Secondary forest buffers the effects of fragmentation on aerial insectivorous bat species following 30 years of passive forest restoration

Passive forest restoration can buffer the effects of habitat loss on biodiversity. We acoustically surveyed aerial insectivorous bats in a whole‐ecosystem fragmentation experiment in the Brazilian Amazon over a 2‐year period, across 33 sites, comprising continuous old‐growth forest, remnant fragments, and regenerating secondary forest matrix. We analyzed the activity of 10 species/sonotypes to investigate occupancy across habitat types and responses to fragment size and interior‐edge‐matrix (IEM) disturbance gradients. Employing a multiscale approach, we investigated guild (edge foragers, forest specialists, flexible forest foragers, and open space specialists) and species‐level responses to vegetation structure and forest cover, edge, and patch density across six spatial scales (0.5–3 km). We found species‐specific habitat occupancy patterns and nuanced responses to fragment size and the IEM disturbance gradient. For example, Furipterus horrens had lower activity in secondary forest sites and the interior and edge of the smallest fragments (1 and 10 ha) compared to continuous forest, and only two species (Pteronotus spp.) showed no habitat preference and no significant responses across the IEM and fragment size gradients. Only the Molossus sonotype responded negatively to vegetation structure. We uncovered no negative influence of forest cover or edge density at guild or species‐level. Our results indicate that reforestation can buffer the negative effects of fragmentation and although these effects can still be detected in some species, generally aerial insectivorous bats appear to be in recovery after 30 years of passive forest restoration. Our findings reinforce the need to protect regenerating forests while conserving vast expanses of old‐growth forest.


Introduction
Representing 45% of all forest cover on Earth (FAO & UNEP 2020) tropical forests are some of the most diverse ecosystems (ForestPlots.net et al. 2021), yet they are under constant threat due to a wide array of anthropogenic pressures (Malhi et al. 2014).Land use change continues to drive global deforestation, with agricultural expansion persisting as the major reason for forest clearance, accounting for at least half of all forest loss-circa 10 million ha/year (FAO & UNEP 2020; Laso Bayas et al. 2022).Human-modified landscapes are increasingly fragmented (Taubert et al. 2018) and the twin devils of habitat loss and fragmentation are eroding biodiversity by contracting species' ranges, causing widespread changes in species abundance and affecting species interactions, which is notably prominent in the species-rich tropical regions (Ceballos et al. 2017).
The Amazon is a deforestation hotspot and while scientific efforts continue to advocate for the significant and permanent reduction of forest clearance, regenerating secondary forests have been identified as the most effective method of passive restoration of the Amazon (Crouzeilles et al. 2017).The majority of Amazonian land use change resulted from the conversion of native forest into pasture for cattle and agricultural expansion (Silveira et al. 2022).However crop fields and pastures are often abandoned, allowing reforestation via successional processes (Chazdon 2014).Reforestation of abandoned lands has the capacity to support displaced forest biodiversity, contribute to carbon sequestration (Heinrich et al. 2023) and provide climate change mitigation (Locatelli et al. 2015).Secondary forests are a significant feature in the Amazon, covering greater than 234,000 km 2 by 2017 (Smith et al. 2021).These landscapes often contain remnant fragments, embedded in a matrix ranging from high-contrast agricultural to low-contrast regrowth forest (Neeff et al. 2006;Numata & Cochrane 2012).Where regenerating forests mature, the negative effects of fragmentation can be buffered, providing valuable habitats which can support a wide variety of taxa (Edwards et al. 2017;Derhé et al. 2018;De Aquino et al. 2022).
Aerial insectivorous bats provide essential ecosystem services and are useful bioindicators of ecosystem health (Jones et al. 2009;Kunz et al. 2011).In Brazil, they account for nearly half (48%) of the greater than 181 described bat species (Garbino et al. 2022).They are best sampled with acoustic methods, as their highly developed echolocation means that, unlike their phyllostomid counterparts, they are rarely captured in mist nets (Appel et al. 2021a;Carvalho et al. 2023).The sensitivity of aerial insectivorous bats to forest fragmentation is largely mediated by their wing morphology and echolocation (Bader et al. 2015;Colombo et al. 2023) which can determine foraging success within a habitat based on the level of clutter (Schnitzler & Kalko 2001).Forest specialists are often more vulnerable to fragmentation (Núñez et al. 2019;L opez-Bosch et al. 2022) as they are adapted to navigate and forage in highly cluttered space, whereas aerial-hawking species use uncluttered open space above the canopy and can fly longer distances to access suitable foraging habitat, making them more resilient to fragmentation (Bader et al. 2015).
Vulnerability of neotropical bats to fragmentation has been widely investigated at the Biological Dynamics of Forest Fragments Project (BDFFP) in the Brazilian Amazon, the world's largest and longest-running experimental study of habitat fragmentation and forest regeneration.While a significant proportion of this research has focused on the responses of phyllostomid bats (e.g.Farneda et al. 2018Farneda et al. , 2022;;Rocha et al. 2018), recent research has also targeted the ensemble of aerial insectivorous bats (e.g.Appel et al. 2021b;L opez-Baucells et al. 2022;Yoh et al. 2022), which comprises a considerable fraction of the BDFFP bat fauna.Multi-dimensional (taxonomic, functional, and phylogenetic) diversity responses of the aerial insectivorous bat ensemble to vegetation and landscape structure indicated subtle scalesensitive associations for functional diversity only (L opez-Baucells et al. 2022).The regenerating secondary forest matrix around the smallest fragments was found to accommodate significantly lower diversity compared to the continuous forest.This biodiversity degradation became less pronounced with increasing fragment size.In contrast, forest edges generally harbored higher taxonomic, functional, and phylogenetic diversity compared to continuous forest interiors.Recent research investigating speciesspecific responses of aerial insectivores to edge effects across the interface of continuous and secondary forest suggested that the regenerating secondary forest might have recovered enough that edge effects are no longer evident for most common species (Yoh et al. 2022).In contrast, Appel et al. (2021b) found lower activity in the regenerating secondary forest suggesting that even after 30 years it remains less attractive as a foraging habitat.
Here, we build on our previous multi-dimensional diversity analysis (L opez-Baucells et al. 2022)  We hypothesized that the 30 years of secondary forest regeneration would buffer the negative effects of fragmentation on the focal species.Specifically, we predicted that; (1) occupancy and activity would be species-specific, with similar responses within guilds, (2) FS would have higher occupancy in CF due to functional trait exclusion, (3) there would be a positive response for EF exploiting feeding opportunities generated around the fragment edges and secondary forest matrix sites, and (4) FF would show no response or preference for any habitat.

Study Area
The study was conducted at the BDFFP, an experimental fragmentation project located approximately 80 km north of Manaus, Central Amazon, Brazil (Fig. S1).The topography of the study area (approximately 1000 km 2 ) is relatively flat, with streamlets cutting through nutrient-poor soil (Laurance et al. 2011).Annual rainfall varies from 1900 to 3500 mm, with a dry season between July and November (Ferreira et al. 2017).The BDFFP began in 1979 when ranchers cleared continuous old-growth terra firme rainforest to accommodate cattle pastures and preserved forest patches of varying sizes (1, 10, and 100 ha) isolated from the CF by distances of 80-650 m (Bierregaard Jr et al. 1992).Within 10 years the cattle ranches were abandoned and secondary forest established in the matrix around the fragments (Laurance et al. 2018).Approximately every 10 years, to preserve isolation, a 100 m-wide strip of secondary forest surrounding the fragments is cleared (Rocha et al. 2020).Canopy height of fragment forest averages 30 and 37 m, with emergent trees exceeding 55 m.The most recent re-isolation of the fragments prior to data collection for this study occurred between 1999 and 2001 (Laurance et al. 2011).There was relatively little structural contrast between CF, fragments, and the advanced-stage (approximately 30 years old) secondary forest at the time of this study.

Acoustic Recordings
Acoustic surveys were conducted between November 2011 and September 2013 across a total of 33 sampling sites: (a) nine control sites in three CF areas (located in Cabo Frio, Florestal, and Km41 camps), (b) eight fragment sites (3 Â 1, 3 Â 10, and 2 Â 100 ha, located in Colosso, Porto Alegre, and Dimona camps) with sampling conducted in the interiors and at the edges of all eight fragments (interior-edge distances 245 AE 208 m [mean AE SD]), and (c) eight secondary forest sites located 100 m into the matrix from the nearest fragment edge.
Bats were acoustically sampled with Song Meter 2 (SM2Bat+) recorders fitted with omnidirectional ultrasonic SMX US microphones (Wildlife Acoustics Inc., Maynard, MA, U.S.A.).Recorders were positioned at 1.5 m height and programmed to record for 12 hours (from 18:00 to 6:00 hours).Recordings were captured in real time at full-spectrum resolution (16-bit), sampling rate of 384 kHz with a high pass filter set at fs/32 (12 kHz) and adaptive trigger level relative to noise floor of 18 signal-to-noise ratio.Each site was surveyed for five consecutive nights, four times annually, with two surveys in the wet season and two in the dry season, amounting to a total of 1021 recording nights.Sampling of the interior, edge, and matrix sites of the same forest fragment was temporally spaced as far apart as logistically feasible.

Bioacoustic Analysis
Recordings were manually analyzed using Kaleidoscope Pro v4.0.4 software (Wildlife Acoustics Inc.) following the acoustic key in L opez-Baucells et al. (2016).Calls were identified to species level whenever possible or sonotypes (groups of taxa with similar calls).It is not possible to count individuals using bioacoustic data, thus activity was used as a proxy for abundance (Rowse et al. 2016) based on bat passes.A bat pass was defined as any call sequence containing at least two distinguishable echolocation pulses during a maximum duration of 5 seconds (L opez-Baucells et al. 2021).Bat activity was quantified by the number of bat passes per night per species/sonotype.We identified 283,242 bat passes belonging to 17 aerial insectivorous bat species/sonotypes, however we only included species that were detected in at least 10% (102 nights) of total nights and in all three sampling years to minimize potential detection bias as per Appel et al. (2021b).This resulted in the selection of 272,201 calls from nine species and one sonotype; Centronycteris maximiliani, Cormura brevirostris, Saccopteryx bilineata, S. leptura, Furipterus horrens, Molossus sonotype (including currentium/rufus calls), Pteronotus alitonus, P. rubiginosus, Eptesicus brasiliensis, and Myotis riparius (Table S1).Following Yoh et al. (2022), we assigned species/ sonotypes to guilds based on ecological requirements and foraging strategies: EF-C.brevirostris, C. maximiliani, S. bilineata, S. leptura; FS-E.brasiliensis, F. horrens, M. riparius; FFboth Pteronotus species; OS-Molossus sonotype (included in the guild analysis for comparison only as represented by a single sonotype).S2; Fig. S2), and was positively correlated with average DBH of trees greater than 10 cm, canopy height and percentage canopy cover, and scores were retained as a predictor variable summarizing vegetation structure (PCA1).

Local Vegetation Structure and Landscape Predictor Variables
Methods for calculating landscape structure metrics followed L opez-Baucells et al. (2022).In brief, we used 2011 LandSat Thematic Mapper satellite images categorized into two land cover classes; CF and secondary forest.To avoid collinearity (a common problem with landscape predictor variables) and to allow for comparison with previous studies (L opez-Baucells et al. 2022), we selected the same three landscape metrics which were considered acceptable based on variance inflation factor calculations (Rocha et al. 2017).R package "landscapemetrics" (Hesselbarth et al. 2019) was used to calculate composition (CF cover) and configuration (patch density and edge density) metrics within circular buffers with radii of 0.5, 1, 1.5, 2, 2.5, and 3 km around 33 sampling sites.Little is known about the home range of the focal species in this study, we therefore took into consideration the scale of effect documented by Jackson and Fahrig (2015) and the utilization of 3 km landscape buffers in other tropical bat studies (e.g.Farneda et al. 2020;Martínez-Ferreira et al. 2020).Spatial scales greater than 3 km were not investigated to minimize spatial overlap between sites (Meyer & Kalko 2008) although it is acknowledged that landscape buffers which overlap may not necessarily violate statistical independence (Zuckerberg et al. 2020).To control for statistical correlation (Trzcinski et al. 1999), we fitted a linear model between the composition variable (CF cover) and configuration variables (edge and patch density) at each spatial scale and extracted the residuals creating new configuration variables (Bélisle et al. 2001;Klingbeil & Willig 2010).

Statistical Analysis
To visualize species-specific frequency of occurrence, activity per habitat type, expressed as percentage, was plotted (Fig. 1).We modeled each of the species' bat activity with (1) the fragment size/IEM variables and (2) the vegetation structure and landscape variables using Monte Carlo Markov Chain generalized linear mixed models via the R package MCMCglmm (Hadfield 2010), specifying a Poisson error distribution, and priors were set to be weakly informative in order to deal with overdispersion (inverse gamma parameters: ν = 0.002 and Ѵ = 1; Kryvokhyzha et al. 2016;Hadfield 2018).For fragment size/IEM variables, we fit a full model for each species, specifying a single categorical fixed effect with combined information on the fragment size and IEM gradient (10 categories: CF interior, 100 ha interior, 100 ha edge, 100 ha matrix, 10 ha interior, 10 ha edge, 10 ha matrix, 1 ha interior, 1 ha edge, 1 ha matrix; L opez-Baucells et al. 2022) and incorporated research camp location as a random effect.To model vegetation structure and landscape variables, we fit a set of models, each set contained 10 full models for each species with four fixed-effect variables (PCA1 and the new residual variables; CF cover, edge density, patch density) for each buffer size (0.5, 1, 1.5, 2, 2.5, and 3 km) and research camp location as a random effect.We also modeled vegetation and landscape variables (as previously detailed) for each guild (FS, ES, and FF).We ran 50,000 iterations, using a thinning-interval of 10 after "burn-in" of 5000.
We evaluated convergence through (1) trace plot visual checks (2) lag k autocorrelation statistic, and (3) Gelman-Rubin diagnostic statistic (Gelman & Rubin 1992).All models achieved good convergence (potential scale reduction factor <1.1).Posterior distributions for the predictor variables were obtained.As model output, we report posterior means, 95% credible intervals and Bayesian p values (p MCMC ) indicating the significance of variables with a threshold of *p MCMC less than 0.05, **p MCMC less than 0.01, ***p MCMC less than 0.001 (Sweeny et al. 2021).

Results
Across 33 sites, 272,201 bat passes from 10 species/sonotypes and 5 families encompassing four guilds were analyzed.All species/sonotypes were recorded across all four habitats, however patterns of occupancy were highly species-specific (Fig. 1).
Eptesicus brasiliensis and Furipterus horrens had occurrence frequencies greater than 50% in CF.There were only three species with greater than 20% occurrence recorded in the secondary forest (Pteronotus alitonus, P. rubiginosus, and Molossus sonotype).Cormura brevirostris had over 75% of its frequency of occurrence in fragment interiors.Across the IEM and fragment size gradients, we observed species-specific differences in activity (Table S3; Fig. 2), with most significant responses seen in the secondary matrix, particularly around the 1 and 10 ha fragments.Significant negative responses were observed in relation to the 100 ha matrix (Centronycteris maximiliani, Saccopteryx bilineata, F. horrens), the 10 ha matrix (C.maximiliani, F. horrens, E. brasiliensis) and the 1 ha matrix (C.maximiliani, S. bilineata, S. leptura, F. horrens, E. brasiliensis) as species reduced activity in these habitats.Five species showed no significant negative responses to either the IEM or fragment size gradients (C.brevirostris, P. alitonus, P. rubiginosus, Myotis riparius, Molossus sonotype) and two of these species (M.riparius and Molossus sonotype) evinced a positive response to the matrix habitat even around the smallest 1 ha fragment.
In particular, C. maximiliani showed a clear pattern with significantly reduced activity in all secondary forest sites and reduced activity at the edges of the 1 ha fragments.The activity of S. bilineata was significantly lower in the matrix around the smallest fragments and in matrix and edge sites of the 100 ha fragments.S. leptura activity was significantly higher in the Figure 2. Comparison of total bat activity for species/guilds across the IEM and size gradients at the Biological Dynamics of Forest Fragments Project (forest fragment interiors, forest fragment edges, and adjoining secondary forest matrix).The predicted differences between each habitat and continuous old-growth forest interior, modeled using MCMCGLMM are plotted with their corresponding 95% CI.Those which do not touch or overlap the vertical dashed line (0) are considered significant (*p MCMC < 0.05, **p MCMC < 0.01, ***p MCMC < 0.001).
interiors of the 1 and 10 ha fragments whereas it was reduced in the matrix adjacent to the 1 ha fragments.F. horrens had significantly lower activity in all matrix sites and the interior and edge sites of the smallest fragments (1 and 10 ha).M. riparius activity responded positively to forest edges, irrespective of fragment size.Neither of the Pteronotus species showed any significant responses to the habitat gradient or fragment size.Molossus sonotype activity was significantly elevated across all matrix and edge sites for all fragment sizes.

Influence of Vegetation Structure and Landscape Predictors
Modeling showed species-specific variation in response to vegetation structure and landscape predictors (Table S4), with seven species (C.maximiliani, C. brevirostris, S. bilineata, S. leptura, F. horrens, Molossus sonotype, and E. brasiliensis) exhibiting significant scale-sensitive associations with the metrics (Fig. 3).Species which showed no significant association with any of the vegetation structure or landscape variables, at any of the scales investigated, included P. alitonus, P. rubiginosus, and M. riparius (Fig. S3).
Vegetation structure was an important predictor of bat activity.C. maximiliani, S. bilineata, and S. leptura, exhibited a significant positive relationship at every scale.F. horrens and C. brevirostris evinced a consistent positive relationship at the larger scales (2.5-3 km).E. brasiliensis responded positively and exclusively at the 2 km scale to vegetation structure.In contrast, the Molossus sonotype exhibited a significant negative relationship at every scale.
Regarding landscape variables, forest cover and patch density were irrelevant predictors, with very little influence on bat activity and no significant associations at any scale.The configurational metric, edge density, exhibited a positive association with C. maximiliani activity at the 1 and 1.5 km scale.
The guild-level modeling results (Table S5; Fig. S4) further highlighted the influence of vegetation structure as FS revealed a positive relationship at scales 2.5 and 3 km.It was also a significant determinant of activity for EF at all scales, and highly so at the 0.5, 1, and 2.5 km scale.The configurational metrics, edge, and patch density, also influenced the EF with a negative relationship with patch density at 0.5 km and positive relationship with edge density at 1 km.

Discussion Habitat Occupancy and Activity Responses to IEM and Fragment Size Gradients
Emballonurids produce shallow-modulated and multi-harmonic echolocation calls, which allow them to forage across a range of habitats including forest edge vegetation, forest gaps, and above canopy and over open landscapes (Jung et al. 2007).The four emballonurids in this study are categorized as EF although their frequency of occurrence suggests a broader habitat preference within the landscape.Our results concur and highlight the niche differentiation outlined by Denzinger et al. (2018) with Saccopteryx bilineata, known to forage above shrub layer, in small forest gaps and in the canopy whereas S. leptura can often be found foraging in subcanopy and canopy.Our results also show that Cormura brevirostris had the highest frequency of occurrence recorded in the forest fragment interiors compared to all other species and is the only edge-foraging species which showed no significant negative activity response across the IEM and fragment size gradients.A study comparing the echolocation calls of S. bilineata and C. brevirostris found that C. brevirostris was better able to adapt its calls in the flight chamber whereas S. bilineata rarely changed its call structure (Jakobsen et al. 2012).This study suggested two plausible possibilities; that C. brevirostris has more call frequency composition control and acoustic dexterity, compared to S. bilineata, or conversely that S. bilineata is more clutter-tolerant and does not need to make changes to its echolocation to navigate such habitats.This suggests that C. brevirostris has more call frequency composition control and acoustic dexterity allowing uninhibited foraging across the BDFFP landscape.
Building on conclusions drawn by Yoh et al. (2022), our results show that Furipterus horrens appears to be more sensitive to the restored secondary forest as activity was negatively affected across the whole IEM gradient and all bar the largest fragment interiors.Compared to other neotropical bats, F. horrens emits extremely high frequency, low intensity pulses which allows it to forage in dense rainforest vegetation (Falcão et al. 2015).However, this may result in F. horrens relying on old-growth forest due to trait-mediated exclusion from secondary forest.Eptesicus brasiliensis exhibited a similar frequency of occupancy to F. horrens but appears to be able to take advantage of a wider variety of habitats.Echolocation calls of E. brasiliensis are similar to its North American congener Eptesicus furinalis (Arias-Aguilar et al. 2018).Studies highlight examples of E. furinalis navigating through acoustically challenging environments and complicated scenes (Simmons et al. 2018(Simmons et al. , 2020)).This acoustic dexterity trait is likely to be shared with E. brasiliensis, allowing it to utilize habitat created by fragmentation.For Myotis riparius, activity was consistently and significantly increased across all edge habitats, independent of fragment size.M. riparius shows significant plasticity in habitat usage (Novaes et al. 2017) and is often recorded at ground level, rather than subcanopy or canopy level (Marques et al. 2016).Due to M. riparius' small body size (Norberg & Rayner 1987) and echolocation call structure (Schnitzler & Kalko 2001) it appears to be taking advantage of a foraging niche in the lowest forest strata which in the case of this study is also extending to edge habitat.
Our results for Pteronotus alitonus and P. rubiginosus showed similar activity across all four habitat types.This absence of habitat preference was also noted in the IEM and fragment size gradient comparison where no significant effects were found.These results are in line with another study conducted at the BDFFP during the same time.Appel et al. (2021b) found elevated activity levels in secondary forest compared to continuous forest, however P. alitonus and P. rubiginosus exhibited foraging plasticity with increased activity in continuous forest on bright moonlit nights, possibly to reduce threats from avian predators.This flexibility was only evident in Pteronotus species which produce high duty cycle constant-frequency echolocation calls, allowing them to accurately detect small mobile prey items in dense vegetation (Estrada-Villegas et al. 2012;de Oliveira et al. 2015;L opez-Baucells et al. 2018).They are well adapted to forage in regenerating forests where understory vegetation may be heavily cluttered.A later study conducted at the BDFFP, sampled aerial bat activity following fragment re-isolation in 2013 and found that Pteronotus spp.were more abundant in forest interiors than in the newly cleared areas or at fragment edges (Núñez et al. 2019).This flexibility across habitat types and response Figure 3. Species exhibiting significant scale-sensitive associations with metrics; EF Â 4; Centronycteris maximiliani (CM), Cormura brevirostris (CB), Saccopteryx bilineata (SB), S. leptura (SL), FS x 2; Furipterus horrens (FH), Eptesicus brasiliensis (EB), and OS Â 1; Molossus sonotype (M2).Bat activity modeled as a function of vegetation structure and landscape variables; vegetation structure (LVS), forest cover (FC), edge density (ED), and patch density (PD).Shown are posterior mean estimates 95% CI.Credible intervals which do not touch or overlap the zero line are considered significant (*p MCMC < 0.05, **p MCMC < 0.01, ***p MCMC < 0.001).

Bat fragmentation responses in restored landscapes
to biotic and abiotic variables illustrates how well adapted these mormoopid species are, occupying a relatively free acoustic niche (Siemers & Schnitzler 2004).
Molossids are associated with fast and economic flight (Castillo-Figueroa 2020).Their mobility is constrained within forest strata consisting of forest gaps and obstacles to navigate and so they are often precluded from foraging in these habitats due to the high energetic cost (Voigt & Holderied 2012).Our results found the Molossus sonotype most associated with fragment edges and secondary forest showing significantly elevated activity in these habitats.Bats of the Molossidae family generally show good tolerance to fragmentation (Estrada-Villegas et al. 2010) and are not overly sensitive to anthropogenic disturbance (Kemp et al. 2019;Meramo et al. 2022).We did not expect to see such bias in occupancy toward regenerating habitat.Molossids generally forage high above the forest canopy in open space (Surlykke & Kalko 2008) which can make detecting their calls difficult if they are a significant distance away from the detectors (Marques et al. 2016).The lower canopy height of the recovering secondary forest and the forest edges may allow molossids to fly lower and for their calls to be within recording range of detectors more often in disturbed forests compared to CF.We highlight that our results could be due to increased detectability from changes in flight behavior across habitat types.

Influence of Vegetation Structure and Landscape Predictors
Forest cover had no significant influence on the activity for any of the species or guilds.Inconsistent results have been reported for the effect of forest cover on aerial insectivorous bat activity (Azofeifa et al. 2019;Rodríguez-San Pedro et al. 2019;Colombo et al. 2023).We also found no negative responses associated with the configurational metrics (edge density and patch density).Aerial insectivorous bats, particularly those adapted to foraging around patch edges, often benefit from increased hunting opportunities and have shown resilience, per se, to fragmentation (Ethier & Fahrig 2011;Rodríguez-San Pedro & Simonetti 2015).Our results also concur with a study on edge effects at the BDFFP which provided evidence that the regenerated secondary forest can support comparable levels of aerial insectivorous bat activity relative to continuous oldgrowth forest as a result of edge sealing (Yoh et al. 2022).The Molossus sonotype consistently showed a significant negative response to vegetation structure across all scales.These open space foragers will find it acoustically and physically challenging to navigate through forested habitat (Voigt & Holderied 2012) and often have reduced activity in mature forest where, on account of their increased mobility and fast flight speed, they can commute longer distances and access anthropogenically altered habitats to take advantage of increased roost and food availability created by human settlements (Bader et al. 2015;Kemp et al. 2019).Again, we acknowledge that detectability of the Molossus sonotype may have affected our results.
The Importance of Forest Restoration for Amazonian Aerial Insectivorous Bats Cattle ranching is the main driver of deforestation in the Amazon (Skidmore et al. 2021), yet farms, akin to the situation at the BDFFP, often become unproductive soon after establishment, resulting in large expanses of abandoned pastureland (Kaimowitz & Angelsen 2008).Restoring forests and recovering degraded pasturelands are key strategies by the Brazilian government to fulfill its ambition to restore 12 million ha of forest by 2030 (Feltran-Barbieri & Féres 2021).Recent studies have estimated that approximately 40% of deforested areas in the Brazilian Amazon show high potential for passive restoration (Vieira et al. 2017) and natural regeneration is also considered the most cost-effective approach to forest restoration (Crouzeilles et al. 2019).The responses of the focal bat species/sonotypes reported here suggest that passive forest restoration can buffer the effects of fragmentation at landscape scale and provide sufficiently suitable habitat to maintain an ensemble of common aerial insectivorous bat species at the BDFFP if enough recovery time is afforded (i.e. at least 30 years).We found evidence of species-specific responses, including higher occupancy patterns for FS in the CF, highlighting there may still be functional trait exclusion based on the recovering habitat structure.We show that many aerial insectivorous bats, those adapted to forage around edge habitat, in open space or with flexible foraging requirements, can successfully exploit habitat features within the regenerating forest landscape.Our results also show limited negative effects of fragmentation across this restored forest at landscape scale.However, while the restored forest may present viable habitat for many of the common species this might not be the case for all species.Our sampling method made it difficult to detect the uncommon species which might be more sensitive to the effects of fragmentation and could require a longer period of time and/or more active restoration approaches to facilitate full recovery.In summary, especially in landscapes with high levels of forest cover such as the BDFFP, passive restoration of degraded pastures, when coupled with the long-term protection of the resulting secondary forest, is an important strategy to maximizing aerial insectivorous bat species richness in human-modified Amazonian landscapes.

For
each of the 33 sampling sites, within three plots of 100 m 2 (5 Â 20 m) around the detector locations (during the same period as the acoustic sampling) we quantified seven vegetation variables (L opez-Baucells et al. 2022): (1) number of trees (>10 cm diameter at breast height [DBH]); (2) number of woody stems (<10 cm DBH); (3) average DBH of trees (>10 cm); (4) percentage canopy cover; (5) liana density (visually classified every 5 m in five categories varying from no lianas to very high liana density); (6) canopy height (based on visual estimation); and (7) vertical stratification in vegetation density.)To reduce the dimensionality of the data, we performed a principal components analysis (PCA) (L opez-Baucells et al. 2022).The first axis explained 38.3% of total variance (Table

Figure 1 .
Figure1.Species sorted by their frequency of occurrence in CF interiors, forest fragment interiors, forest fragment edges, and secondary forest.Percentage of occurrence was calculated using bat activity in each site per habitat type.Differences in sampling effort between habitat categories were accounted for in the percentage calculations.