Introduction

Animals modify their behaviour in response to seasonal changes in both the environment (e.g. weather) and their physiology (e.g., reproductive condition). For many mammalian taxa, behavioural patterns change at parturition. This may be due to different amounts of time invested in parental care and promoting offspring survival. Strategies to maximize offspring survival vary interspecifically and may include selecting areas of lower predation risk when offspring are most vulnerable at the expense of forage quality1, seeking out higher forage quality in areas of higher predation risk to meet the energetic demands of lactation2. Social behaviours may also vary with an individual’s changing reproductive status3,4, as the benefits of group living vary throughout the reproductive cycle. For example, ungulates often socially isolate prior to parturition2,5 while similarly in Scottish blackface sheep (Ovis aries), individuals with offspring take a more peripheral position in a flock6. The nature of interactions between individuals may also vary with breeding season, breeding status, and sex. For instance, females may increase agonistic interactions to protect offspring, while males increase affiliative interactions to improve access to breeding opportunities7. Additionally, within one sex, breeding status has been shown to be a predictor of behaviour, where mothers may increase their distance from conspecifics while nonmothers in the same group do not6.

Temperate bats have distinct seasonal cycles of movement and reproduction which includes a spring-summer season where females aggregate in large maternity groups8,9. At least some females switch roosts frequently throughout this period, resulting in changes in the composition of roosting subgroups each day10,11. While meeting the energetic demands of pregnancy and lactation, bats balance energy conservation via torpor with the potential costs of torpor to the timing of parturition and offspring survival12,13,14, and regularly employ social thermoregulation to minimize torpor use15,16. Accordingly, female bats may select warmer roosts or roosts that are used by a greater number of conspecifics to facilitate social thermoregulation15,17. Lactation represents a period of higher energetic demand for adult females than pregnancy18, but as ambient temperature is typically higher during lactation compared to pregnancy, it is not clear how energetic demands of different reproductive periods may influence bat roost selection. Not only is roost selection influenced by thermoregulatory needs, bats also need to balance foraging and parental care after parturition. Female little brown myotis (Myotis lucifugus) have been shown to travel shorter distances during foraging following parturition as females frequently return to roosts to nurse their pups19. Changes in foraging behaviour, along with the limited mobility of pre-volant young may impact the ability of females to change roosts.

Roosts used by bats may vary in temperature regime and proximity to forage sites, among other factors, and as such bats may switch roosts regularly20. Selecting appropriate roosts seems particularly important for reproductive females due to its influence on torpor use and therefore offspring development21. For example, Lausen and Barclay22 have shown that pregnant temperate bats select roosts with warmer and more stable temperatures late in the pregnancy period and during lactation22. Therefore, reproductive females should select roosts with appropriate microclimates that maximize energy balance given their physiological condition and the environmental conditions. However, as external environmental conditions change, conditions within the roost may also change requiring females to change roosts. This was demonstrated in big brown bats (Eptesicus fuscus), where high ambient temperatures increased the frequency at which females switched roost buildings or moved within buildings23.

Bat roosting decisions also influence, and are influenced by, social relationships24,25,26,27,28,29. Individuals may select roosts based on the presence of certain conspecifics29 and thereby influence their social environment by the information they bring, or physically by their effect on space and roost temperature. Social grouping may confer additional benefits, such as energy conservation, cooperation during foraging, or by reducing inter-individual conflict30. The benefits and effects of group behaviour, and thus the structure of bat groups may also change between seasons. Variation in network structure and subsequently social relationships between seasons and reproductive periods has been shown in other social mammal species including Tasmanian devil (Sarcophilus harrisii)31, rhesus macaques (Macaca mulatta)32, and black-tailed prairie dogs (Cynomys ludovicianus)7. However, these studies focused on differences between mating and nonmating seasons in both male and female individuals rather than relationships between only females. Previous studies of female bats have shown greater aggregation of individuals at roosts during pregnancy compared to lactation in some species11,33, but smaller roosting groups during pregnancy in others34, and colony fragmentation during post-lactation in little brown myotis35. Further, it is expected that movement patterns of females are affected during lactation due to nursing requirements19, and thus reproductive phases may have distinct impacts on the social structure of bat groups.

Little brown myotis colonies are typically comprised of interconnected and sympatric social communities, where the frequencies with which individuals share roosts with one another are not entirely explained by roost fidelity36. There has been some suggestion in other Myotis species that these interconnected communities may facilitate information transfer37,38,39, and it is known that social thermoregulatory benefits of co-roosting improve offspring survival and development15,22. We hypothesized that changes in the physiological requirements of female little brown myotis due to reproductive condition will affect the roosting decisions and social structure of a maternity colony. Based on this hypothesis, we tested the prediction that (1) the likelihood of adult female bats using the same roost on consecutive days would be highest during lactation, when females have non-volant offspring even when environmental pressure (i.e., weather) to switch roosts is accounted for. We also predicted (2) that, given the time and energetic requirements to revisit roosts at night to nurse pups19 as well as to carry non-volant pups between roosts, not only would females generally switch roosts less frequently during lactation, but that they would also use a smaller subset of monitored roosts more frequently and others less frequently than during pregnancy. Finally (3), we predicted that the network-level patterns of co-roosting associations in the maternity colony would change between the pregnancy and lactation periods as less frequent switching, as in predictions (1) and (2), would reduce overall connectivity of the maternity colony. As bats return to the same roosts within and between years, there is the potential for social familiarity and preferences to form among individuals, preferences which may become more apparent when the number of associates is reduced, as we predict may occur during the lactation period.

Results

Three hundred and seventy seven adult female little brown myotis met the criteria of having at least ten observations before 25th percentile of parturition dates and at least 10 observations after the 75th percentile of parturition dates in a single year (Table 1). Of the 377 individuals, 218 were included in multiple years and two individuals were included in eight of the nine years of the study. Numbers of tagged individuals who we suspect entered or left our monitored system following parturition, individuals for which there were at least ten day roost observations in one reproductive period but zero observations in the other reproductive period in a single year, did not correlate with the timing of the parturition cut-off date.

Table 1 Results of Akaike’s Information Criterion (AICc) analysis of candidate models explaining the probability of female little brown myotis reusing the same roost on consecutive days in Newfoundland Canada. The difference from the top-ranked models (∆AICc), balanced by the number of parameters (K), Akaike weights (wi) for all models and the sum of Akaike weights (Σwi) for models comprising ≥ 95% are also calculated. Parameters included in ‘Maternal care’ include reproductive period and a random effect of year, ‘Thermoregulation’ includes parameters of change in temperature, change in humidity, and an interaction between temperature and humidity, and ‘Flight conditions’ includes parameters of maximum wind gust and total precipitation.
Full size table

Variation in daily roost switching behaviour

Results of the model selection analysis suggested the candidate model that included maternal care and thermoregulation best explained the probability of bats reusing the same roost on consecutive days (AICc = 21530.8, model weight = 0.64), and the global model which also included flight conditions was the second best (AICc= 21532.0, model weight = 0.36) and only other model included in the 95% confidence set (Table 1). Our results showed that standard errors of parameter estimates did not overlap zero and were thus likely significant for changes in absolute humidity between consecutive days and reproductive period as predictors of the probability of bats using the same roost (Table 2). Our results specifically showed that a greater change in absolute humidity between consecutive days reduced the likelihood that a bat would reuse a roost (Fig. 1A). Also, as predicted, bats were more likely to reuse a roost on consecutive days following parturition (Fig. 1). Model-averaged parameter estimates revealed temperature, maximum wind gust speed, and the interaction between temperature and humidity were not significant parameters for predicting the probability of using the same roost on consecutive days as the model estimates plus or minus the standard errors overlapped zero (Table 2). However, it must be noted that there are likely many unaccounted for factors that also explain roost switching behaviour as model fit was low, with a corrected R2 of 0.24 for both the global model and the model including thermoregulation and maternal care.

Table 2 Model-averaged parameter estimates and standard errors to predict the probability of female little brown myotis reusing the same roost on consecutive days in Newfoundland, Canada. Parameters are averaged from all models included in the 95% confidence set which included a model of thermoregulation (changes in absolute humidity and temperature) and the global model.
Full size table
Fig. 1
figure 1

A) Predicted probability of bats reusing the same roost on consecutive days as a function of change in absolute humidity from one day to the next, change in temperature from one day to the next, and reproductive period. Changes in temperature had little effect on the probability of roost reuse while change in absolute humidity was highly influential. B) The proportion of observed bats that reused a roost on consecutive days across all dates and weather conditions in the pregnancy and lactation periods from 2012–2017, the period when hourly weather data were available. Each point represents the proportion of bats that reused the same roost on any given consecutive day period.

Full size image

Variation in roost fidelity and dyadic roosting relationships with reproductive period

On average, roost fidelity of individuals was significantly correlated (average = 0.66, SE = 0.02, range: 0.56—0.72) before and after parturition based on Spearman rank correlation (p < 0.001). This means that there was moderate correlation in the frequency that individual bats used each monitored roost before and after parturition, although roost use patterns by some individuals changed between reproductive periods. The nature of the change in behaviour varied greatly among bats, where for some individuals, this change was primarily a reduction in the number of different roosts used, while in others there was a change in the frequency that different roosts were used. For example, in 2017, one bat was detected in seven different roost boxes during the pregnancy period and only two boxes during the lactation period while many other bats showed no change in the total number of boxes used during a season. Meanwhile, box use frequencies also occasionally changed where, for example, one bat in 2019 used box 106 most frequently (15 of 25 detections) during the pregnancy period but box 108 (15 of 27 detections) most frequently during the lactation period.

Association strengths (SRI) between dyads during the pregnancy period were significantly correlated with association strengths during the lactation period (p < 0.001), indicating that, on average, many relationships between females were also consistent throughout the maternity season. However, there was considerable variability among dyads as some individuals that associated frequently during the pregnancy period never associated during lactation and vice versa. The average Mantel R score based on a Spearman Rank Correlation of SRI between reproductive periods was 0.51 (Range = 0.42–0.67; SE = 0.03) indicating that while some dyads showed consistent association patterns throughout the entire maternity period, others experienced changes in association patterns between pregnancy and lactation.

Changes in association patterns among bats did not appear to be related to changes in individual roost use patterns (Fig. 2; β = 0.84, SE = 0.51, p = 0.15). Analyses were also run using the half weight index (HWI) and yielded comparable results.

Fig. 2
figure 2

Relationship between correlation in roost use patterns and correlation in association strengths (Simple Ratio Index; SRI) between the pregnancy and lactation periods of each year. Linear regression revealed a significant but weak effect of correlation in roost use patterns between reproductive periods on the similarity in association strengths (SRI; β = 0.84, SE = 0.51, p = 0.15, R2 = 0.17).

Full size image

Variation in network structure among reproductive periods

Each of the Coefficient of Variation of the Simple Ratio Index (CV SRI), graph density, and clustering coefficient changed significantly between reproductive periods (all p < 0.05). The CV SRI increased during lactation while graph density and clustering coefficient each decreased (Fig. 3). Analyses using HWI and the CV HWI yielded comparable results.

Fig. 3
figure 3

The change in (A) coefficient of variation of the simple ratio index (CV SRI), (B) graph density, and (C) clustering coefficient between pregnancy and lactation periods. Each line connects points relating to a single year. P-values are reported based on a paired t-test comparing values between reproductive periods. CV SRI showed a significant increase following parturition while graph density and clustering coefficient both showed a significant decrease.

Full size image

Discussion

In animals that invest significantly in parental care, it is expected that behavioural patterns will change once offspring are present. We demonstrated that female little brown myotis generally maintained similar roost fidelity during lactation as during the pregnancy period (prediction 2) but were less likely to switch roosts during the lactation period in response to changing weather conditions than during the pregnancy period (prediction 1). Roosting association strengths between individuals during pregnancy were also predictive of these relationships during lactation, and our results suggested that although patterns in roost use may change once pups are present in the system, individual variation in selection for both associates and roosts were generally maintained throughout the maternity season (prediction 3). However, just as bats may be limited in their ability to switch roosts during the lactation period, likely due to the need to frequently revisit the same roost at night to nurse non-volant pups19, bats in our study did not maintain all co-roosting social connections that existed prior to parturition into the lactation period, as evidenced by declines in network density and clustering coefficient. Overall, our findings suggest that reproduction may be an important factor in the patterns of co-roosting associations of female little brown myotis and influence the wider structure of the maternity network.

In support of our first prediction, regardless of weather conditions, bats were more likely to reuse a roost on consecutive days following parturition. These results could reflect that females are either more limited in their ability to change roosts when caring for a pup and/or that there is a benefit to keeping a pup in the same roost. Importantly, changes in weather conditions did have an impact on the probability of switching roosts across both reproductive periods, and we found changes in absolute humidity to be the most influential environmental factor in determining whether female bats switched roosts between consecutive days, with no significant effect of changes in temperature. The influence of absolute humidity is also supported by findings demonstrating absolute ambient moisture as the most important factor in determining evaporative water loss and the surface temperature of animals40, and thus switching roosts during periods of high absolute humidity may allow females to avoid torpor use and the associated delays in gestation14. It remains to be determined how reduced roost switching during lactation at periods of high absolute humidity may impact survival of both the mother and offspring.

Meanwhile, given that female Bechstein’s bats (M. bechsteinii) preferred warmer roosts during the lactation period41, it is surprising that temperature did not have a significant effect on roost switching patterns in little brown myotis. This is unlike the findings of Ellison et al. (2007), and that the effect of temperature appeared similar in both reproductive periods. These results may be due to the similar construction of the monitored roost boxes along with the presence of other bats, therefore making social thermoregulation more influential on the temperature of the roost. Wind gusts and precipitation also did not appear to be influential in affecting daily roost switching of bats in this study. Therefore, conditions outside the roost may not be important determinants of roost selection and strong winds or heavy rains not limiting to mobility, particularly given the close proximity of the roosts we monitored. However, our results also suggest that there are many unaccounted-for factors influencing the roost switching decisions of female little brown myotis. Factors such as changes in foraging sites and other social mechanisms may represent other, unmeasured factors in roosting decisions. Roost type has also been shown to affect roost switching in response to precipitation in other temperate bat species in Atlantic Canada42, and therefore the use of artificial roost boxes may also have impacted our results. The distance between our study site and the nearest weather station at Pippy Park (≈ 54 km) and the variable nature of precipitation and wind across a landscape compared to temperature and humidity may also have limited our ability to make inferences about precipitation and wind as factors influencing roost switching decisions, particularly given that Pippy Park is much closer to the Atlantic Ocean (≈ 4 km) than Salmonier Nature Park (≈ 20 km).

We found that although individuals maintained fidelity to the same roosts across the pregnancy and lactation periods, there was evidence of less roost switching during the lactation period. Although previous research has found that maternity group formation is relatively stable in little brown myotis and Yuma bat (M. yumanesis) groups35, it is unsurprising that roost use patterns differed slightly between reproductive periods given previous evidence for differential roost preference41, reduced mobility and roost switching during lactation19, and that seasonal changes in body composition and hormone status, such as those experienced throughout a reproductive cycle, can impact an individual’s thermoregulatory needs43. It is additionally possible that during the pregnancy period, bats regularly use more roosts, and then reduce to this “core” set during lactation to which they have high fidelity as our results on roost switching and changes in roost fidelity suggest. As we do not know how many unmonitored roosts are used by the bats in Salmonier Nature Park, further study would be needed to quantify the number of different roosts used by reproductive females in both pregnancy and lactation periods, and how females may be selecting the roosts that they use the most frequently. However, given the similarity in construction of monitored roost boxes and lack of information on prey availability, we can only speculate as to why bats differed from each other in their potential roost preferences.

It is important to note that in social animals, not only are behavioural decisions influenced by the environment an individual uses, but that these decisions can also alter the environment itself, a concept now known as social niche construction44,45. Based on the relationship between individuals and their social environment, decisions made by parents about site selection and social association with conspecifics throughout their reproductive cycle can also influence the wider population and therefore the behaviour of conspecifics. The observed changes in roosting decisions here, in turn resulted in changes in the social structure of the animal group, specifically a decline in network connectivity, and it is unknown whether or how this reduction in connectivity during lactation improves outcomes for mothers and pups. We reported a weak relationship in which years with higher correlation in roost use patterns between reproductive phases also displayed a higher correlation in association patterns. Given this, we cannot rule out our third prediction, that in addition to limitations in roosting decisions imposed by parental care, the social needs and preferences of mothers may change once pups are present. We found an overall decline in network connectivity, an expected result given the less frequent roost switching of lactating females. The higher network connectivity during pregnancy may also facilitate information transfer as in rhesus macaques, females played an important role in information transfer during the mating season32. Our results were also consistent with other network studies on bats that report little brown myotis and big brown bat roosting groups are larger prior to parturition11,33. However, as we were only focused on individuals which met minimum criteria over the study period and did not have information on the number of pups in the system or the number of untagged bats, it remains uncertain whether changes in roosting group size, as reported by Olson and Barclay46, also occur in this system.

While it remains unclear what the benefits or consequences of reduced network connectivity during lactation may be, for the endangered little brown myotis, it may be important to consider these patterns when supplementing roosting habitat. As such, while roosts close together as recommended by Holroyd et al.47 may allow greater connectivity and reduce the costs of switching roosts when pups are present, a greater availability of more distant roosts in variable locations may be important to allow pregnant females to accommodate their thermoregulatory needs. We also suggest that even in small populations, multiple roost options is important for ensuring that females can reproduce successfully for years to come. Given that parturition in this species is coincident with seasonal changes in weather conditions, we cannot rule out that some of the observed changes in roost use patterns may be attributable to seasonal differences in the environment independent of reproductive needs. That said, given the consistency of these changes with our expectation based on reproductive pressures, and the importance of reproductive period when assessing roost switching behaviour even when environmental factors are controlled for, it is reasonable to infer that the reproductive cycle is an important influence on network structure. As the reproductive status of many individuals in our study is uncertain, we could not determine how a shift in reproductive condition may affect roosting decisions at an individual level. Future study investigating parturition timing in conjunction with behavioural changes at an individual level will be crucial for gaining a deeper understanding of the relationship between reproduction and social behaviour as parturition in little brown myotis and other temperate bat species is largely asynchronous10,48. Thus our data represents an approximation of when behavioural shifts are expected. However, individual bats may change their habitat or social needs at different times. As our analysis represents a broad overview of what may be occurring in the population at an average parturition date, it is expected that changes would be even more pronounced when analyses are refined to the level of the individual. As changes in roosting patterns are expected to occur at the time of parturition, it would be of interest to investigate whether the association patterns between individuals may be based on similarity in reproductive timing.

Other limitations of our research include that our study focused on individuals present throughout the study period and transient individuals may be influencing network structure and interacting at unmonitored locations. We do not know how the inclusion of nearby, unmonitored roosts would affect these results but expect that patterns of association and roost-use of our included bats are reflective of their general behaviour patterns given that we required a minimum number of observations for inclusion in network analysis. Finally, we assumed that years were sufficiently independent from each other due to the large number of individuals that entered or left the study system between years, but the consistency in individual roosting decisions and social relationships between years remains to be determined. Similar studies in systems where a known, high percentage of individuals are monitored over multiple years or in a closed system would be highly beneficial for more confidently characterizing the causes and consequences of the patterns we identify here.

Overall, our data support the contention that female little brown myotis roosting patterns and thus the social landscape of maternity groups are influenced by reproductive processes. Females expectedly switched roosts less frequently and co-roosted with a smaller subset of the maternity group during the lactation period, but what the fitness consequences or benefits may be for these behaviour changes remains to be determined. Consistent with studies in other mammal species1,5,49,50, these findings lend additional support to the contention that in animals demonstrating maternal care, the presence of offspring has a strong influence on decisions parents make, and that individuals require flexibility in responses to their environment throughout a reproductive cycle.

Methods

Data collection

Using mist nets (Avinet, Dryden, New York, USA) and harp traps (Austbat Research Equipment, Lower Plenty, Victoria, Australia) little brown myotis were captured in and around Salmonier Nature Park (n = 1604; Lat: 47.3º, Long: -53.3º) Newfoundland, Canada from 2012 to 2021 between May 15th and August 19th of each year. Reproductive status51 and age52 were determined for each individual and then a passive integrative transponder (PIT) tag (0.09 g; EID-ID100 implantable transponders, EIDAPInc, Sherwood Park, Alberta, Canada and Trovan Electronic Identification Systems, North Ferriby, UK) was subcutaneously implanted between the scapula. Transponder antennas (LID650, Dorset Identification, The Netherlands) were deployed at eleven roost boxes within a 1.1 km2 area from April to September each year to collect data on roost use. One roost box was placed on a building while the other ten were paired on five poles throughout the study area. Distances between boxes ranged from centimeters, for those placed on the same pole, 10–20 m between poles within a section of the study area, and 100 m to 1 km between sections. Further details on the study system can be found in Sunga et al.36. The last detection before sunset, but after sunrise of the same day, was used to infer the roost box in which an individual spent the day and represents an ‘observation’. Due to potential errors in PIT-tag recording, an unknown number of unmonitored roosts, and an unknown proportion of the population being untagged, it is expected that many instances of roost use were unrecorded.

Across the entire study period, there were 1083 tagged adult females, including some females that were initially tagged as juveniles. Males were not included in these analyses as they were rarely detected at maternity roosts and do not have the same variation in energy requirements as females throughout the maternity season18. Juveniles were excluded as it is uncertain whether juvenile movements are independent from their mothers37,53, and juveniles would only be present for the second portion of the maternity season. Among all adult females, the number of day roost records varied between 1 and 88 (mean = 22) day roost records across a maternity season, and the number of tagged adult females within a monitored roost varied anywhere from 0 to 110 individuals on any given day. Adult females that may have been non-reproductive in a given year were included in our analyses. Data were constrained to include detections between 15 May and 15 August of each year and only individuals that were observed > 10 times in each of the pregnancy and lactation reproductive periods in a single year were included54. For each year the pregnancy period was determined as the period from 15 May until the date when up to 25% of femles in which parturition could be detected were expected to have had their pups in each year (Table 3)55. The lactation period was determined as the period from the date when at least 75% of females were expected to have had their pups until 15 August. Observations between these dates were not included to reduce contamination of data regarding the pregnancy and lactation periods. As parturition is relatively asynchronous, these cutoffs were chosen to reduce the number of individuals already lactating included in the parturition period while also ensuring that there were not many still pregnant individuals for a significant portion of the lactation period. Further, a later cut-off would risk more individuals beginning to depart for swarming and hibernation sites during the observation window and thus fewer observations on their lactation-state roost use and association patterns. Individuals were included in each year in which they met these criteria, and so some individuals were included in multiple years of analysis.

Table 3 The number of individuals included in each year of analysis by meeting the criteria of at least 10 observations at monitored day roosts both before the end of the defined pregnancy period (25th percentile of population parturition dates) and after the start of the defined lactation period (75th percentile of population parturition dates)55.
Full size table

Variation in daily roost switching behaviour

To assess the extent to which reproductive period may influence daily roosting decisions in bats, we applied an information theoretic approach using Conditional Akaike’s Information Criterion43 of a candidate set of mixed effects models to explain variation in the likelihood of reusing the same roost on consecutive days based on logistic models. Despite some roosts providing some internal microclimate variation 21, we expected that large changes in external weather conditions are more likely to impact internal roost conditions or nearby foraging opportunities, and therefore lead to roost switching. However, we also predicted that reproductive period would also influence the likelihood of roost switching, even when external weather conditions are accounted for. We quantified the probability of individual bats reusing the same roost on consecutive days in each reproductive period and in response to daily weather conditions between 15 May and 15 August. This analysis included information for all individuals for which we observed day roosting in our monitored boxes on consecutive days on at least one instance. As hourly precipitation data were not available for the entire study period, this analysis included data only to 2017, with dates included in the analysis only if precipitation data was available on consecutive days.

Each bat on each day was assigned a 0 or 1 based on its location on the consecutive day where a 1 indicated a bat that was in the same roost on day one and day two while a 0 indicated that a bat was in a different, monitored roost on day one than on day two. Any bats that were recorded on day one, but not recorded on day two were not included in the analysis due to the possibility that the disappearance may actually be a missed read at the same roost, creating contamination of our dataset. Although it is expected that the number of bats that reuse a roost between consecutive days may be underestimated due to missed observations, it is expected that these observations were missed randomly and do not covary with our independent variables.

Our candidate model set consisted of multiple, nonmutually exclusive explanations for the roosting decisions of female bats. A candidate model of thermoregulatory effects driving roost use included change in mean nightly temperature between consecutive days (⁰C), change in mean nightly absolute humidity between consecutive days (g/m3), and an interaction between these terms as it is expected that a night that is both cold and humid would pose a different thermoregulatory challenge than a night that is cold and dry13. Both temperature and absolute humidity were calculated based on the average of hourly recordings taken between 02:00 and 07:00 on each day, when we expected bats would make their day roosting decisions. The change in temperature (∆Cº) and change in absolute humidity (∆AH) was then calculated as the absolute difference between the mean on each day, with the expectation that a greater change in either of these variables would decrease the likelihood of a bat reusing the same roost. This is because roosts may have different microclimates and therefore if weather drives21,] at least in part, roost selection decisions, individuals are less likely to reuse the same roost as weather conditions show greater variability one day to the next. A candidate model of flight conditions included maximum nightly windspeed and total precipitation, as it was expected that very high winds or heavy precipitation would result in more energetically expensive flight and increase the likelihood of bats reusing a roost. The maximum nightly windspeed was calculated based on the maximum reported wind gust between 02:00 and 07:00 on each day, and total precipitation as the sum of all precipitation also between 02:00 and 07:00. Finally, a candidate model of maternal care was tested broadly with a binary variable of whether the date was before (0) or after (1) the population parturition date estimate, separating the pregnancy and lactation periods for this study (Table 3), where bats were predicted to more frequently reuse a roost on consecutive days during the lactation period, when more females are expected to be nursing young. The random effect of year was included to account for annual changes based on factors such as spring climate conditions and timing of emergence from hibernation13,56,57, and it is unknown whether roost preferences vary annually.

A random effect of individual was also included to account for the fact that some individuals may be included in this analysis more than others due to greater fidelity to our monitored roosts. All weather data with the exception of hourly precipitation information were obtained from the Environment Canada weather station at St. John’s International Airport, approximately 57 km northeast of Salmonier Nature Park, through “weathercan”58. Precipitation data was provided from the Pippy Park weather station (Government of Newfoundland and Labrador, Department of Environment and Climate Change, Water Resources Management Division (WRMD)), approximately 54 km northeast of Salmonier Nature Park. Although available weather stations are not immediately near our study site, given we are looking at changes in weather conditions rather than absolute conditions, we expect that the stations are close enough such that large changes at these nearby weather stations equate to large weather changes at our study sites.

Our candidate model set al.so included models that represented combinations of these potential explanations for roost switching behaviour, such as torpor (temperature and humidity) with flight conditions (wind and precipitation) in one model (Table 1). Additional two-way interaction terms were not included to avoid overfitting of models without clear biological precedent. Further, due to a correlation between variables, ∆AH and precipitation were never included in the same model and precipitation was omitted from the global model and the combined model of maternal care and flight condition model. Using a Corrected Akaike’s Information Criterion (AICc) we calculated AICc values and determined the best performing model. We then selected the 95% confidence set of models, and applied multi-model averaging of variable estimates59.

Variation in roost fidelity and dyadic roosting relationships with reproductive period

To assess possible changes in roost preferences between reproductive periods in each year, we generated matrices of roost fidelity, the proportion of days in which individuals roosted in each monitored roost box. Only individuals that were observed > 10 times in both of the pregnancy and lactation reproductive periods in a single year were included to ensure sufficient sampling of each period within individuals. For each reproductive period-year combination, we generated an individual-by-roost matrix populated with the proportion of days an individual bat was observed in each monitored box calculated against the total number of days that bat was observed. We then compared these roost fidelity matrices between reproductive periods for each year using a Spearman Rank Correlation to quantify the extent to which roost preferences during pregnancy may be indicative of roost preferences during lactation.

Next we calculated the association strength between pairs of individuals (dyads) based on the simple ratio index (SRI) which is a proportion of the number of instances a dyad was found to be associating compared to the proportion of instances dyads were not associating60,61. We applied the gambit-of-the-group assumption61 such that dyads were assumed to be associating if they were recorded in the same roost on the same day. To assess associate preferences, we generated an individual-by-individual matrix with values corresponding to the SRI between each dyad using the function get_network in the package “asnipe”62. In each year, we conducted a Mantel test using the function mantel in the package “ecodist”63 with 10,000 permutations to test the similarity of associate preferences before and after parturition. Mantel R was calculated using the Spearman Rank Correlation to account for skew in the distribution of SRI values.

We then performed a simple linear regression to determine if the correlation in roost use by individuals before and after parturition was predictive of the correlation in SRI between reproductive periods. This allowed us to make inference on how a change in roost preferences may impact changes in social relationships and vice versa.

Variation in network structure among reproductive periods

Using individuals that were detected at least 10 times in each of the pregnancy and lactation periods, we assessed the changes in network metrics between pregnancy and lactation periods. As we only used individuals which met the above-criteria in both periods, the same number of individuals were used to construct each reproductive period network within a year. For each reproductive period-year network based on co-roosting associations, we calculated several metrics. The coefficient of variation of the SRI (CV SRI) was calculated by computing the standard deviation of all SRI values over the mean of all SRI values then multiplying by 100. Graph density, a proportional measure of the number of connections present in the network versus the number of potential connections, was calculated using the function edge_density. Clustering coefficient, the probability that individuals with a common associate are also themselves connected64 was calculated with the function transitivity in the package “igraph”65. We also calculated community assortativity (Rcom) a measure of the reliability of the assortment of individuals between potential social communities66. For each year, we conducted a paired t-test for each of the following network metrics; graph density, clustering coefficient, and CV SRI, to determine if there was a significant difference before and after parturition.

All analyses were conducted and figures created in R version 4.0.0 (R Core Team 2020).