The overlap between the roosting areas of each cluster- deﬁned group of bats was depicted graphically using ArcView GIS 3.2 (Environmental Systems Research Institute, Inc., CA, U.S.A.) and spatial analysis performed with the Animal Movement extension (Hooge & Eichen- laub 1997). Coordinates of all roost trees were plotted in a digital orthophoto of the Mar´ıa Luisa Park with 0.5-m resolution (Ortofotograf´ıa digital de Andaluc´ıa, Junta de Andaluc´ıa 2004). We superimposed on to the photographs detailed digital maps of the park (Servicio de Parques y Jar- dines, Ayuntamiento de Sevilla 2002) to precisely locate trees. As not all trees were used by all bats, and not all trees were used with the same intensity, we identiﬁed ‘core roosting areas’ of each cluster-deﬁned group of bats by drawing 50% contour lines using the ﬁxed kernel estima- tion method (Worton 1989). We counted the number of days that each bat roosted in each tree, which is equiva- lent to weighting by intensity of use. The 95% kernel use distribution was used to describe the overall roosting range. The smoothing parameter (h) was ﬁxed to a value of 15 m.
Loyalty to Roosting Areas
Using recapture and transponder data, we assessed whether adult females were loyal to their roosting areas between years and if young females returned to their natal roosting areas in subsequent years.
Data collected in a preliminary study in 1992 were used to assess the long-term stability of roosting patterns. In July 1992, two adult female bats captured emerging from a tree roost in the Mar´ıa Luisa Park were ﬁtted with transmitters and followed to their roosts for 13 and 16 days, respectively. Several other trees were checked in
1992 and in subsequent years for the presence of bats.
Ninety per cent of adult females breed in the study area but reproductive status was not apparent at the end of April when bats were captured and tagged (see Results).
Based on the timing of parturition in the study area (see Results), we calculated roost-switching frequency for the prelactation period (from when bats were tagged until 31
May), and for the lactation period (from 1 June until the signal was lost). We used a paired t test to assess differences in roost-switching frequency between reproductive pe- riods, for individuals that were continuously tracked (i.e. every day) during both periods in 2004. The power of the test was calculated using Power and Precision Ver- sion 2.0 (Borenstein et al. 2000). Nonsigniﬁcant deviation from normality and homogeneity of variances were evalu- ated using KolmogoroveSmirnov and Levene tests. To as- sess if some individuals switched roosts at a different rate than others, we used a Friedman test (for repeated measures) with individuals as a factor and the number of consecutive days they remained in each successive tree before each roost-switch as the repeat level factor. For this calculation, only data from the prelactation period were used, and to obtain a balanced design, we only tested a subsample of 10 in- dividuals with equal numbers of repeat levels (N ¼ 12).
Three-hundred and twelve individuals were captured and marked between 1999 and 2006, of which only 16 were adult males. Parturition took place between late May and beginning of June, and 90% of adult females captured during June and July (N ¼ 244) were lactating. Tagged bats switched roosts frequently and used 73 different trees (Fig. 1, Table 1). They nearly always returned to the park for day roosting (only in 5 of 885 attempts did we fail to ﬁnd the signal for a bat). Night roosting inside the park was also observed, but we did not attempt to locate night roosts precisely. The number of bats emerging from counts at ﬁve roosts varied from 14 to 60 (mean ¼ 27, N ¼ 13) in the breeding season.
Social Groups and their Roosting Areas
All clustering methods separated the 25 bats into three main groups based on roost use (Fig. 2). Hereafter, we des- ignated these as social groups I, II and III (Fig. 2, Table 1). The four methods grouped bats in the same way (cluster- ing following Ward’s method is shown; Fig. 2): all bats that were captured at the same or at very nearby trees were assigned to the same cluster, except for bat 17 which was captured together with members of social group I, but cluster analysis placed it in social group II.
Based on pairwise associations, all hierarchical trees also revealed three main assemblages (clustering following Ward’s method is shown; Fig. 3). All 15 bats were distrib- uted in the same clusters as the previous approach, again with the exception of bat 17, which was placed into all three social groups depending on the clustering method. All roost trees were identiﬁed and mapped allowing us
to assign each of the 312 bats captured in 1999e2006 to one of the three social groups, depending on the tree where it was ﬁrst captured. Of 60 recaptured individuals, only one (a juvenile) was recaptured in a tree belonging to a different group from where it was ﬁrst caught, and it is
(a) (b) (c)
0 100 200 m
Figure 1. Distribution of roosts in the Mar´ıa Luisa Park (Seville, Spain). (a) Trees used for roosting by three social groups; black dots: social group I; black squares: social group II; grey circles: social group III. Arrows indicate the position of the two automatic reading devices. (b) Roosting areas of the three groups. Dark grey: 95% kernel of group I; middle grey: 95% kernel of group II; light grey: 95% kernel of group III; black: 50% roosting areas of each group. (c) Trees used for roosting by two bats radiotracked in 1992. Black dots: trees still used in 2003e2004; crosses: trees that have been felled. Roost trees characterized in 2003e2004 are shown as grey dots for comparison. T1 and T2 represent trees later used by social group III (see text). Roads are marked as lines for aiding comparison between subﬁgures.
unclear whether this individual actually emerged from the roost or merely got entangled on the net while passing by (unlike most bats, this individual was not seen emerging from the cavity). The automatic readers detected 116 of the 256 bats marked with transponders. Based on our cluster analysis, readers were located in trees used by social groups I and II, respectively, and we designated them as readers 1 and 2 (Fig. 1a). Reader 1 detected 61 bats belong- ing to social group I (75% of all marked bats in this group) and only one outsider, an adult female assigned by capture site to group II (Table 2). Reader 2 detected mostly bats marked at roosts of social group II (48 bats, i.e. 79% of bats marked from this group; Table 2), and only six
‘foreign’ bats, ﬁve adult females and one adult male,
assigned by capture site to social group III (5% of all marked bats from social group III). This pattern is signiﬁ- cantly different from random (Chi-squared test: c2 ¼ 249, P < 0.001), that is bats from all three social groups did not have the same probability of being de- tected at readers 1 and 2: bats from social group I were de- tected at reader 1, bats from social group II were detected at reader 2, and bats from social group III, with no reader located at their roosts, were negligibly detected.
To compare activity patterns of ‘native’ versus ‘foreign’ bats visiting the reader-equipped trees, we counted the number of days each bat was detected (Table 2). For reader
1, ‘native’ bats (those assigned to social group I) visited the roost an average of 47 (±45 SD) days per bat
20 19 18 12 11 17 10 8 16 9 7 13 15 6 14 5 23 24 22 21 25 4 3 2 1
Figure 2. Results of cluster analysis (Ward’s method) for the 25 bats radiotracked (2003e2004) revealing three main subunits (social groups I
and III). Bats are assembled together depending on the degree of similarity in their roost use.
25 24 23 22 21 17 14
15 16 13 20 18 19 12 11
Figure 3. Results of cluster analysis (Ward’s method) for the 15 bats radiotracked in 2004. Aggregations are based on pair-wise associations between all bats.
(noncontinuous days, as the bats left the roost and re- turned to it several times), resulting in 2904 bat visits* days. Although individual variation was large, only ﬁve
bats (w8%) were detected on fewer than 3 days. For reader
2, ‘native’ bats used the tree on average 40 noncontinuous days (±48 SD) for a total of 1919 bat visits*days. Only ﬁve bats (w10%) were detected on less than 3 different days.
Bats emerged soon after sunset but activity by bats enter- ing and leaving usually continued for the whole night. Some individuals made more than one emergence-return event. In contrast, ‘foreign’ bats had only one record each, mostly between 0500 and 0600 hours, except one bat that was detected on each of two consecutive nights. Except for this case, in which the bat presumably day- roosted in the tree, there was no evidence that the ‘for- eign’ bats roosted in the tree. They could merely have sat in the entrance to explore the cavity, as a bat roosting in a tree should have at least two records (entrance and emergence).
Roost trees were assigned to the social group where the bats that used them belonged (Fig. 1a). Bats from social group I roosted in 39 trees; social group II used 27 trees; and social group III used 34 trees. Each social group gener- ally roosted in a spatially distinct area of the park, al- though 14 trees (19% of all 73 used trees) were used by bats from several social groups (two trees shared by social groups I and II, 7 by social groups I and III, two by groups II and III and three shared by all social groups; Fig. 1a). These shared trees were either located in the core roosting area of one social group and only occasionally some ‘for- eign’ bat roosted in them, or they were located at the boundaries between roosting areas and used occasionally by one or a few bats of each social group (Fig. 1b). Core roosting areas of each group, designated by 50% kernel use distributions, did not overlap (Fig. 1b). These areas contained few roost trees, but these trees were the most important (i.e. most used) by the different social groups. When spatial overlap between social groups took place, one or all of them used the area only marginally (95%
kernel; Fig. 1b). Percentage surface area of roosting ranges of each social group (designated by 95% kernels) shared with other social groups was 15% for social group I, 17% for social group II and 9% for social group III (Fig. 1b).
Loyalty to Roosting Areas and Trees
Sixty bats were recaptured, all of them in their distinct roosting areas, some 1 year (N ¼ 15), 2 years (N ¼ 15), 3 years (N ¼ 13) or 4 years (N ¼ 1) after their ﬁrst capture. Bats were also found repeatedly in their roosting areas across different seasons. At least 32 out of 49 adult females (65%) carrying transponders from social groups I and II (with one reader each) before 2006, returned in subse- quent years to their roosting areas. At least 17 females re- turned to their natal roosting areas based on detections at the readers, out of 27 juvenile females marked from social groups I and II before 2006 (55%).
The two individuals radiotracked in 1992 were followed to their roosts for 13 and 16 days, respectively, and used nine different trees. Some trees including the one where they were captured, no longer exist but were situated inside the main roosting area of social group I. Three other trees used in 1992 were still used in 2003 and 2004 by bats from social group I. One tree where reader 1 was located has been checked every year since 1992 and has been occupied by giant noctules every year. Another tree occupied in 1992 (T1; Fig. 1c) was reused in 2003 and
2004 by bats from social group III, and roost T2 was shared in 2003 and 2004 by both social groups I and III (Fig. 1a). These two last roosts, which were not checked every year, are near the boundary between the current core areas of groups I and III. Although the park was not intensively surveyed for giant noctules in 1992, several other trees with bats were detected. Speciﬁcally, trees T3, T4 and T5 (Fig. 1c), which in 2003e2004 belonged to the core area of social group II (Fig. 1b), were occupied in 1992 but not used by the two radiotagged bats.
Bats spent an average of 2.68 (± 0.82 SD; range: 1e31; mode ¼ 1; N ¼ 13 bats) days in a given tree before moving to another roost (Table 1). Most roost trees were reused during the study period (less than 30% were used only once). Only seven bats provided data during the lactation period because transmitters failed or fell off before June (Table 1). Average roost-switching frequency was not signiﬁcantly different before (2.52 days/roost ± 0.74 SD) versus during lactation (4.88 ± 1.91 SD; paired t test, t6 ¼ —1.97, P ¼ 0.096). Power of the test was 0.38. Roost- switching rate (counted as the number of consecutive days that each bat roosted in each successive tree) did not differ signiﬁcantly between individuals (Friedman
test: c2 ¼ 7.06, N ¼ 12, P ¼ 0.63).
Although group composition in a particular tree varied from day to day, roost mates rarely switched to another tree simultaneously. Thus, bats did not seem to move as a group. On 22% of the days that we located bats in their roosts (excluding date of capture), all tagged bats belong- ing to a social group were scattered in different trees. On only 19% of days did all tagged bats from a social group roost in the same tree. Mean number of different trees used simultaneously per social group (considering days when all ﬁve radiotagged bats of a social group could be located) was 2.76 (range ¼ 1e5 trees; N ¼ 45 observa- tions). These represent minimum values (i.e. minimum number of trees used simultaneously per social group), as we radiotracked only a small subsample of each social group (ﬁve bats out of an estimated 80e100 adult females per social group). Values varied with social group; for social group I (the group that used the largest number of tree roosts), all tagged individuals roosted separately on more than 40% of days and were together in the same tree only on the day of capture. Five trees were used con- tinuously throughout the radiotracking period: two of which were the trees equipped with automatic transpon- ders readers (Fig. 1a), where continuous bat activity was recorded from March until November.
Bats radiotracked in 1992 showed a similar roosting behaviour to that described above, with frequent roost switching (Table 1, Fig. 1c).
Giant noctules formed ﬁssionefusion societies similar to those of chimpanzees, elephants, dolphins and some forest bats. Group members were spread between several roost trees on a daily basis with frequent remixing through roost switching. Colonies were thus dynamic entities exceeding the limits of single trees.
Cryptic, Stable Social Groups: Close
Together. but Still Apart
Giant noctule colonies were not restricted to a single tree; however, all bats in Mar´ıa Luisa Park did not belong to a single colony. Three clearly deﬁned social groups coexisted in close proximity. The absence of apparent
barriers between the social groups and the ﬁssionefusion behaviour, with frequent roost switching and remixing of individuals, would make this population structure difﬁcult to detect through roost monitoring and checking group size. Roosting areas of the three social groups were generally distinct but with some degree of overlap (9e17%). Distinct forest patches used by different social groups have been identiﬁed for several forest bat species (Kronwitter 1988; O’Donnell 2000; Willis et al. 2003). Roosting areas for these species are typically nonover- lapping and occupy several hundred hectares. For giant noctules, overlapping roosting areas inside 20 ha can probably be explained by the high density of roosting opportunities within an otherwise barren landscape.
According to the deﬁnition of maternity colony (i.e. a group of reproductive females roosting together), the giant noctule population of Mar´ıa Luisa Park consists of three distinct maternity colonies. Our results suggest that these are stable over time, as adult females remained in their roosting areas and returned to them year after year for at least 5 years. Juvenile females also returned to their natal roosting areas in subsequent years. Female philopa- try is common in social mammals (Greenwood 1980). In bat societies, including those using ﬁssionefusion, female philopatry can restrict female-mediated gene ﬂow between nearby social groups, creating matrilineal socie- ties; male dispersal is usually the rule, preventing genetic isolation between social groups and populations (Petit & Mayer 1999; Kerth et al. 2000, 2002a; Castella et al. 2001; Metheny 2006). Fissionefusion maternity colonies of Bechstein’s bat, M. bechsteinii, are closed matrilineal socie- ties with extreme female natal philopatry and practically no immigration (Kerth et al. 2000, 2002b). In contrast, so- cial groups of E. fuscus are more lax and there is immigra- tion of females (Metheny 2006). Although giant noctules occasionally visited roosts of neighbouring colonies, we did not detect permanent changes of roosting area by any bats. It remains unknown whether immigration and genetic mixing between social groups occur.
Separate social groups and roosting areas have poten- tially existed in the park since at least 1992, with speciﬁc trees reused over the long term (up to 14 years).
Why do Forest Bats Switch Roosts?
Before lactation, giant noctules switched roosts on average every 2.52 days, a value similar for all-year male and autumn mixed populations of the similar species Nyctalus noctula, the common noctule, in Germany (2.57 days/roost; Kronwitter 1988). In our study, reproductive period did not inﬂuence signiﬁcantly the rate of roost switching, contrary to barbastelle bats Barbastella barbas- tellus in forests (Russo et al. 2005). However, a tendency towards slower rates during lactation was observed, and the low power of the test (0.38) for the small sample size (N ¼ 7) may have prevented signiﬁcance. Roost-switching frequency did not differ signiﬁcantly between individuals. Such as in B. barbastellus, for which roost switching was not inﬂuenced by age, sex or body condition (Russo et al. 2005), our results thus do not support that variable
requirements (e.g. thermoregulatory) are a major force shaping roost-switching behaviour by giant noctules.
Parasite load poses energetic costs for bats, partly ex- pressed by extra time spent in grooming (Giorgi et al.
2001). Costs might be more pronounced for individuals in maternity colonies, as decreased immunocompetence in gestating females and in juveniles is linked to increased parasite infestation (Christe et al. 2000). Roost switching could thus function as an antiparasite strategy (Lewis
1995, 1996; Reckardt & Kerth 2006). However, giant noc- tule subgroups rarely abandoned the roost trees simulta- neously. This situation, also observed in populations of N. noctula (Kronwitter 1988), differs from other forest bats (Lewis 1996; O’Donnell & Sedgeley 1999; Russo et al.
2005). At least the two trees equipped with automatic readers were used continuously from March, when females start congregating at roosts, until SeptembereNovember. If avoidance of parasites was driving roost switching, tree cavities should remain vacant for enough time to dis- rupt parasite life cycles (Lewis 1996), at least for parasites that do not accomplish their entire cycle on hosts. Although the effectiveness of this behaviour has been shown for some forest bats (Reckardt & Kerth 2006), it does not seem the case in giant noctules. Likewise, contin- ued use of tree roosts does not support the hypothesis that escape from predators learning about bat emergence is the main cause of roost change.
Rather, our results support the hypothesis that roost switching is a means of maintaining social bonds between colony members spread over multiple roosts e which is consistent with other studies (O’Donnell 2000; Willis & Brigham 2004; O’Donnell & Sedgeley 2006). In addition, we suggest that this behaviour may serve to retain and transfer knowledge about a large number of roosts (Kerth
& Reckardt 2003; Russo et al. 2005). The hypothesis that forest bats switch roosts because trees are ephemeral does not contradict this view, as the knowledge and pres- ervation of a pool of roosts would make the colony less vulnerable to roost losses due to natural processes or human management.
Why do Forest Bats form FissioneFusion
Several reasons might promote sociality in breeding females, for example better thermoregulation in larger groups, increased protection from predators or coopera- tive breeding (Kalcounis & Brigham 1994; Kerth et al.
2001; Kunz & Lumsden 2003). Obviously these direct ben- eﬁts of group living would only concern bats sharing the same roost simultaneously. What then is the advantage of belonging to a group larger than the cluster that is actu- ally roosting together, and that needs roost switching to maintain cohesion?
The ﬂuidity of ﬁssionefusion social systems allows
animals to counterbalance resource competition by split- ting in small groups and yet aggregate when it is beneﬁcial. Wild chimpanzees, for example, live within a large ‘community’ but form smaller ‘parties’ of variable size according to environmental, social or demographic
conditions which also vary in time and space (e.g. Chap- man et al. 1995; Lehmann & Boesch 2004). The existence of the ‘community’ allows chimpanzees to respond effec- tively to changes that alter their ﬁnely tuned balance between costs and beneﬁts of group living. In elephant populations, ‘core’ social groups (‘families’) fuse in re- sponse to variable levels of food competition or risk of pre- dation on calves, forming larger, ‘bond’ groups, which can also aggregate into ‘clans’ (Wittemyer et al. 2005). If direct beneﬁts of group living for forest bats also depend on subgroup size (such as thermoregulation or protection of young), or if optimum subgroup size varies in response to factors such as climate or roost characteristics, then sud- den changes in the environment (e.g. roost alterations, roost losses or climatic changes) or in the composition of subgroups (e.g. death, dispersal or immigration) might compromise these beneﬁts. As in primate and elephant ﬁssionefusion societies, belonging to a larger group allows restructuring bat subgroups in response to these changes, and in the long term, this would beneﬁt all members of the group.