Study area

The study area was located in an experimental section (450–940 m elevation) of Chichester State Forest, 200 km north of Sydney, Australia. The experimental section, which comprises eight small (13–97 ha) catchments, was established in 1974/75 to investigate the effects of logging on water flow and quality (see [37] for site description), embedded within extensive forests at varying stages of post-logging regeneration. The site average rainfall from 1974–2013 was high (1547 mm), and despite five consecutive years of below average rainfall during the study, only 2002 recorded < 1000 mm.

Six of the eight catchments were clearfelled in 1983, with the remaining two catchments of old growth left undisturbed [37]. Unlogged buffers of various widths lined creeks (20 m minimum each side of creek). Prior to experimental logging, tall wet sclerophyll forest (> 35 m) covered most of the catchments, dominated by Sydney blue gum Eucalyptus saligna, silvertop stringybark E. laevopinea and rainforest. Rainforest species dominated creek-lines and much of the understorey. In 1997, after 14 years of regrowth, trees occurred at a much greater density in the regrowth catchments (3500–6100 stems ha^-1) compared to unlogged catchments (744 stems ha^-1) [37]. Rainforest understorey contributed most stems (700 ha^-1) in the unlogged catchments, with overstorey eucalypts occurring at a density of just 44 ha^-1.

Experimental design

Because of the area requirements of bats, each site replicate spanned two adjacent catchments of the same forest treatment, extending the small spatial scale of the individual catchments in the original hydrology experiment [37] to an average of 78 ha per treatment. Unlogged treatments were dominated by old growth forest (includes rainforest) (a mean of 68% of a 500 m circular buffer centred on unlogged treatments) while eucalypt regrowth (excludes riparian buffers) represented an average of 56% of the buffer surrounding regrowth treatments. We compared bat population dynamics in four areas of regrowth forest with two areas of unlogged forest (Fig 1). One regrowth area and unlogged area sampled lower elevation forest, one regrowth area sampled mid-elevation, while two regrowth and one unlogged area sampled higher elevations. Three different clearfell logging practices yielded the regrowth forest: logging plus a regeneration burn at Corkwood, logging without a regeneration burn at Raingauge 9, and logging followed by eucalypt plantation establishment at Kokata and Coachwood. Riparian buffers were retained on creeks and scattered old, hollow trees and unharvested rainforest were also retained. Although logging practices varied, vegetation structure in logged treatments was similar, based on estimates of stem density and vegetation cover, after 16 years of regrowth when our study began [25, 37].

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 1. High elevation eucalypt forest in the climate refuge study area.

(a) unlogged forest catchment showing tall emergent eucalypts above a dense rainforest understorey (b) 27 year old regrowth catchment showing dense stands of mixed eucalypts and rainforest species with a small 4WD trail used for harp-trapping bats.

https://doi.org/10.1371/journal.pone.0191471.g001

Bat sampling

The six study locations were sampled every year in autumn (~ four months after parturition) for 14 years (1999–2012). At each site, we standardised trapping effort, with each capture event consisting of two consecutive nights of trapping using 3 harp-traps set within a 200 m length of a 4WD trail. Time of year was standardised to the same two week period at the beginning of autumn (March) each year. We conducted one trapping event per year to minimise trap avoidance behaviour, except the first two years of the study when we employed three trapping events each autumn (separated by ~ two weeks) for the purpose of maximising the number of bands in the population for recapture in later years. Also, we did not extend trapping beyond two nights for each site to minimise trap avoidance. The same locations were trapped each year of the study.

We applied approved flanged-metal bat bands from the Australian Bird and Bat Banding Scheme to the forearm of captured bats. Study protocols were approved by the Department of Industry-Lands and Forestry, Forestry Corporation animal ethics committee (Animal Research Authority: 12/17). Band injuries on recaptured bats were noted (0 = no injury, 1 = minor abrasion, 2 = minor swelling, 3 = serious swelling of forearm requiring band removal). We recorded < 1% of recaptures with a swelling that required us to shift the band to the other forearm, indicating that band injuries were rare (unlike split metal types, [38]). Bats were sexed, weighed, forearm lengths measured and teeth wear noted as an independent, though coarse, method of aging bats (1 = shiny and sharp, 2 = dull, but sharp, 3 = dull and noticeably worn). We note that teeth wear will vary based on a species’ diet, but the species we studied have a mixed diet of small moths and beetles. As epiphyses had typically fused by the time of Autumn trapping, we were unable to reliably identify juveniles by this method. Eight percent of bats were classified as juveniles (3–4 months old) at capture, but on some occasions none were identified as juveniles.

Modelling bat recaptures

All 18 capture events, three in each of 1999 and 2000 and one per year thereafter, were treated as distinct capture events in the capture-mark-recapture (CMR) analyses. Within-year capture events in each of 1999 and 2000 were two weeks or a month apart, while successive capture events from 2001–2012 were one year apart.

We employed program MARK to construct Cormack-Jolly-Seber (CJS) models for apparent survival and recapture with time, species, sex, and logging treatment (regrowth versus unlogged) as groups or effects and a two-week time step. We then used the Akaike information criterion (AIC_c) [39] for model comparison and selection. This approach provides a systematic method to select parsimonious models, as measured by the number of parameters in the model, which trade some bias for precision in parameter estimates. We reduced model complexity in order to capture the strongest biological signals in the data and avoid being misled by poorly estimated parameters [40]. One outcome of this approach was that site elevation could not be incorporated into our model because of the additional complexity it would have added given the number of sites and bats sampled and the need to allow for the presence of transients in the modelling. Nevertheless, by having logging treatment represented at both high and low elevation, we avoided confounding logging treatment with an elevation effect.

Because we could not reliably discriminate adults from juveniles when banding in autumn we did not distinguish age classes and this heterogeneity may have introduced some bias into our survival estimates. However, most males (81%) had swollen and descended testes at this time of the year and so were functionally adult. Bats are highly mobile, suggesting that models distinguishing between transients and residents might be relevant, especially as some transients may be sub-adults.

We used program U-CARE [41] for goodness-of-fit testing, which indicated the presence of transients (i.e., bats with zero apparent survival) but provided no evidence for a trap effect or overdispersion requiring the use of QAIC when species and time were included as effects in the global model (see Supplement for details). We therefore employed AIC_c for model ranking and utilized the time-since-marking (TSM) model of MARK [42], §7.4) to distinguish between newly marked (mixture of transients and residents) and already marked bats to implement the model of Pradel et al. (1997) [43]. For such models, survival of already marked bats is interpreted as survival of residents. The survival of newly marked bats is used to compute the proportion of newly marked bats that are residents (as the ratio of survival of newly marked bats to that of already marked bats). Note that as recapture probability is conditioned on the presence of marked individuals, it is unaffected by the presence of transients.

The TSM model with time*species*sex*treatment dependence served as our global model but the very large number of model parameters and the typical sparseness of mark-recapture data entailed it served primarily as a heuristic model. The CJS models and AIC_c comparisons led us to choose the structure sex*time for recapture (no evidence for logging treatment or species effects) but we initially retained time*species*sex*treatment for survival in the TSM model. While maintaining a distinction in survival between newly and already marked bats and between capture events within years and capture events between successive years, we simplified the dependence of survival on effects iteratively until we obtained models for which MARK estimated all remaining parameters. Some simplifications reflected insufficient data to discriminate effects, e.g., to discern effects on survival between capture events within years. Time dependence remaining in models was simplified by modelling it with logistic regression using the weather covariates described in Table 1. We hypothesized that survival over years might covary with over-winter conditions, as measured by MinTempWint when bats are torpid [21, 22]. We also conjectured that spring-summer conditions, both temperature and rainfall, prior to capture when bats are active and foraging would be important, e.g., through influences on insect abundance [3] for both survival and recapture, while temperature and rainfall during the capture period might influence recapture (Table 1). Note that 6laggedRain was negatively correlated (-0.8) with MaxTempSumm so these covariates were never employed in the same model. The same value of MaxTempSumm was employed for capture events within the same year. In a preliminary analysis, we also tested the mean maximum daily temperature during the previous winter, mean minimum daily temperature in the previous summer, rainfall in the previous 12 months (definitions analogous to those in Table 1), rainfall during the capture period and 6- and 12-month lagged measures of Southern Oscillation Index, but none of these covariates proved competitive with those listed in Table 1 and were discarded.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 1. Covariates employed in modelling time dependence in survival and recapture.

https://doi.org/10.1371/journal.pone.0191471.t001

This procedure for simplifying both group structure and time dependence yielded a model singled out by AIC_c ranking, i.e., which was more competitive than more complex models or similarly complex models with alternate structure. The recapture structure was p(sex+ MaxTempSumm + AvMinTempDur) and the survival structure denoted here φ(1) is described in Table 2. This model was further simplified by reducing dependencies of covariates on group effects, and/or removing group effects or covariates, resulting in a collection of 35 models over which AIC_c increased gradually. Thus, many slight variations of the model φ(1)p(sex+ MaxTempSumm + AvMinTempDur) structure were competitive alternatives (S2 Table). We adopted the AIC-cut-off notion with a value of four for a confidence set of models [39]:170). Throughout our analysis we exploited Burnham and Anderson’s [39] parsimony argument to reject as competitive a model that resulted by the addition of one parameter to a model already under consideration without changing the deviance.

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Table 2. Survival structure in model φ(1)p(sex+ MaxTempSumm + AvMinTempDur).

The two-letter species abbreviation indicate survival rates for that species, qualified if the rate was sex specific by adding M or F and treatment specific by adding R(regrowth) or U (unlogged), e.g., CmFU, designates a survival rate for female Chalinolobus morio in unlogged treatments. The absence of a qualifier means the survival rate is independent of that effect. Coincidences between survival rates are expressed with equality signs. All the survival parameters in the second column were modelled as logistic regressions on the covariate MaxTempSumm with the same regression coefficient but their own intercept, independently of whether they were newly or already marked. Note that survival rates within the same year only apply to the capture season, and should not be extrapolated beyond this period.

https://doi.org/10.1371/journal.pone.0191471.t002

Our final estimates of survival and recapture were obtained by model averaging over the confidence set [39], a compromise between accommodating the lack of a clear-cut model structure and excluding poor models from contributing to parameter estimation [44]. Standard errors for parameters derived from MARK output (e.g., survival rates converted to annual time steps, proportion of residents) were estimated using the delta method.

Abundance (population size)

Modelled recapture probability is the probability of capturing a marked individual present at the capture event. For abundance estimation, we assume that, apart from the influence of modelled effects on recapture, all bats present at the capture event, whether marked or unmarked, transient or resident, were equally likely to be trapped, i.e., we assume that the modelled recapture probability also estimates the probability of capture of any bats present at the capture event. For each sex of each species, at each site, abundance of bats at a capture event was estimated by dividing the number caught at that event by the relevant probability of capture at that event, and the SE of this estimate was estimated by the delta method ([45], equations (18.6) and (18.23)). To estimate the number of residents at a capture event, the numbers of already and newly marked bats caught were each divided by the recapture probability to estimate their separate numbers at the capture event. The estimated number of newly marked bats was multiplied by the estimate of the proportion of residents amongst the newly marked bats and this estimate added to the estimate of already marked bats to obtain the overall estimate of the number of residents at the capture event (and SEs again obtained by the delta method).

Mean life expectancy

Mean life expectancy after banding was calculated from survival estimates for resident bats of different species, all of which were banded in autumn when they were at least three months old. We used the formula of Seber (1982) [46], though note that use of apparent survival will underestimate life expectancy:

Analysis of logging effects: Abundance and body condition

For abundance, we observed general patterns relating to sex, species, and treatment (site) by averaging over capture events. To test whether logging history influenced estimated bat abundance at a site, we summed estimated male and female abundance separately for each species in each year and then averaged across years per site. We used pre-planned contrasts to compare between logging treatments using trap-site elevation as a co-variate. Body condition was calculated for each individual by dividing body mass by forearm length. Mass was measured with a Pesola spring balance (50 g) and forearm was measured with a Vernier caliper. All measurements were standardised by handling bats in the afternoon, after retrieval from harp traps in the early morning. However, we acknowledge that different individuals would have foraged for different periods prior to capture on the previous night and this may have led to additional variation in body condition measures. Body condition was used as the response variable in a mixed model analysis with logging treatments, species and sex and all interactions as fixed effects and individuals as a random factor. A significant interaction with logging was taken as evidence for a logging effect. Only recaptures were analysed as these were known residents that experienced conditions that could be confidently allocated to a treatment.

Analysis of time series: Abundance and body condition

To examine trends in abundance over time, for each capture event, we summed over sex and species obtaining a vector of six site-specific time series and also summed over sex and sites to obtain a vector of four species–specific time series. Dynamic factor analysis (DFA; [47]) expresses a vector of time series as a linear combination of random walks and covariates (here MaxTempSumm) plus residual, and thus searches for common trends in multiple time series. DFA was implemented in package MARSS in R [48]. We used the AIC_c output of MARSS to select amongst simplifications of each global model. A DFA was also conducted to assess trends in body condition over time with species and sex kept separate, but summed over sites, to give a vector of eight time series.

Survival model structure

Initial AIC_c ranking of TSM models eliminated more complex models in favour of a model denoted φ(1)p(sex+ MaxTempSumm + AvMinTempDur). We describe the structure of this model in detail as final model selection is conveniently described in terms of variations on this particular model structure. The survival component φ(1) is given in Table 2 while recapture was modelled as a logistic regression on the two covariates MaxTempSumm and AvMinTempDur (Table 1 and S1 Table). For each of these two covariates, the intercept, but not the slope, depended on sex; i.e., each covariate had an additive structure with respect to the interaction with sex, and the slope was positive.

Of the 35 models obtained by further simplifying the model φ(1)p(sex+ MaxTempSumm + AvMinTempDur), 22 fell within four AIC_c units of the AIC_c-best-ranked model, including the model φ(1)p(sex+ MaxTempSumm + AvMinTempDur) itself at 2.5 ΔAIC_c units. These 22 were selected as the models over which survival and recapture parameters were averaged (S2 Table). Since the models that we averaged were simplifications of the model structure in Table 2, we employed that structure to describe the model-averaged survival and recapture parameters.

Variation in recapture probability

Recapture probability was modelled as covarying (Table 1 and S1 Table) with weather conditions prior to (MaxTempSumm) and during (AvMinTempDur) annual trapping and additively on sex (species and treatment groups were not supported by AIC_c selection). The additive structure made male recapture slightly larger than female recapture (≈ 4%) for a given capture event. While both regression coefficients were positive, the standardized coefficient of MaxTempSumm was about 5 times larger than that of AvMinTempDur. Male recapture probability varied from 0.2–0.51 across the study, being higher after warm summers (MaxTempSumm) with a subtle negative influence of cold nights when trapping (AvMinTempDur) (Fig 2).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 2. Male recapture probability and its covariates.

Male recapture probability (●) at the n’th capture event (n = 1 is 99_2, the second capture event in 1999; n = 2, 99_3, the third capture event in 1999; n = 3, 4, and 5, the first, second, and third capture events in 2000, respectively; and n > 5 is the capture event in year 2000 + (n-5), through 2012); MaxTempSumm (▲) and AvMinTempDur (■). Since recapture probability was modelled as additive for sex, the graph for female recapture probability is similar but slightly less (≈ 4%) at each capture event.

https://doi.org/10.1371/journal.pone.0191471.g002

Variation in resident survival

A single, effect-independent survival of 0.89 ±0.04 per two weeks was estimated for survival of already-marked bats between capture events within the same year (applicable only to the period of capture). All resident survival rates from the (final) capture event of one year to the (initial) capture event of the next year are expressed as annual rates. Species, sex and treatment effects influenced survival. In line with our prediction that the forest refuge would buffer survival from climate extremes, modelling assigned the species C. morio and V. darlingtoni, and also V. regulus in regrowth, a common constant annual survival rate estimated at 0.60 ±0.02 while for V. pumilus females in regrowth, annual survival was estimated to be 0.41 ±0.12. The annual survival rates for male V. pumilus, female V. pumilus in unlogged, and V. regulus in unlogged covaried inversely with MaxTempSumm, with the same regression coefficient, differing only in intercept (means ± SD were 0.46 ± 0.13, 0.30 ± 0.12, and 0.73 ± 0.03, respectively; see also Fig 3). Yet variation with MaxTempSumm was small for V. regulus in unlogged, and only substantial for V. pumilus, which experienced low survival throughout the study (typically < 0.5). Logging treatment was influential on just two cases of survival, being higher in unlogged than regrowth catchments for V. regulus (mean survival = ~0.73 vs 0.60), while the reverse was the case for V. pumilus females (mean = 0.30 vs ~0.41).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 3. Time varying survival parameters and their covariate (MaxTempSumm).

The value plotted over year n (n = 1999–2011) is survival over the year beginning with the capture event of year n to year n+1. VrU = V. regulus in unlogged; VpM = V. pumilus males; VpFU = V. pumilus females in unlogged. Note that survival values are averages over modelled estimates assuming covariation with the weather covariates.

https://doi.org/10.1371/journal.pone.0191471.g003

Mean life expectancy and longevity

Mean life expectancy for resident bats that were > 3 months in age at banding was just 1–3 years; C. morio = 2.0 years, V. darlingtoni = 2.0 years, V. regulus = 2.0–3.2 years and V. pumilus = 0.8–1.3 years. This is considerably less than the maximum longevity we recorded; C. morio = 8⁺ years, V. darlingtoni = 9⁺ years, V. regulus = 8⁺ years and V. pumilus = 6⁺ years. Although use of apparent survival will underestimate life expectancy (though we accounted for the effect of transients), the extent of teeth wear in the population provides supporting evidence for low mean life expectancy. Across all bats (excluding retraps), 85% showed no signs of teeth wear (index = 1) and just 1% of bats had very worn teeth (index = 3; 6% of retrapped bats), confirming that few bats, even those not banded, live to an old age.

Variation in abundance

The proportion of residents amongst newly marked bats was computed as an intermediate step in the estimation of the abundance of residents (Supplement). Combining the sexes, V. darlingtoni was the most abundant species at high and mid altitudes, with V. pumilus most abundant at low altitude (S3 Table). Elevation was negatively related to abundance of V. pumilus (r = -0.99; P<0.001) and positively related to the abundance of V. darlingtoni (r = 0.93; P<0.01) and V. regulus (r = 0.83; P<0.05), but not significant for C. morio abundance (r = 0.522; P = 0.29). As a consequence, V. pumilus was negatively associated with V. darlingtoni (r = -0.88) and V. regulus (r = -0.78). The abundance of V. darlingtoni and V. regulus were positively associated with each other (r = 0.95). These patterns occurred in both treatments. After adjusting for elevation, pre-planned contrasts found no significant difference in abundance averaged across 14 years between unlogged and regrowth catchments for any species: V. regulus (t = 0.41; P = 0.71), V. pumilus (t = 1.76, P = 0.12), V. darlingtoni (t = 0.06, P = 0.96) and C. morio (t = -0.30, P = 0.89).

We predicted that abundance would be relatively insensitive to weather extremes over time. The DFA indicated that the time series of abundances behaved randomly and similarly across sites and across species (see Supplement). We attribute the peak in abundance in the 4^th capture (Fig 4), early February 2000, to high numbers of juveniles (39% of captures in 2000) still present during the only capture event to take place so early in the year. Removing abundance values for February 2000 did not alter DFA patterns. One site (Kokata) showed a major dip in abundance in the 15^th sampling event (2010), while an adjacent catchment (Coachwood) showed a less distinct dip (Fig 4), coinciding with silvicultural thinning of the dense forest at Kokata in 2009–2010, including surrounding trap locations. Abundance recovered in the following year (Fig 4).

Download:

• PPT
  PowerPoint slide
• PNG
  larger image
• TIFF
  original image

Fig 4. Plots of estimated abundance (●) and number of residents (▲) over all four species per site with respect to capture event, numbered as in Fig 3.

https://doi.org/10.1371/journal.pone.0191471.g004

Body condition

Body condition for each species/sex fluctuated only slightly over time (CVs: 0.03–0.08). The DFA analysis detected no evidence for any trend with covariates and only the simplest random walk structure, which again is consistent with our prediction that the forest refuge buffered against weather extremes. Logging treatment did not influence the body condition of residents (P = 0.48), nor was there a significant logging by species interaction (P = 0.29) or logging by species by sex interaction (P = 0.22). Condition was not correlated with bat age (time elapsed since banding; r = 0.17).

Bat population dynamics

From the highly complex global model with time*species*sex*treatment dependence, we derived relatively simple models (at most 20 parameters) whose model-averaged parameters nevertheless portray a relatively complex picture of variation in survival. We caution that more data might paint an even more subtle picture of the dynamics of these four species than our data could reveal. For example, elevation could be an additional influence on survival that we chose not to model in preference to other factors given the number of sites sampled and bats caught per site. We purposely minimised trapping effort to minimise trap avoidance, yet still achieved a 32% retrap rate. It is also important to point out that we estimated apparent survival, as mortality and permanent emigration are confounded. However, by distinguishing between residents and transients, our modelling does not ascribe all disappearances of bats caught only once to apparent mortality of residents, which is therefore a less biased estimate of actual mortality than it would be otherwise. Transience is suspected to be greatest in autumn when juveniles disperse [17]. Our estimates of the proportion of residents (S1 Appendix and S1 Fig) was high (82–90%, with the exception of V. regulus in unlogged) among trap events within autumn, suggesting minor transiency over a 2–3 month period. These results are consistent with the fact that small vespertilionids typically have stable short-term foraging ranges [51], though these can shift locally in relation to patch dynamics of food [52]. Residency among years was also high, on average, but it varied considerably across the study (S1 Appendix and S1 Fig; see below). Just one retrap was recorded beyond the immediate surrounds of our study area, which was an 88 km straight line movement for F. tasmaniensis [53], but this was for the largest species that we banded and such movement is unlikely to be undertaken by smaller species.

Despite these uncertainties, our results are broadly consistent with previous studies on bat survival. First, our survival estimates averaged 0.60–0.73 per annum, except for the smallest species, V. pumilus, which averaged 0.30–0.41. These estimates (except V. pumilus) are relatively high compared to similar-sized mammals, but comparable to other bats, whose survival tends to vary between 0.6–0.9 per annum [3, 15, 17, 18, 20, 21, 22, 38]. Although we included 3–4 month old bats in our models, and it is well known that juveniles have elevated mortality and dispersal [3, 17, 21], we found a single, effect-independent survival parameter of 0.891 ±0.035 per two weeks of resident bats during the capture periods of 1999 and 2000. This survival rate is only applicable to the capture period and should not be extrapolated beyond it. An attrition rate of some 11% per two weeks would not be sustainable over many consecutive fortnights, but may reflect apparent mortality, including ongoing dispersal (even accounting for transients), of the cohort of sub-adults during the two extended autumn periods of capture. Pooling of sub-adults (often unrecognisable) with adults likely negatively biases apparent survival. We found no evidence for a sex effect on survival over years for the three larger species, but a small positive effect was present for males of the smallest species (V. pumilus). Differential survival between the sexes of bats has been found in some studies [15, 21], but not others [17].

Our smallest species, V. pumilus, presents the greatest anomaly in our results, but we suggest there are a number of explanations for lower survival at our study site. First, the species is small (~ 4 g) and is known to give birth to twins [54], both of which are traits related to lower survival [15]. Second, V. pumilus is a species of the coastal forests of NSW and the higher elevation of our study area is likely to be at the distributional limits of the species. Survival was lower for female V. pumilus, perhaps suggesting limited resources for breeding. Notably, it was the only species whose abundance was correlated negatively with elevation in our study area, which is also consistent with activity levels measured by acoustic detectors [25]. We suggest that our high elevation study area may be operating as a sink [8] at the edge of the range for V. pumilus, while negative associations with co-occurring species could imply additional negative effects from competition.

Population dynamics in disturbed forests

Long-term changes in population dynamics have been described for bird communities in relation to changes in vegetation structure in regenerating forest and responses tend to be species idiosyncratic, though survival estimates are rare [55, 56]. Results from our long-term bat banding found that survival and abundance showed minor differences at best between regrowth (varying in age over the study from 16–30 years) compared to unlogged catchments. There was no detectable effect of logging history on survival of two species (C. morio and V. darlingtoni), a positive effect for one (V. pumilus) and negative for the other (V. regulus), while there was no detectable effect of logging history on the abundance or body condition of any of these species. All four of the bat species we investigated can be classified as edge-space species with medium to high frequency echolocation calls that commonly avoid clutter, such as in dense regrowth [25]. Radio-tracking also supports our results in that V. pumilus will roost and breed in regrowth forests [51], while V. regulus preferentially roosts in patches of mature forest [57]. Acoustic detectors revealed that V. regulus was more active off tracks in unlogged forest than in regrowth [25], suggesting either that it is more sensitive to clutter than V. pumilus or it displaces the latter from these areas.

We suggest that survival was not consistently lower in cluttered regrowth forests because tracks provided open linear flyways where bats could forage efficiently. A number of studies have suggested that tracks facilitate the use of regrowth forest by many bat species that would otherwise be too cluttered for efficient foraging by echolocation [25, 27, 28, 29]. However, as tracks only represent ~ 3% of the landscape in timber production forests (e.g. [28]), local availability of unlogged forest is also likely to be important. Unlogged forest was retained within informal reserves along streams to protect this environment and as rainforest because this forest type is no longer logged in NSW [58]. Systematic landscape protection of unlogged forest is a key recommendation for ecologically sustainable management of timber production forests and is often referred to as multi-scale forest management [16, 59]. We note that our results pertain to areas of multi-scale forest management where ecological sustainability is a goal. Also, our conclusions focus on the most abundant local species rather than rare or threatened bat species, which we were unable to band in sufficient numbers for analysis.

The regrowth forest substantially self-thinned over the course of our study coincident with drought, even though rainfall was higher than areas outside of our study, with reduced stem density of both overstorey and understorey species being evident by 2011 [60]. This partly corresponds to trends of increased abundance over the study at two of the four regrowth sites where self-thinning was most evident (Corkwood and Kokata; authors pers. obs.). We also detected a dip in abundance at Kokata in 2010, coinciding with silvicultural thinning at that site. Recovery in the following year suggests this was a behavioural rather than a population effect. Such a pattern may represent lower trap success on trails during thinning, due to bats avoiding the disturbance.

Variation in population dynamics in relation to weather extremes

Can regenerating forest function as a climate refuge during climate extremes? A key result from our study was little variation in survival over the 14 years of banding, except for V. pumilus and to a smaller extent for V. regulus, our two smallest species. Survival rates for both of V. pumilus and V. regulus recovered as soon as weather conditions improved. Body condition also showed little variation over the study. This was despite the study spanning a severe El Niño period that resulted in the worst drought in recorded history for south-eastern Australia as well as spanning high rainfall years of La Niña periods. Not only were our species resilient to severe drought, variation in adult survival with extreme weather conditions has been suggested to be a major driver of population dynamics in other bat species and temperate passerines [49]. For example, survival of little brown bats Myotis lucifugus peaked in years with wet summers, suggesting drier summers associated with global climate change could pose future risks for the species [3], although the inverse was found for ghost bats Macroderma gigas in Australia [23]. European studies have either reported near constant survival in bats over years [18] or annual fluctuations only weakly related to weather [22], though they have not spanned extreme droughts. We suggest that the location of our study site in a climate refuge buffered survival, body condition and abundance from the effects of weather extremes. Annual rainfall at our wet site remained relatively high even in the driest year of 2002 (877 mm). These results are similar to tropical cloud forests where relatively constant capture rates of bats have been reported over a 27 year period [61]. Floodplains are another relatively moist environment that represented a climate refuge for birds during the Australian millennium drought, with an analysis of reporting rates rather than population dynamics finding that fewer species declined in floodplains (19%) than non-floodplains (29%) [62]. We predict that bat survival in Australia is likely to decline outside of such climate refugia during droughts, given canopy invertebrates decline during droughts [63], and collapses were observed in other fauna groups [23, 33, 34]. Climate change is forecasted to be widespread in the forests of eastern Australia [64] and forests may become increasingly vulnerable to tree mortality in response to future warming and drought [65].

There is some evidence that weather conditions had more subtle effects on the bat populations. Hot and dry spring-summer conditions immediately prior to autumn trapping were associated with increased capture (recapture) rate. One explanation is that insect abundance was maintained in the forest refuge over the drought and the warmer than average temperatures led to greater activity and a higher trap rate. Weather conditions during trapping were also important with cooler temperatures being associated with lower trap rates, noting that we avoided periods of heavy rain during trapping. In addition, transiency between years (one minus proportion of resident newly marked bats) was higher in the hot and drier years of 2002–2003 and 2009–2010 compared to cooler, wetter years, when transiency was low between years (S1 Appendix and S1 Fig). Male V. pumilus were an exception to this pattern, possibly indicating greater site fidelity. During drought, residents were possibly over-whelmed by intruders from nearby areas that experienced harsher conditions than at our buffered site, rather than residents leaving in search of better conditions (given survival was mostly unaffected by weather).

As previously noted, our autumn bat trapping prevented us from reliably discriminating juveniles from adults, which prevented estimation of recruitment. Variable weather patterns are known to affect recruitment in bats. In North America, bat reproduction declines during hotter and drier periods, especially during lactation, with more than 50% of females being non-reproductive during the worst drought years [66]. In European bats, juvenile survival (recruitment) across years is relatively constant, though data on pre-weaning mortality is lacking and extreme drought has not been sampled [17, 22]. The population dynamics of other temperate bats are driven by variation in birth-timing, which is delayed in very cold and wet spring and summer weather [67, 68]. We suggest recruitment likely varied in our climate refuge given small fluctuations in abundance across years and mostly constant adult survival rates. Mortality would be most likely post-parturition as we have found > 90% of females are typically pregnant based on trapping in spring at a range of sites and years throughout NSW (see also [3]).

Conclusion

We found that apparent survival was not strongly influenced by extreme weather patterns, except for the smallest species, nor was abundance or body condition. These results are consistent with the notion that our study area was located in a climate refuge, buffering bat population dynamics from weather extremes. The relative resilience of study populations to the ambient weather pattern had no detectable interaction with logging history, further suggesting that the apparent climate refuge effect was not compromised by the past logging activity [10]. This differs from limited post-drought recovery observed for birds in wetlands, possibly because of very large-scale disturbances to the system, especially altered flooding regimes [63]. Continued refinement of management practices to improve ecological sustainability, following an adaptive management approach, will be fundamental for managing competing needs in identified refugia [69].