. Maps dating to 1674 indicate that Predation/Isolation Variables the area has been almost completely clear of woodland for at least 300 years. There is evidence of human habi- Bats avoid open areas for both commuting and forag- tation in the region dating back 5000 years, and most of ing, preferring to travel along linear vegetation (Ekman & the woodland might have been cleared when the Romans de Jong 1996; Entwistle et al. 1996; Walsh & Harris 1996), occupied the area (between AD 79 and AD 410). Most perhaps because of the risk of predation (Limpens et al. woodland during the study was found along rivers, and 1989; Ekman & de Jong 1996). They may therefore avoid the western boundary of the study area was marked by a roosts that are isolated (that is, those that do not have JENKINS ET AL.: ROOST SELECTION IN P. PIPISTRELLUS 911

Table 1. Characteristics of 28 roosts and 17 randomly selected control buildings in the Mearns region in northeast Scotland, which were recorded between 12 June and 20 July 1997 Variable Description

Habitation Inhabited by people or not Heating Central heating or fires used regularly during the summer or not. Buildings that were uninhabited were classed as heating not used Roof material Slate, tiles, or corrugated iron Lining material Presence or absence of full lining of rough wooden planking (sarking), on the ceiling of the attic underneath the slates was recorded Insulation within attic Presence or absence of added insulation within the attic: coded as present if the attic floor was lined with either glass fibre or vermiculite grains, otherwise coded as absent Wall material Stone, concrete or bricks Cavity walls Presence or absence of cavity walls Age of house Categories: 1=under 50 years; 2=50–100 years; 3=100–200 years; 4=over 200 years Disturbance As a measure of possible disturbance the presence or absence of cats in the loft was recorded

Table 2. Predation/isolation variables recorded for all roosts and random buildings Variable Description

Isolation The presence or absence of linear vegetation elements (e.g. hedges and tree lines) leading towards woodland was noted. If linear elements were not present the building was classed as isolated Distance to cover The distance (m) to the nearest tree over 10 m tall was measured with a 30-m tape measure in the field. Where the distance was too great to measure in the field it was measured directly with a ruler from a 1:25000 OS map Percentage cover The percentage canopy closure was estimated within a radius of 50 m of the building. Only trees over 10 m tall were included Predators Presence or absence of kestrels, Falco tinnunculus, and owls (Tyto alba, Strix aluco and Asio otus; the most common predators of bats, Speakman 1991): householders were asked whether these were regularly seen or heard

sufficient vegetation surrounding them). To test this, we Where areas of woodland were represented as lines on the made several measurements of the habitat near each roost land classification maps, they were allocated a standard and random building (see Table 2). The presence or width of 25 m, and where areas of woodland were repre- absence of predator species was also recorded (Table 2). sented as points on land classification maps, they were allocated a standard area of 0.0625 ha. Where the wood- land was also shown on the standard OS maps, we Habitat Variables calculated the area using these maps. We counted the The area around all buildings was divided into three number of land classes within each band (a measure of concentric distance bands (up to 0.5 km, between 0.5 and habitat complexity), and measured the distance, on the 1.5 km, and between 1.5 and 2.5 km). We considered it maps, from the building to the nearest area of woodland unnecessary to characterize the area beyond 2.5 km as greater than 1 ha, and to the nearest water (including P. pipistrellus in northeast Scotland travel at most 2.5 km streams, rivers and ponds). We also recorded whether the from the roost and an average of only 1 km (Racey & site was situated within 0.5 km of one of the three main Swift 1985). To quantify the areas (ha) of each land class rivers crossing the area. We investigated the habitat on (deciduous woodland, coniferous woodland, arable land, foot in five areas to ensure that the land classification improved grassland, scattered trees, low scrub, heathland, maps (which are based on aerial photos taken in 1986) smooth grassland, built-up land, ‘other’) in each dis- were up to date, and to clarify areas that were not clearly tance band, we used 1:25 000 maps (Ordnance Survey, shown on the maps. and Scottish land classification maps; Macaulay Land Research Institute 1993). Any woodland classed as mixed Data Analysis woodland on the Scottish land classification maps was considered half deciduous and half coniferous and we We compared roosts and nonroosts by using uni- divided the area between these two woodland types. variate statistics. Continuous variables were normalized 912 ANIMAL BEHAVIOUR, 56, 4

logarithmically or by arcsine transformation where poss- Table 3. Comparisons of features of roost buildings (N=21) with ible, so that parametric tests could be used. Owing to the those of randomly selected control buildings (N=17), using G tests large number of tests carried out, a sequential Bonferroni (William’s correction, Fowler & Cohen 1990)

correction (Rice 1989) was used for each set of tests. Variable df Gadj Significance Roost buildings were occasionally situated within 200 m of one another (that is, they were found at the same location). These roosts were possibly used by bats Habitation 1 1.392 NS Heating 1 0.003 NS from the same colony, and thus may not be independent Roof material 2 2.482 NS of each other (see Thompson 1992), and some of the Lining material 1 0.150 NS predation/isolation and habitat variables (the presence/ Insulation within attic 1 2.392 NS absence of owls and kestrels, isolation and all land class Wall material 2 1.265 NS areas) were identical. Therefore, we made each of the Cavity walls 1 1.323 NS Age of house 6 9.717 NS comparisons twice, once treating all 21 roosts indepen- Disturbance 1 0.048 NS dently, and a second time to avoid pseudoreplication, treating all roosts found within 200 m of one another as A sequential Bonferroni correction was used to establish the level of single roosts. For the second set of tests the values for the significance (Rice 1989). distance to cover and percentage cover within 50 m of the roost, and the distances to water and woodland were averaged for roosts found in the same location. (in all buildings where they were present) and crevices in In all of the analyses, only habitat variables that stone walls in steadings (farm buildings of stone construc- occurred in more than 50% of roosts or random buildings tion). The majority of buildings with roosts were inhab- were used. Therefore, heathland, scrub, smooth grass- ited by humans (17 of the 21 buildings). One roost was in land, built-up areas and those habitat variables classed as a church and two were in steadings. The majority of ‘other’ were excluded from analyses. buildings with roosts had slate roofs (20) with wooden Roost buildings were given a score of 1 and control lining (19), but only six had cavity walls. Buildings buildings a score of 0. We used forward stepwise logistic ranging from 13 to 750 years old were occupied by bats. regression (using SPSS) to quantify the effect of the All but two roosts had some form of linear vegetation predation/isolation and habitat variables of the binary leading towards woodland areas, or were situated imme- dependent variable, presence or absence of bats. We used diately adjacent to woodland and were therefore classed stepwise logistic regression analysis because variables may as ‘not isolated’. Eight of the 21 roost sites were situated be nonparametrically distributed, and categorical data within 0.5 km of one of the three main rivers in the may be included (Sharma 1996). region. All roosts were relatively close to a permanent We investigated the influence of both distance to cover water body (the maximum distance was 550 m). and percentage cover within 50 m of the roost on the emergence behaviour of bats by correlating each of these variables with the light level when the first bat emerged. Features of the Building Distance to cover and light level at first emergence were No significant difference was found between roosts and normalized by a logarithmic transformation, and the random buildings in any of the structural variables percentage cover within 50 m of the roost was normalized recorded (Table 3). by arcsine transformation. Where roosts were observed on more than one occasion (N=2) an average value for the light level at emergence was taken. Predation/Isolation Characteristics In both sets of tests, we found that roosts were less RESULTS isolated than random buildings, were significantly closer Seven of the 28 roosts used by P. pipistrellus were occupied to cover, and had a greater percentage cover within a by the 45-kHz phonotype and 21 by the 55-kHz phono- radius of 50 m (Table 4). When all 21 roosts were analysed type. Since these two phonotypes are separate species we found that a greater proportion of roosts than random (Barratt et al. 1997) with different habitat preferences buildings had tawny owls, Strix aluco, near the building (Vaughan et al. 1997), we omitted data for roosts occu- (Table 4). This was not the case when roost buildings pied by the 45-kHz phonotype from our analyses. It was found at the same site were treated as single roosts (Table impossible to determine whether the roosts were occu- 4). There was no difference in the distribution of kestrels, pied by maternity colonies or nonreproducing individ- Falco tinnunculus, between roosts and random buildings uals. The only indication as to the status of roosts was the for either set of tests (Table 4). number of bats (that is larger numbers were more likely to be breeding colonies). However, this is only a rough Habitat Characteristics gauge of roost status. All roosts were situated within buildings. The 55-kHz In both sets of analyses roosts were situated signifi- phonotype bats roosted in a variety of locations within cantly closer to an area of woodland greater than 1 ha buildings, including the spaces between slates and than were random buildings (Table 5). The total area of wooden sarking (boards behind the slates), cavity walls decidous woodland and the total area of coniferous JENKINS ET AL.: ROOST SELECTION IN P. PIPISTRELLUS 913

Table 4. Difference between roosts and randomly selected control buildings (N=17) in predation/isolation variables, tested using t tests and G tests (with William’s correction, Fowler & Cohen 1990)

Variable Roost Random house Gadj/t

All roosts (N=21)

Presence/absence of linear vegetation elements 90% with linear vegetation 29% with linear vegetation G1=31.31** leading towards woodland

Presence/absence of kestrels 67% with kestrels 47% with kestrels G1=0.71 Presence/absence of owls 86% with owls 35% with owls G1=10.27* ± ± ± Log10 distance to cover (m; X SE) 0.93 0.06 1.60 0.17 t18=3.74** ± ± ± − Arcsine %cover within 50 m of roost (X SE) 0.56 0.37 0.18 0.04 t35= 7.28**

Neighbouring roosts pooled (N=11)†

Presence/absence of linear vegetation elements 82% with linear vegetation 29% with linear vegetation G1=7.37* leading towards woodland

Presence/absence of kestrels 64% with kestrels 47% with kestrels G1=0.30 Presence/absence of owls 73% with owls 35% with owls G1=3.64 ± ± ± Log10 distance to cover (m; X SE) 1.02 0.06 1.60 0.17 t20=3.22** ± ± ± − Arcsine %cover within 50 m of roost (X SE) 0.59 0.05 0.18 0.04 t19= 6.58**

Sequential Bonferroni corrections were used on each set of tests to establish significance, hence P values are not shown. †Locations with more than one roost treated as a single roost. *P<0.05; **P<0.01.

woodland within the 0.5-km distance band were both DISCUSSION significantly greater for roosts than for random buildings (Table 5). When all roosts were treated independently Features of the Building there was a significantly greater area of deciduous wood- land within the 0.5–1.5 km distance band but this differ- Occupied roosts did not differ from randomly selected ence was not found when roosts at the same location buildings in any of the features of the actual buildings were treated as single roosts (Table 5). Roosts were more that we measured. This suggests that pipistrelles were likely to be situated within 0.5 km of a major river than not selective in their choice of roost with respect to were random buildings, but only when roosts found these specific characteristics of the building. A similar within 200 m of one another were treated as single roosts study in an adjacent area of northeast Scotland found (Table 5). No other differences in habitat variables were that long-eared bats (which roost within the roof space found. of buildings) preferred buildings that were older, had We did the logistic regression only once, treating roosts roof spaces divided into more roof compartments and located within 200 m of one another as single roosts. This that were more likely to be fully lined with rough was to avoid biases in the result due to identical values for wooden planking than randomly selected buildings certain attributes. One of the predation/isolation and (Entwistle et al. 1997). Long-eared bats thus appear to habitat variables (percentage cover within 50 m of the be more specialized in their roosting habits than pipis- building) was significantly related to the presence or trelles. Pipistrelles have previously been described as absence of bats in the stepwise logistic regression model generalists in terms of their roosting habits (Avery (Table 6). Roost sites had a greater percentage cover 1991). This is illustrated by the variety of roost locations within 50 m of the roost than random buildings. used by pipistrelles during the present study which included cracks in walls, beneath slates and within cavity walls. It is likely that common pipistrelles will Emergence Behaviour of Bats generally not be limited by the availability of suitable buildings. Bats emerged between 33 min before sunset and 26 min Bats might have been selecting roosts using criteria we after sunset, and at light levels between 16 and 183 lx. did not measure, such as the temperature and humidity There was a strong positive relationship between percent- within the roost. Ambient conditions play an impor- age cover within 50 m of the roost and light level at first tant role in balancing energy budgets (Kunz 1980) and emergence (log10 (light level)=1.13+0.896 arcsine (% influence metabolic rate (Speakman & Racey 1987; 2 cover), r =0.38, F1,18=12.68, P<0.01; Fig. 1a), and a nega- Rodri´guez-Dura´n 1995) and evaporative water loss tive relationship between percentage cover and time after (Webb et al. 1995). Selection of roosts with a specific sunset of first emergence (time=19.0"27.9 arcsine (% temperature and humidity has been demonstrated in 2 cover), r =0.15, F1,18=4.41, P<0.05; Fig. 1b). Distance to other bat species (e.g. Plecotus auritus: Entwistle et al. cover was not related to light level at first emergence 1997; Rhinonycteris aurantius: Churchill 1991), and (r= "0.05, N=20, NS), or time after sunset or first these attributes may also influence roost selection in emergence (r= "0.18, N=20, NS). pipistrelles. 914 ANIMAL BEHAVIOUR, 56, 4

Table 5. Comparisons of habitat characteristics between roosts and randomly selected control buildings (N=17)

Variable Roost Random house Gadj/U/t

All roosts (N=21) ± ± Log10 distance to deciduous woodland (m) 1.75 0.13 2.96 0.06 t28=8.57** ± ± Log10 distance to water (m) 2.01 0.08 2.16 0.11 t31=1.14 Within 0.5 km of main river or not 38% within 0.5 km 0% within 0.5 km G1=10.50 0.5-km band around roost Area of deciduous woodland (ha) 13.41 (6.05, 14.92) 0.13 (0.06, 0.69) U=10** Area of arable land (ha) 63.56 (33.00, 63.56) 72.88 (60.88, 77.50) U=92 Area of improved grassland (ha) 0.00 (0.00, 13.94) 0.00 (0.00, 9.63) U=195 Area of coniferous woodland (ha) 2.13 (1.59, 14.02) 0.31 (0.00, 1.22) U=50** ± ± − Complexity‡ 4.57 0.34 3.71 0.27 t35= 1.99 0.5–1.5-km ring ± ± − Area of deciduous woodland (ha) 24.76 3.3 6.67 1.01 t23= 5.24** Area of arable land (ha) 486.3 (441.7, 579.1) 531.7 (441.2, 579.1) U=155 Area of improved grassland (ha) 31.15 (0.00, 58.28) 21.90 (0.00, 82.23) U=166 Area of coniferous woodland (ha) 17.88 (8.09, 42.54) 4.40 (1.55, 15.88) U=116 Area of scattered trees (ha) 8.09 (1.35, 39.56) 4.24 (0.00, 11.24) U=121 ± ± − Complexity‡ 7.67 0.31 7.06 0.34 t34= 1.32 1.5–2.5-km ring ± ± Log10 area of deciduous woodland (ha) 1.02 0.09 1.06 0.10 t35=0.29 Area of arable land (ha) 682.5 (16.2, 924.6) 980.8 (666.0, 1122.1) U=98 Area of improved grassland (ha) 193.5 (108.3, 1127.9) 168.4 (17.3, 206.5) U=106 Area of coniferous woodland (ha) 9.06 (7.71, 49.65) 24.25 (8.05, 111.35) U=165 Area of scattered trees (ha) 13.23 (7.32, 38.89) 16.70 (0.93, 22.32) U=146.5 ± ± − Complexity‡ 9.24 0.14 8.59 0.31 t22= 1.92

Neighbouring roosts pooled (N=11)† ± ± Log10 distance to broadleaved woodland (m) 1.67 0.23 2.96 0.06 t11=5.50** ± ± Log10 distance to water (m) 1.98 0.13 2.16 0.11 t21=1.03 Within 0.5 km of main river or not 45% within 0.5 km 0% within 0.5 km G1=78.58** 0.5-km band around roost ± − ± − Log10 area of deciduous woodland (ha) 0.69 0.16 0.65 0.15 t23= 6.04** Area of arable land (ha) 42.88 (29.81, 68.50) 72.88 (60.88, 77.50) U=44 Area of improved grassland (ha) 5.94 (0.00, 13.94) 0.00 (0.00, 9.63) U=69 Area of coniferous woodland (ha) 4.59 (2.03, 14.88) 0.31 (0.00, 1.22) U=19** ± ± − Complexity‡ 5.09 0.41 3.71 0.27 t18= 2.81 0.5–1.5-km ring ± ± − Area of deciduous woodland (ha) 25.09 5.69 6.67 1.01 t10= 3.19 Area of arable land (ha) 442.53 (404.9, 501.5) 531.7 (441.2, 579.1) U=62 Area of improved grassland (ha) 46.10 (31.15, 72.05) 21.90 (0.00, 82.23) U=119 Area of coniferous woodland (ha) 30.79 (9.76, 58.44) 4.40 (1.55, 15.88) U=50.7 Area of scattered trees (ha) 15.73 (1.35, 49.64) 4.24 (0.00, 11.24) U=50 ± ± − Complexity‡ 7.82 0.54 7.06 0.34 t17= 1.20 1.5–2.5-km ring ± ± − Log10 area of deciduous woodland (ha) 1.08 0.15 1.06 0.10 t18= 0.08 Area of arable land (ha) 843.9 (562.2, 927.4) 980.8 (666.0, 1122.1) U=59 Area of improved grassland (ha) 182.7 (91.4, 237.3) 168.4 (17.3, 206.5) U=66 Area of coniferous plantation (ha) 49.43 (9.06, 259.57) 24.25 (8.05, 111.35) U=74 Area of scattered trees (ha) 17.21 (7.32, 46.57) 16.70 (0.93, 22.32) U=71.5 ± ± − Complexity‡ 9.18 0.23 8.59 0.31 t25= 1.55

Differences were tested with G tests (with William’s correction, Fowler & Cohen 1990), t tests, and Mann–Whitney U tests. Means±SE are shown where t tests were used and medians (interquartile range) where Mann–Whitney U tests were used. A sequential Bonferroni correction (Rice 1989) was used for each set of tests to establish the level of significance. †Locations with more than one roost treated as a single roost. ‡Number of land classes present. *P<0.05; **P<0.01.

Table 6. Logistic regression model of presence/absence of bats at roosts (N=11 buildings (sites) with bats, and 17 without bats) Predation/Isolation Characteristics χ2 Variable Coefficient 1 P Roosts had more linear vegetation elements leading away from them towards woodland areas than random Constant −7.0118±3.3263 4.4435 <0.05 buildings had. It has been suggested that bats tend to Arcsine %cover 0.5259±0.2702 3.7867 <0.05 avoid open areas and prefer to travel along linear veg- etation elements (Limpens et al. 1989). Bats may more JENKINS ET AL.: ROOST SELECTION IN P. PIPISTRELLUS 915

2.5 minimizing the amount of time for which they are (a) exposed to predators. The presence of trees close to the roost may also provide cover from predators, allowing bats to emerge 2 from roosts with a lowered probability of detection. As bats are thought to be under the greatest risk of predation from diurnal avian predators, such as falcons, as they 1.5 emerge from roosts (Gillette & Kimbrough 1970; Fenton et al. 1994), cover near the roost may increase survival, light level (lx) particularly for young bats which are unable to fly fast 10 (Racey & Swift 1985). Furthermore, cover close to the Log 1 roost may allow bats to emerge earlier, when it is lighter, and when the risk of detection by diurnal avian predators would otherwise be high. By emerging earlier, pipistrelles can forage for longer and exploit the many insects that 0.5 0 0.2 0.4 0.6 0.8 1 are at their peak abundances before dusk (Speakman 1995; Rydell et al. 1996). The average energy expendi- 30 ture of pipistrelles is approximately 20 kJ/day (Racey (b) & Speakman 1987). With an absorption efficiency of 20 approximately 76.5% (Barclay et al. 1991), the equivalent food intake each evening would be 26.1 kJ/night. Pipis- 10 trelles spend about 225 min foraging each evening (Swift 1980). Therefore, the average feeding rate is 0.117 kJ/min. 0 At dusk pipistrelles have been observed feeding at a rate of 10 insects per min (Racey & Swift 1985). If each of –10 these insects has a dry weight of 1 mg and an energy content of 23.2 kJ/g dry weight, the energy intake rate at –20 dusk is 0.232 kJ/min. Thus, the feeding rate at dusk is Time after sunset (min) Time approximately twice that of the overall nightly feeding –30 rate. Therefore, extra foraging time during this part of the evening may be important, particularly for reproducing –40 females. 0 0.2 0.4 0.6 0.8 1 In the present study bats emerged earlier when there Arcsine %cover was a greater percentage cover within 50 m of the roost. Figure 1. The relationship between (a) light level at first emergence Using the regression equation to predict the emergence and (b) time after sunset of first emergence and percentage cover time, we calculated the emergence time of bats at roosts within 50 m of the roost. In (a), m denotes two points in the same with 29% cover (the mean percentage cover at roosts) to place. be 3 min after sunset and at roosts with 3% cover (the mean at randomly selected buildings) to be 14 min after easily find roosts that are located along linear elements sunset. Thus, on average an extra 11 min of foraging time (Limpens & Kapteyn 1991). was gained at roosts with 29% cover. At the energy intake Compared with random buildings, roosts were closer to rate for dusk foraging (0.232 kJ/min calculated above) an trees over 10 m tall and they had a greater percentage additional 2.55 kJ is consumed during the extra 11 min of cover within 50 m of the building. The presence of trees foraging time at roosts with 29% cover. This is about 10% in the vicinity of roosts may offer foraging opportunities of the average energy intake each night (26.1 kJ/night). as insects will probably be more abundant than in open The observation that owls were more likely to be found areas (Racey & Swift 1985). Foraging behaviour and the near roosts than near random buildings is to be expected distribution and abundance of pipistrelles are both influ- as roosts had more woodland near them, where tawny enced by insect distribution and abundance (Racey & owls are likely to be found. Swift 1985; de Jong & Ahle´n 1991), with bats found foraging in greater numbers around trees and over water Habitat Characteristics where prey are abundant. However, as bats rarely forage close to their roosts, it is unlikely that trees situated The shorter distance to, and greater area of deciduous within 50 m of the roost will represent an important woodland found within, a radius of 1.5 km from the foraging site (Racey & Swift 1985). roosts suggest that the proximity to and amount of Alternatively, the presence of trees around the roost woodland influence roost selection. Pipistrelles prefer to may provide shelter from environmental extremes, forage in deciduous woodland (Racey & Swift 1985; de which are known to influence bat activity (Fenton et al. Jong & Ahle´n 1991) where insects are more abundant 1977), and may result in a more stable environment than in coniferous woodland (Walsh & Harris 1996). The within the roost itself. Shelter near the roost may greater area of coniferous woodland found within 0.5 km thus allow bats to emerge and commute faster, thus of roosts than that found within 0.5 km around random 916 ANIMAL BEHAVIOUR, 56, 4

buildings indicates that coniferous woodland is also and to Dr I. Mitchell for valuable discussion. Thanks also important, even though bats avoid it (Swift 1980). This to Peter Lutz and two anonymous referees for their result may be an artefact of the technique we used for comments. quantifying woodland, in which we divided mixed wood- land between deciduous and coniferous woodland. Much of the woodland surrounding roosts was mixed woodland References rather than separate stands of deciduous and coniferous Avery, M. I. 1991. Pipistrelle Pipistrellus pipistrellus.In:The Handbook woodland. Mixed woodland is more diverse than conifer of British Mammals (Ed. by G. B. Corbet & S. Harris), pp. 124–128. plantations (see Butterfield & Malvido 1992), and thus Oxford: Oxford University Press. may represent suitable foraging sites. Barclay, R. M. R., Dolon, M. & Dyck, A. 1991. The digestive The fact that there were no differences in any of the efficiency of insectivorous bats. Canadian Journal of Zoology, 69, habit variables beyond 1.5 km corresponds closely with 1853–1856. observations of the distances travelled by foraging bats Barratt, E. M., Deaville, R., Burland, T. M., Bruford, M. W., Jones, (Racey & Swift 1985). Therefore the number of suitable G., Racey, P. A. & Wayne, R. K. 1997. DNA answers the call of pipistrelle bat species. Nature, 387, 138–139. foraging areas within this distance is likely to influence Butterfield, J. & Malvido, J. B. 1992. Effect of mixed-species tree roost selection. A study of P. auritus in northeast Scotland planting on the distribution of soil invertebrates. In: British Eco- showed that this species selects roosts that are surrounded logical Society Special Publications No. 11: The Ecology of Mixed by a greater area of deciduous woodland than randomly Species Stands of Trees (Ed. by M. G. R. Cannell, D. C. Malcolm selected available buildings, within a radius of 0.5 km. & P. A. Robertson), pp. 255–265. Oxford: Blackwell Scientific This species has also been observed to spend most of its Publications. foraging time within this distance. Cerveny´,J.&Bu¨rger, P. 1989. Density and structure of the bat Roosts were not situated closer to water than were community occupying an old park in Zihobce (Czechoslovakia). nonroosts. This probably reflects the fact that all the In: European Bat Research 1987 (Ed. by V. Hana´k,I.Hora´cek & buildings in the area were located close to water rather J. Gaisler), pp. 475–488. Praha: Charles University Press. Churchill, S. K. 1991. Distribution, abundance and roost selection of than indicating that water is not required by pipistrelles the orange horseshoe bat, Rhinonycteris aurantius, a tropical cave (the maximum distance to water for nonroosts was 475 m dweller. Wildlife Research, 18, 343–354. and for roosts 550 m). In the study area, the availability of Davis, R. B., Herreid, C. F. & Short, H. L. 1962. Mexican free-tailed roost sites close to water was probably not a limiting bats in Texas. Ecological Monographs, 32, 311–342. factor. Roosts were more likely to be situated within Ekman, M. & de Jong, J. 1996. Local patterns of distribution and 0.5 km of one of the three main rivers than were non- resource utilization of four species (Myotis brandti, Eptesicus roosts. This suggests that the distribution of roosts may nilssoni, Plecotus auritus and Pipistrellus pipistrellus) in patchy and by determined to a certain extent by different types of continuous environments. Journal of Zoology, 238, 571–580. water. Entwistle, A. C., Racey, P. A. & Speakman, J. R. 1996. Habitat Understanding the roosting requirements and es- exploitation by a gleaning bat, Plecotus auritus. Philosophical Transctions of the Royal Society of London, Series B, 351, 921–931. pecially maternity roost requirements for any bat species Entwistle, A. C., Racey, P. A. & Speakman, J. R. 1997. Roost is essential to their conservation in today’s changing selection by the brown long-eared bat Plecotus auritus. Journal of agricultural landscape. In this study, the factors that Applied Ecology, 34, 754–763. determined roost selection, in general, related to tree Fenton, M. B. 1983. Roosts used by the African bat, Scotophilus cover and deciduous woodland around the roost build- leucogaster (Chiroptera: Vespertilionidae). Biotropica, 15, 129– ing. Deciduous woodland has also been shown to be 132. important for other species (e.g. P. auritus: Entwistle et al. Fenton, M. B. & Rautenbach, I. L. 1986. A comparison of the 1997). Therefore, attempts should be made to conserve roosting and foraging behaviour of three species of African insec- woodland if the status of pipistrelles and long-eared bats tivorous bats (Rhinolophidae, Vespertilionidae, and Molossidae). Canadian Journal of Zoology, 64, 2860–2867. is to improve. Fenton, M. B., Boyle, N. G. H., Harrison, T. M. & Oxley, D. J. 1977. Activity patterns, habitat use, and prey selection by some African Acknowledgments insectivorous bats. Biotropica, 9, 73–85. Fenton, M. B., Rautenbach, I. L., Smith, S. E., Swanepoel, C. M., We thank all householders who allowed us access to their Grosell, J. & van Jaarsveld, J. 1994. Raptors and bats: threats and homes, and who showed us a great deal of hospitality opportunities. Animal Behaviour, 48, 9–18. while we conducted our work. Thanks also to Jason Fowler, J. & Cohen, L. 1990. Practical Statistics for Field Biology. Reynolds, Donna McRae, Stan Hughes, Claire Murphy, Chichester, U.K.: J. Wiley. Marie Kelly, Sally Oldfield, Bromwyn Kineavy, Craig Fox, R., Lehmkuhle, S. W. & Westendorf, D. H. 1976. Falcon visual Turner, Steve Hutton, Emma Pettitt, Beckie Leaper, Alison acuity. Science, 192, 263–265. Holt, Diane Jackson, Paula Redman, Ian Mitchell, Sarah Gillette, D. D. & Kimbrough, J. D. 1970. Chiropteran mortality. In: About Bats (Ed. by B. H. Slaughter & D. W. Walton), pp. 262–283. Bowe, Ela Krol, Guy Robertson, Georgette Shearer, Nick Dallas: Southern Methodist University Press. Cox and Stuart Blackman for their help with counts. This van Heerdt, P. F. & Sluiter, J. W. 1965. Notes on the distribution work was supported by a grant from the Mammal Con- and behaviour of the noctule bat (Nyctalus noctula)inthe servation Trust Bursary. We are grateful to Professor I. Netherlands. Mammalia, 29, 463–477. Jolliffe (Department of Statistics, University of Aberdeen) Jones, G. & van Parijs, S. M. 1993. Bimodal echolocation in for statistical advice, to Dr Gall (Macaulay Land Use pipistrelle bats: are cryptic species present? Proceedings of the Research Institute) for providing land classification maps, Royal Society of London, Series B, 251, 119–125. JENKINS ET AL.: ROOST SELECTION IN P. PIPISTRELLUS 917

Jones, G., Duverge´, P. L. & Ransome, R. D. 1995. Conservation Speakman, J. R. & Racey, P. A. 1987. The energetics of pregnancy biology of an endangered species: field studies of greater horse- and lactation in brown long-eared bats (Plecotus auritus). In: shoe bats. In: Symposium of the Zoological Society of London: Recent Advances in the Study of Bats (Ed.byM.B.Fenton,P.A. Ecology, Evolution and Behaviour of Bats (Ed.byP.A.Racey&S.M. Racey & J. M. V. Rayner), pp. 367–393. Cambridge: Cambridge Swift), pp. 309–324. London: The Zoological Society of London. University Press. de Jong, J. & Ahle´n, I. 1991. Factors affecting the distribution Speakman, J. R., Racey, P. A., Catto, C. M. C., Webb, P. I., Swift, pattern of bats in Uppland, central Sweden. Holarctic Ecology, 14, S. M. & Burnett, A. M. 1991. Minimum summer populations and 92–96. densities of bats in N. E. Scotland, near the northern borders of Kunz, T. H. 1980. Daily energy budgets of free living bats. In: their distributions. Journal of Zoology, 225, 327–345. Proceedings of the 5th International Bat Research Conference (Ed. by Swift, S. M. 1980. Activity patterns of pipistrelle bats (Pipistrellus D. E. Wilson & A. L. Gardner), pp. 369–392. Lubbock: Texas pipistrellus) in north-east Scotland. Journal of Zoology, 190, 285– Technical Press. 295. Limpens, H. J. G. A. & Kapteyn, K. 1991. Bats, their behaviour and Thompson, M. J. A. 1992. Roost philopatry in female pipistrelle bats linear landscape elements. Myotis, 29, 39–48. Pipistrellus pipistrellus. Journal of Zoology, 228, 673–679. Limpens, H. J. G. A., Helmer, W., Van Winden, A. & Mostert, K. 1989. Bats (Chiroptera) and linear landscape elements. Lutra, 32, Tuttle, M. D. 1976. Population ecology of the gray bat (Myotis 1–20. grisescens): factors influencing growth and survival of newly volent Macaulay Land Use Research Institute 1993. The Land Cover of young. Ecology, 57, 587–595. Scotland 1988. Final Report. Aberdeen. Vaughan, N., Jones, G. & Harris, S. 1997. Habitat use by bats Morrison, D. W. 1980. Foraging and day-roosting dynamics of (Chiroptera) assessed by a broad-band acoustic method. Journal of canopy fruit bats in Panama. Journal of Mammalogy, 61, 20–29. Applied Ecology, 34, 716–730. Racey, P. A. & Speakman, J. R. 1987. The energy costs of pregnancy Vaughan, T. A. 1970. Flight patterns and aerodynamics. In: Biology and lactation in heterothermic bats. In: Symposium of the Zoologi- of Bats, Vol. 1 (Ed. by W. A. Wimsatt), pp. 195–216. New York: cal Society of London: Reproductive Energetics in Mammals (Ed. by Academic Press. A. S. I. Loudon & P. A. Racey), pp. 107–125. London: The Vaughan, T. A. 1987. Behavioural thermoregulation in the African Zoological Society of London. yellow-winged bat. Journal of Mammalogy, 68, 376–378. Racey, P. A. & Swift, S. M. 1985. Feeding ecology of Pipistrellus pipistrellus (Chiroptera: Vespertilionidae) during pregnacy and Vaughan, T. A. & O’Shea, T. J. 1976. Roosting ecology of the pallid lactation. I. Foraging behaviour. Journal of Animal Ecology, 54, bat, Antrozous pallidus. Journal of Mammalogy, 57, 19–42. 205–215. Vonhof, M. J. & Barclay, R. M. R. 1996. Roost-site selection and Rice,W.R.1989. Analyzing tables of statistical tests. Evolution, 43, roosting ecology of forest-dwelling bats in southern British 223–225. Columbia. Canadian Journal of Zoology, 74, 1797–1805. Rodri´guez-Dura´n, A. 1995. Metabolic rates and thermal conduct- Walsh, A. L. & Harris, S. 1996. Foraging habitat preferences of ance in four species of neotropical bats roosting in hot caves. vespertilionid bats in Britain. Journal of Applied Ecology, 33, 508– Comparative Biochemistry and Physiology, 110A, 347–355. 518. Rydell, J., Entwistle, A. & Racey, P. 1996. Timing of foraging flights Waters, D. A. & Walsh, A. L. 1994. The influence of bat detector of three species of bats in relation to insect activity and predation brand on the quantitative estimation of bat activity. Bioacoustics, risk. Oikos, 76, 243–252. 5, 205–221. Sharma, S. 1996. Applied Multivariate Techniques. New York: J. Wiley. Webb, P. I., Speakman, J. R. & Racey, P. A. 1995. Evaporative water Speakman, J. R. 1991. The impact of predation by birds on bat loss in two sympatric species of vespertilionid bat, Plecotus populations in the British Isles. Mammal Review, 21, 123–142. auritus and Myotis daubentonii: relation to foraging mode and Speakman, J. R. 1995. Chiropteran nocturnality. In: Symposium of implications for roost site selection. Journal of Zoology, 235, the Zoological Society of London: Ecology, Evolution and Behaviour of 269–278. Bats (Ed. by P. A. Racey & S. M. Swift), pp. 187–201. London: The Wunder, L. & Carey, A. B. 1996. Use of the forest canopy by bats. Zoological Society of London. Northwest Science, 70, 79–85.