Foraging behaviour of a frugivorous bat helps bridge landscape connectivity and ecological processes in a fragmented rainforest

Authors

  • MICKAËL HENRY,

    1. Département Ecologie et Gestion de la Biodiversité, MNHN/CNRS, UMR 5176, 4, av. du petit château, F-91800 Brunoy, France; *Origine, structure et évolution de la biodiversité (UMR 5202), Département Systématique et Evolution, Muséum National d’Histoire Naturelle, Case Postale 51, 55 rue Buffon, F-75005 Paris, France; and †INRA-EFPA, UMR Centre de Biologie et de Gestion des Populations (CBGP), Campus International de Baillarguet, CS 30016, F-34988 Montferrier/Lez cedex, France
    Search for more papers by this author
  • JEAN-MARC PONS,

    1. Département Ecologie et Gestion de la Biodiversité, MNHN/CNRS, UMR 5176, 4, av. du petit château, F-91800 Brunoy, France; *Origine, structure et évolution de la biodiversité (UMR 5202), Département Systématique et Evolution, Muséum National d’Histoire Naturelle, Case Postale 51, 55 rue Buffon, F-75005 Paris, France; and †INRA-EFPA, UMR Centre de Biologie et de Gestion des Populations (CBGP), Campus International de Baillarguet, CS 30016, F-34988 Montferrier/Lez cedex, France
    Search for more papers by this author
  • and * JEAN-FRANÇOIS COSSON

    1. Département Ecologie et Gestion de la Biodiversité, MNHN/CNRS, UMR 5176, 4, av. du petit château, F-91800 Brunoy, France; *Origine, structure et évolution de la biodiversité (UMR 5202), Département Systématique et Evolution, Muséum National d’Histoire Naturelle, Case Postale 51, 55 rue Buffon, F-75005 Paris, France; and †INRA-EFPA, UMR Centre de Biologie et de Gestion des Populations (CBGP), Campus International de Baillarguet, CS 30016, F-34988 Montferrier/Lez cedex, France
    Search for more papers by this author

M. Henry, Centro de Investigaciones en Ecosistemas, Universidad National Autónoma de México, Apartado Postal 27–3 (Xangari), Morelia, Michoacán, Mexico. E-mail: mickael.henry@ac-rennes.fr

Summary

  • 1Landscape connectivity may greatly influence the distribution of animals when it alters their movements and their ability to reach food patches. Depending on their foraging behaviour, organisms may or may not adapt to anthropogenic changes in landscape connectivity and may eventually undergo local extinctions.
  • 2Recent studies underlined the need to use indicators of functional landscape connectivity based on the behaviour and movement abilities of studied animals to better link landscape structure to ecological processes in disturbed and fragmented areas.
  • 3The objectives of this study were: to elaborate an index of functional connectivity for Rhinophylla pumilio, a Neotropical understorey frugivorous bat; to use this index to investigate the possible mechanisms controlling its distribution and sustainability in a fragmented landscape; and to test whether this index could be applied to other species of the same guild.
  • 4We pursued a 10-year bat mist-net survey, coupled to local estimates of food availability, in a mature forest of French Guiana that was recently fragmented by the completion of a reservoir lake. The 18 sampling sites range from undisturbed continuous forest sites to small remote forest fragments. A connectivity value, based on radio-tracking surveys, was attributed to each site. Connectivity measures mean forest cover within neighbouring landscape units, weighted by the probability that bats would use them, as estimated by frequency distribution of flight distance data.
  • 5The abundance of R. pumilio was positively correlated with landscape connectivity and not correlated with local food availability. Its foraging strategy has evolved in response to the highly scattered distribution of its fruit resource. In spite of its high mobility, R. pumilio apparently failed to exploit a food resource that is distributed patchily over a low-connective habitat because its foraging movements are not well adapted to habitat disruptions.
  • 6The connectivity index contributed to explain general tendencies of abundance variations in other understorey frugivorous bats, although the spatial scale we examined was probably too small for these species. We make recommendations to adapt a functional connectivity index to species whose large-scale movements are difficult to survey.

Introduction

One of the most crucial concerns of tropical ecologists is to assess the impact of habitat loss and fragmentation on species diversity, as well as the underlying ecological processes leading to local species extinctions (Stouffer & Bierregaard 1995a,b; Laurance & Bierregaard 1997; Gascon et al. 1999; Laurance et al. 2002; Ewers & Didham 2006). Habitat fragmentation may affect animal species’ densities through a variety of direct and indirect mechanisms, such as pure habitat loss, increasing edge effect and associated microclimatic changes on small fragments, isolation of habitat fragments, or greater extinction risks of small populations (Saunders, Hobbs & Margules 1991; Andrén 1994). To satisfy the needs of both theoretical and applied conservation biology, various hypotheses have been proposed to explain the density and distribution of animal species within fragmented landscapes. Early hypotheses were based on the equilibrium theory of island biogeography (MacArthur & Wilson 1967), stating that as in true islands, the occurrence of species in habitat fragments is a function of fragment area and isolation distance from the mainland (Simberloff 1988).

However, the theory of island biogeography has its own limits, and the ‘island metaphor’, consisting in treating habitat fragments as islands embedded in an inhospitable matrix, gives imperfect insights about the distribution of organisms in fragmented areas. Border shape of both fragments and mainland may also greatly influence colonization and dispersal rates, independently from fragment size or isolation distance (Taylor 1987a,b). Moreover, the occurrence of some species in fragments may depend more on dispersal movements from other fragments rather than from the mainland (Fahrig & Merriam 1994), making the notion of isolation distance difficult to apply. Conversely, some mainland areas bordering the inhospitable matrix are likely to undergo noticeable faunal changes as well, due to habitat disruptions, edge effect, or invading species. Thus, whenever possible, it is important to consider ecological processes along a continuous gradient of landscape disturbance rather than in a patch-based approach. Finally, the assumption that the matrix surrounding fragments is inhospitable regarding the focal species is often violated (Norton, Hannon & Schmiegelow 2000; Ricketts 2001; Brotons, Mönkkönen & Martin 2003; Kupfer, Malanson & Franklin 2006). For instance, forest birds may find valuable food resources in pastures or logged areas surrounding their natural habitat, leading to greater population densities than expected in fragmented areas (Brotons, Herrando & Martin 2004; Brotons et al. 2005). Therefore, it is of special interest to control for possible mitigating effects related to matrix quality.

The recent discipline of landscape ecology, using GIS-based analyses of landscape structure, offers solutions to side-step these limitations. Landscape analyses can provide integrative measurements of spatial heterogeneity, combining many environmental variables and their spatial interactions. Among the descriptors of landscape structure, landscape connectivity may provide valuable insights for understanding ecological processes related to habitat fragmentation. Landscape connectivity describes the degree to which the landscape facilitates or impedes movements of animals among resource patches (Taylor et al. 1993) and may be used to predict the distribution of their movements and activity in fragmented areas. Landscape connectivity has been successfully linked to ecological processes in an increasing number of studies (With, Gardner & Turner 1997), including analyses of dispersal distances and genetic relatedness among animal individuals or populations (Manel et al. 2003; Coulon et al. 2004).

More recently, Bélisle (2005) reviewed the notions of structural and functional connectivity. On one hand, landscape parameters that measure ‘the degree to which some landscape elements of interest are contiguous or physically linked to one another’ refer to structural connectivity (With et al. 1997; Tischendorf & Fahrig 2000). On the other hand, landscape parameters that take into account dispersal or movement abilities of the studied organisms – i.e. that measure landscape connectivity with respect to how organisms may perceive landscape – refer to functional connectivity. Species-specific studies would gain in efficiency and applicability if behaviour-based functional connectivity was favoured. Without any tight link to behaviour, it is necessary to consider in connectivity analyses several to many landscape descriptors, as well as their respective interactions. Furthermore, several trials may be necessary to determine which spatial scale is the most relevant for the target species. Overall, batteries of nonindependent analyses repeated at different spatial scales may be required. This increases risks of misinterpretations due to type I statistical errors (null hypothesis rejected while it is true), or even to type II errors (null hypothesis accepted while it is false) if probability values are corrected for too many tests.

The objectives of this study were: (1) to develop an index of functional landscape connectivity for a Neotropical understorey frugivorous bat Rhinophylla pumilio Peters (Phyllostomidae), whose foraging pattern has been well described (Henry & Kalko 2007); (2) to test whether functional connectivity, jointly with resource availability, may contribute to explain the distribution and sustainability of these bats within a fragmented rainforest; and (3) to assess whether the functional connectivity index may be also applied to the other species of the guild of understorey frugivorous bats, including the genera Carollia and Sturnira (Phyllostomidae). These understorey frugivorous bats act as important seed dispersers in tropical forests (Henry & Jouard 2007) and are sensitive to modifications of landscape structure (Estrada, Coates-Estrada & Merrit 1993; Brosset et al. 1996; Kalko 1998; Schulze, Seavy & Whitacre 2000; Estrada & Coates-Estrada 2001, 2002; Faria 2006). Previous studies (Gorresen & Willig 2004) succeeded to link the abundance of frugivorous bats to various indicators of structural landscape connectivity (e.g. forest cover, densities of fragments). In this study, we further introduce keystone behavioural components to develop an indicator of functional landscape connectivity.

Understorey frugivorous bats feed on well-scattered food items, mainly small fruits produced in small amounts and for extended periods of time (‘steady-state’ fruit crops; Kalko 1998; Thies & Kalko 2004) by understorey plants. This forces bats to be constantly engaged in flights devoted to food search, and to rarely commute over long distances (Fleming, Heithaus & Sawyer 1977; Heithaus & Fleming 1978; Thies, Kalko & Schnitzler 2006; Henry & Kalko 2007). This foraging behaviour appears incompatible with the patchy distribution of resources in fragmented forests. Thus, we hypothesized that the distribution of understorey fruit bats was more dependent on the degree of landscape connectivity (ensuring spatial continuity of resource distribution) than on local conditions of resource availability.

For that purpose, we pursued a long-term bat survey, coupled with local estimates of resource availability, that was initiated 10 years ago in Saint-Eugène, French Guiana (Granjon et al. 1996; Cosson, Pons & Masson 1999a; Cosson et al. 1999b; Pons & Cosson 2002). In this study area, forest fragments are land bridge islands recently isolated from the continuous forest by a reservoir lake. Therefore, one can consider that the landscape is composed of two elements only – the mature forest and an aquatic matrix devoid of fruit resources that could modify the local conditions of resource availability and spatial continuity (Leigh et al. 2002). After validating a connectivity modelling in this simplified system, one may transpose it into more complex landscape mosaics with matrix habitats of various quality.

Methods

study area and time periods

Surveys were undertaken at the Saint-Eugène study area (4°51′N, 53°04′W), northern French Guiana. The mature forest surrounding Saint-Eugène was fragmented by the creation of the Petit-Saut hydroelectric dam built on the Sinnamary River, 60 km downstream from the study area in early 1994 (Cosson et al. 1999a). The subsequent flooding transformed 465 km2 of continuous forest into a reservoir lake covered by 100 km2 of tiny forested islands, mostly < 10 ha in area (Claessens et al. 2002). Total annual rainfall averages 3250 mm, with a main dry season from August to November and a shorter, less marked one in early March.

R. pumilio is the most abundant understorey frugivorous bat in this area, the other species from the guild being Carollia brevicauda (Schintz), C. perspicillata (Linnaeus) and Sturnira tildae (E. Geoffroy) (Cosson et al. 1999a). All bat surveys were conducted during 1–1·5-month sessions during main dry seasons of years 1995–97 and 2002–04. These two periods corresponded, respectively, to second to fourth and ninth to 11th years following fragmentation. They were termed periods of ‘recent’ and ‘older’ fragmentation, respectively.

bat surveys

We selected four mainland sites and 14 islands (size 0·8–7·5 ha, Fig. 1) surveyed at least twice in 1995–97 (Cosson et al. 1999a; Cosson & Pons, unpublished data), and repeated the same sampling methods in 2002–04. The 18 capture sites are located within a 4 × 4-km area encompassing a portion of the flooded lake and the adjacent mainland. A pre-fragmentation survey in 1993–94 ensured that fruit bat abundances and communities were homogeneous across the study area before fragmentation (Cosson et al. 1999a). Within each site, bats were captured simultaneously at two to five capture stations located at least 50 m apart and at least 20 m away from the shoreline. Each capture station consisted of a group of three mist nets (12 × 2·5 m, mesh 38 mm) set at ground level in T pattern when possible, or in line otherwise. To avoid biases due to trap-shy behaviour of bats, each site was sampled a single night per field session, but four to six times out of the six field sessions. From one field session to the next, we tried to set stations at the same place, but this was often difficult due to tree falls and vegetation regrowth on trails that remained unattended between field sessions. The spatial extent occupied by capture stations at the mainland sites roughly equalled that in islands (except for island 22, which was too small to accommodate more than two capture stations).

Figure 1.

Study area map showing the location of the fragments where the radio-tracking survey was undertaken, as well as the 18 capture sites: four mainland sites (letters) and 14 forested fragments (numbers).

During capture nights, nets were opened from dusk to dawn (18.30–06.30 h) and were continuously checked during the first and the last 2 h of the night, and every 2 h otherwise. Capture interruptions due to heavy rain were rare and short because we worked during dry seasons. No netting was done during the 6-day periods encompassing the full moon owing to possible slowdown in bat captures due to lunar phobia (Lang et al. 2006). Captured bats were kept in cloth bags before being identified to species following a key derived from Charles-Dominique, Brosset & Jouard (2001) and Simmons & Voss (1998), and released on capture sites.

food resource availability

In mature forests of French Guiana, the different understorey fruit bat species consume the same fruit resources but in different proportions. To quantify food available to understorey fruit bats, we focused on several epiphytes (Cyclanthaceae and Philodendron spp., Araceae) and on Piper (Piperaceae) plants that are known to constitute keystone resources for R. pumilio and Carollia spp., respectively (Fleming 1982, 1985; Cosson 1994; Cockle 1997; Charles-Dominique & Cockle 2001; Thies & Kalko 2004; Delaval, Henry & Charles-Dominique 2005). Sturnira spp. consumes mostly Solanum fruits within forest edges and second growth in lowland disturbed rainforests of Guiana and Amazon (Marinho-Filho 1991; Cosson 1994; Lobova & Mori 2004). These plants are scarce and probably restricted to large tree fall gaps within undisturbed rainforests. They were so rarely encountered in our study sites that we could not assess their density. However, S. tildae mostly feeds on epiphytes and Piper sp. within undisturbed rainforests (Cosson 1994; Cockle 1997; Charles-Dominique & Cockle 2001; Delaval et al. 2005), making our food availability survey appropriate for it as well.

Plant resources were censused within four to five 200-m2 plots (5 × 40 m) per site, uniformly distributed along capture stations. Botanical surveys occurred in November 1999 and 2004 and were associated with recent and older fragmentation periods, respectively.

Only shrubby Piper individuals > 50 cm tall were counted, i.e. the estimated minimum size required for fruit production. To better estimate potential fruit production on each Piper individual, we counted the number of terminal branches that could potentially bear fruits, and the number of flowering and fruiting spikes. Piper spikes ripen in the evening and are generally removed by bats within the following hours or else fall off in the morning (Thies & Kalko 2004), which rendered it difficult to introduce removal rate as an additional variable in our analyses.

Epiphyte infructescences constitute the main diet of R. pumilio and are frequently consumed by Carollia and Sturnira species in French Guiana. Following diet descriptions provided by Cockle (1997), we concentrated our interest on Asplundia heteranthera, Evodianthus funifer and Thoracocarpus bissectus (Cyclanthaceae), and several Philodendron species (Araceae), including P. billietae, P. grandifolium, P. insigne, P. linnaei, P. pedatum, P. squamiferum and P. subgenus Pteromischum (P. duckei, P. guianense, P. placidum). Most of these are epiphytes whose adventitious roots develop on trunks, at understorey to subcanopy levels (1–8 m above ground). Therefore, we could visually census adult individuals with reasonable accuracy, but the presence of fruits could not be documented as in Piper.

descriptors of landscape connectivity

To describe the local level of forest connectivity at each capture site, we used two landscape descriptors derived from Hewison et al. (2001) and Coulon et al. (2004), a landscape connectivity index and a habitat remoteness index. The connectivity index measures the extent of forest cover within a certain radius around capture sites, while the remoteness index is an indicator of isolation measuring the potential difficulty for bats to reach this site from the nearest area of continuous forest. These calculations were based on a SPOT satellite image (resolution 20 m) of the study area taken in 1996, transformed into a binary map (water vs. forest habitat) of 250 × 250 pixels and exported as a binary text matrix using the software ImageJ 1·33u (National Institutes of Health, USA; URL: http://rsb.info.nih.gov/ij). Connectivity and remoteness values were calculated in three steps (Fig. 2). First, we assigned to each landscape unit, i.e. each map pixel corresponding to a 20 × 20 m plot, an arbitrary suitability value equalling 0 or 200, for water and forested units, respectively. Second, we computed for each landscape unit the mean suitability value of all neighbouring landscape units within a given radius. A mean value of 200 indicates that the considered landscape unit is completely surrounded by forested units within the chosen radius, while a value of 100 indicated that only half of the landscape units are forested. These mean values denote the proportion of forest cover within the chosen radius but do not take into account the size of each forest island. To overcome this limitation, we assigned higher weighting coefficients to closer landscape units, and lower weighting coefficients to farther ones, for the calculation of mean suitability values (see below). The resulting forest connectivity index ranges from 0 (no forested habitat within the chosen radius) to 200 (only forested habitat) and decreases sharply when forested habitat becomes scarce in the immediate vicinity of the considered landscape unit.

Figure 2.

Treatment of the map for calculation of the landscape connectivity index (CI) and remoteness index. Step 1: initial binary map where each pixel (or landscape unit) can take only two values, 0 for ‘water’ and 200 for ‘forest’. Step 2, we assigned to each landscape unit the mean value of all neighbouring units located within a 400-m radius. Step 3 is a similar process, but values of neighbouring units in step 2 are weighted by a coefficient depending on their respective distance to the considered landscape unit (see Results, Fig. 3) to transform the step 2 matrix into the CI matrix. Upper graph shows the CI profile of all landscape units transected by the segment [AB] bridging island 22 to the nearest landscape unit of maximum CI. Lower graph is the same, but showing the reverse values 1/(CI + 1) that are summed to produce the final remoteness index.

Once the connectivity matrix was computed, we assigned to each capture site the connectivity index (CI) of their central landscape unit. The remoteness index (RI) was calculated as the sum of the inverse [1/(CI + 1)] of all landscape units a bat has to cross when reaching the capture site by flying in a straight line from the nearest landscape unit of maximum connectivity index (CI = 200). Thus, RI equals c.0 when the capture site has a CI = 200, and otherwise increases sharply when isolated by large numbers of landscape units with CI = 0.

Following this method, two important parameters are needed for CI and RI calculations, namely (1) the length of the radius enclosing the so-called neighbouring landscape units, and (2) the mathematical function determining the distance-dependent weighting coefficients assigned to these neighbouring units. To define weighting coefficients, we sought a mathematical function describing the frequency distribution of minimal distances bats cover when they search for food. Search flights devoted to food localization are an important component of foraging activity of understorey fruit bats (Fleming et al. 1977; Henry & Kalko 2007), and their lengths give an indication of the extent of habitat bats can investigate in a single flight to find fruiting plants. The longest distances bats cover in a single flight set the radius length, i.e. the distance after which weighting coefficients would become zero. Detailed flight distance data of R. pumilio were obtained by radio-tracking surveys.

movement pattern of r. pumilio

To assess flight distances, we radio-tracked five individuals. The radio-tracking survey was carried out in December 1995 on a small 7·5-ha fragment (one male and one female) and in November 1999 on a larger 28-ha fragment (two males and one female) (Fig. 1). The latter, rather flat, fragment was crossed by numerous trails forming a 100-m spacing grid, which greatly facilitated radio-tracking. Bats were mist-netted on these fragments and fitted with 0·70-g radio-transmitters (Biotrack, Dorset, UK) representing < 7·5% of their body mass. Transmitters were attached to the back of bats using surgical SkinBound® (Smith and Nephew Inc., Mississauga, Ontario, Canada) after a small amount of dorsal fur was trimmed. Bats were released at the capture site within 30 min after capture. No tracking data were taken on the capture night to avoid biases resulting from stress response to manipulation. Bats were tracked afterward for 3–4 nights (19.00–06.00 h) by two observers in radio-contact and each equipped with a CE-12 receiver (Custom Electronics, Urbana, IL, USA) and a four-element Yagi antenna. Radio-tracking nights involved determining whenever possible the hanging locations of bats by triangulation. Bats were considered to be hanging when the signal intensity was judged constant in direction and intensity for at least 1 min. Triangulation data were computed and analysed with the software Tracker 1·1 (Camponotus AB, Solna, Sweden 1994) after invalid bat hanging locations were discarded, i.e. points > 400 m away from observer positions (maximal estimated detection range). Flight distances are defined by the linear distance between each pair of successive hanging locations visited by bats. As discussed in Henry & Kalko (2007), the particular diet of R. pumilio in our study area, mostly epiphyte fruits, makes search flights easy to monitor. As epiphyte infructescences are too large to be removed and transported in feeding roosts, bats consume them in situ (Cockle 1997). Therefore, a part of the hanging locations reported by radio-tracking on R. pumilio may correspond to the location of consumed fruiting plants.

To improve the reliability of the mathematical modelling of the flight distance frequency distribution, we supplemented our data set with similar data collected on six nonreproductive individuals of R. pumilio using the same methodology (two males and four females; Henry 2005; Henry & Kalko 2007) in the undisturbed forest of the Nouragues natural reserve (110 km south-east from Saint-Eugène; Charles-Dominique 2001). Flight distance data were compared among study areas using a nested anova to measure to what extent data sets should be considered separately to produce two distinct series of connectivity index.

statistical analyses

The purpose of our study was to discriminate between the respective contributions of landscape connectivity, food resource availability and age of fragmentation in explaining the local abundance (capture rate) of R. pumilio, and to determine if the same connectivity index could be applied in a similar way to the other understorey frugivorous bats. Owing to low capture rates, data on the other understorey frugivorous bats were pooled together and referred to as ‘shrub-frugivorous bats’ to mark the contrast with R. pumilio that is rather specialized on epiphytes.

Raw data of capture rate (number of captures per unit of capture effort) were strongly asymmetrical due to low capture rate in most of forest fragments. Instead, we used as indicators of bat abundance the residual values of the linear regression of capture numbers against capture effort (expressed in ‘station-night’, i.e. three mist-nets open during one night). As residuals can be either positive or negative, they tend to have a symmetrical distribution. This procedure re-established normality of capture data when both capture numbers and capture efforts were transformed using log(value + 1). We produced two indicators of bat abundance. Abundance 1 refers to residual values where the regression line of capture numbers against effort is forced to pass through the origin of axes (intercept = 0) as what we would logically expect. In abundance 2, the regression line had no constraints, which resulted in slightly negative intercepts, steeper slopes, and offered data sets with greater variance. In all cases, the log-log regression of capture numbers against effort was significant (n = 36 and P < 0·001, R2 = 0·29–0·67).

In a first step, we used generalized linear models (GLMs) to determine if explicative variables (connectivity index, remoteness index, numbers of epiphytes and of Piper branches and spikes per census plot) were linked to each other, and particularly whether food availability varied along gradients of forest connectivity and from one study period to the next. Then, we averaged resource availability values of the various census plots within each capture site, and introduced in models the bat abundance values as dependent variables. We used a forward selection process and report the amount of variability introduced variables explained at each step. Statistical analyses were performed on Systat 9·0.

Results

bat surveys

The six field sessions totalled a capture effort of 255 station-nights (or 765 net-nights) and yielded 267 understorey fruit bats (Table 1) belonging to our four target species, namely the epiphyte-specialist R. pumilio (59·2% of all captures), and the three shrub-frugivores: C. brevicauda (18·7%), C. perspicillata (10·9%) and S. tildae (11·2%). Only 6·5% of the shrub-frugivores were captured in fragments (of a total of 109) against 31·6% of R. pumilio individuals (out of 158). Therefore, the decrease in abundance in fragments compared with the mainland was significantly more pronounced for the former species than for the latter ones (χ2 = 24·366, d.f. = 1, P < 0·001). While shrub-frugivores were detected in five of the 14 (35·7%) fragments during the recent fragmentation period, their capture rates remained zero within fragments during the older fragmentation period. No intersite movements were recorded in the course of our bat surveys (which were not designed for that purpose). Conversely, two males R. pumilio were recaptured at the same capture site at which they were initially banded as long as 7–9 years before.

Table 1.  Detailed bat capture data (Effort: capture effort in numbers of station-nights; Rhp: number of captured R. pumilio, Shf: number of captured shrub-frugivorous bats), descriptors of landscape connectivity (connectivity and remoteness indexes) and mean values of food resource availability (number of Piper branches, Piper spikes and epiphytes per survey plot) within each site and each fragmentation period (recent 2–4 years post-reservoir construction vs. older 9–11 years)
PeriodSiteEffortRhpShfP. branchesP. spikesEpiphytesConnectivity indexRemoteness index
Recent fragmentation 51320146·2514·00 71·461·97
 61200160·7527·5019·21 94·671·17
 8 631 43·0010·0016·56121·430·40
 9 420 68·650·43
12 310175·2552·75 92·780·97
16 771 81·0017·5012·75129·200·34
17 680 73·7520·5015·68122·690·51
19 730 98·00 7·50117·130·47
20 801 83·5014·00118·570·77
21 610 87·989·50
22 420 57·3315·00 53·240·63
24 613 91·639·28
34 410135·201·78
53 631 68·67 4·67 81·987·43
Cp112238 97·5731·5720·97199·170·03
Oi21169 48·50 8·75200·000·00
Ro16281 87·0019·25176·190·20
Vi 510 41·75 3·7510·40193·480·09
Older fragmentation 5 800 54·00 6·7517·75 71·461·97
 6 600 14·25 1·7519·50 94·671·17
 8 330 14·00 0·7510·75121·430·40
 9 610 68·650·43
12 300 46·50 2·7511·25 92·780·97
16 320 25·50 6·7512·25129·20·34
17 300 31·50 3·2511·50122·690·51
19 640 25·50 4·0014·50117·130·47
20 830 23·25 1·5020·75118·570·77
21 800 87·989·50
22 400 32·50 7·0012·00 53·240·63
24 300 91·639·28
34 320135·21·78
53 310 30·50 1·50 9·75 81·987·43
Cp173150 47·1310·5017·13199·170·03
Oi1332  7·75 4·00 6·50200·000·00
Ro 930 26·00 4·5013·25176·190·20
Vi 442 34·50 2·7521·75193·480·09

movement pattern and forest connectivity indices

Radio-tracking surveys conducted on R. pumilio were not equally successful on the two fragments. The three individuals from the large 28-ha fragment could be conveniently tracked all night (contact time > 95%) because they apparently never left the fragment for foraging. A total of 93 flight distances between successive hanging locations were obtained along tracking sessions of 3–5 nights per individual. On the other hand, the two individuals from the small fragment 20 regularly commuted between adjacent fragments (16, 19 and 22 see Fig. 1), leading to a fairly low contact time and highly disrupted, intractable, tracking data sets due to the difficulty of navigating by night among unstable snags. The female roosted once on island 20 and twice on island 19. Therefore, analyses were restricted to individuals from the large 28-ha fragment.

Like the six individuals tracked in Nouragues (Henry & Kalko 2007), R. pumilio individuals in Saint-Eugène had rather small home ranges and displayed short flight distances between successive hanging locations (90% of flight distances < 200 m). Mean flight distances were slightly longer in Saint-Eugène than in Nouragues (116 ± 72 m, n = 93, and 102 ± 75 m, n = 231, respectively). However, a nested anova performed on the square-root transformed flight distances revealed that this difference could be attributed to a significant interindividual variability (F7,315 = 2·317, P = 0·026), rather than an effect of the study area (F1,315 = 0·248, P = 0·248). Therefore, we felt comfortable using tracking data from Nouragues together with those from Saint-Eugène to improve our modelling of flight distances of R. pumilio.

To model frequency distribution of minimal flight distances, we first pooled all of the 324 flight distance values and calculated for each 5-m distance class ranging from 15 to 415 m the proportion of flights observed (Fig. 3). Then, we applied a logistic regression on these values as a function of distance, which describes the probability that a given flight will at least encompass a certain distance. Owing to the absence of long commuting flights, flight distances displayed a homogeneous unimodal distribution, and the logistic regression explained a high proportion of variability (R2 > 0·99). The logistic regression was then used to determine the distance-dependent weighting coefficients required for the calculation of the forest connectivity index. According to this function, the weighting coefficient becomes nearly zero at a distance of 400 m. In other words, for each landscape unit, the connectivity index is calculated over a 400 m radius circular area (50·3 ha). The resulting connectivity values for the 18 capture sites ranged from 53·2 (the smallest island) to 200 (one of the mainland sites) and averaged 119·7 ± 45·8 (Table 1). The remoteness index was minimum (< 0·01) for the latter mainland site, peaked at 9·3–9·5 for two remote islands, and averaged 2·0 ± 3·2.

Figure 3.

Graphical representation of the weighting coefficient attributed to neighbouring pixels (or landscape units) as a function of their distance to the considered landscape unit when calculating landscape connectivity. This curve was determined as the logistic regression of the observed frequency distribution of minimum flight distances R. pumilio covers between two successively visited hanging locations (dots). For instance, the probability that an individual covers at least 100 m from a hanging location to the next is about 0·42. The upper graph is a three-dimensional representation of the same function, X and Y being the spatial coordinates of neighbouring landscape units from the considered landscape unit.

food resource availability

A total of 886 Piper individuals was censused on one hundred and fifteen 200-m2 plots (14 sites per period). Epiphyte estimates are available for only 78 of these 115 plots (six and 14 sites in recent and older fragmentation periods, respectively) and totalled 1085 individuals. Piper productivity within plots (number of fruiting or flowering spikes per terminal branch) did not differ between mainland and islands (Mann–Whitney U = 1466, d.f. = 1, P = 0·211, n = 109). Assuming that the number of terminal branches indicates the potential fruiting rate (correlation between number of branches and number of fruiting spikes: Pearson r = 0·81, P < 0·001), we found that Piper resource availability was negatively and significantly influenced by the forest connectivity (F1,111 = 7·25, P = 0·008), and also underwent a significant 67% decrease from recent to older fragmentation periods (F1,111 = 4·12, P = 0·045, Fig. 4). Epiphyte density did not vary with landscape connectivity or between fragmentation periods (F1,74 = 0·60, P = 0·441, and F1,74 = 1·22, P = 0·272, respectively).

Figure 4.

Significant negative relation between landscape connectivity and the number of Piper branches per sampling plot, as shown by linear regressions. Piper branches were significantly less abundant in older fragmentation period, but the slope of the relation with the connectivity index remained unchanged (Table 2).

determinants of bat abundances

Variation in bat abundance was analysed using GLM, after resource availability values were averaged to produce a single value per capture site. As the numbers of terminal Piper branches and the numbers of flowering and fruiting spikes were tightly correlated (see above), we only retained the latter variable as an indicator of Piper resource availability. Thus, five explicative variables were introduced in models: connectivity index, remoteness index, mean number of epiphytes and of Piper spikes per plot, and study period.

In all combinations of species (R. pumilio on one hand and shrub-frugivores on the other hand) and abundance indicators (abundance 1 and 2), landscape connectivity was the best, if not the only, candidate variable retained in models (Table 2). The connectivity index alone explained 19–38% of the variance of bat abundance data. Shrub-frugivores were also positively influenced by Piper resource availability and their abundance decreased over time according to one of the two abundance indicators.

Table 2.  Sources of variation in the abundance of the epiphyte-specialist R. pumilio and of the shrub-frugivorous C. brevicauda, C. perspicillata and S. tildae within the different capture sites. Results are outputs of a forward multiple regression (GLM) where the introduced explicative factors were: landscape structure (landscape connectivity and remoteness), resource availability (mean number of epiphytes and of piper spikes per plot), and study period (recent vs. older fragmentation period). Factors contributing to explain a significant part of the total variation of bat abundance are given by order of appearance in the model, with their effect sign and their respective contribution to total R2. Abundance 1 and 2 refers to the two indicators of bat abundance (see Methods)
FactorsAbundance 1Abundance 2
F-ratioPSignR2F-ratioPSignR2
Rhinophylla pumilio
1. Connectivity indexF1,34 = 18·6< 0·001(+)0·35F1,34 = 10·80·002(+)0·24
 Shrub-frugivorous bats
1. Connectivity indexF1,24 = 14·70·001(+)0·38F1,34 = 8·070·008(+)0·19
2. Piper spikesF1,24 = 9·260·008(+)0·12    
3. PeriodF1,24 = 4·570·048(–)0·11    

Discussion

Our results have shown that we can rely on radio-tracking data to develop an index of functional landscape connectivity that may help in explaining the distribution of bats in heterogeneous landscapes or may predict their future variations in changing landscapes. Using landscape descriptors based on the foraging pattern of R. pumilio, we have shown that this species is sensitive to the loss of landscape connectivity per se, unrelated to food resource availability. The equations of regression lines in Fig. 5 may be used to predict the abundance of R. pumilio in landscapes with similar habitats and matrix. Furthermore, the same connectivity index contributed to explain variation in the abundance of other understorey frugivorous bat species, suggesting that one may use well-documented species as study models to make crude predictions about the distribution of other species of the same guild.

Figure 5.

Effect of landscape connectivity on abundance index 2 of understorey fruit bats during the recent and older fragmentation periods. All periods combined, equations of linear regressions are [y = 0·0039x – 0·4623] and [y = 0·0034x – 0·4046] for R. pumilio and shrub-frugivorous bats, respectively.

A functional connectivity index measures the degree of landscape connectivity with respect to how organisms may perceive the landscape structure (Bélisle 2005). In the present study, we expected the continuity of forest habitat to be an important feature for R. pumilio given its diet and foraging behaviour. Roost availability was not regarded herein as an essential feature as the roosting behaviour of R. pumilio appears flexible. Individuals roost in small groups under large leaves of fairly common epiphyte or palm tree species, located within or close to their foraging area. They can change roost every few days (Charles-Dominique 1993; Zortéa 1995; Simmons & Voss 1998; Henry & Kalko 2007). In other case studies, like cave-dwelling bats displaying high roost fidelity, roost availability may appear as the proximal factor influencing individuals’ distribution in landscapes. Thus, a functional connectivity index should have a special link with roosting behaviour, taking into account roost locations or densities and/or distances between roosts and foraging habitats.

r. pumilio, a case study

In the simplified system of Saint-Eugène composed of a forested habitat and an aquatic matrix, we found that the distribution of R. pumilio individuals was strongly influenced by our proximal index of forest connectivity based on foraging behaviour. Results support the prediction that local abundance of R. pumilio is more dependent on habitat fragmentation per se, i.e. the loss of landscape connectivity, than on resource availability. Landscape connectivity was the main determinant of its abundance, while food resource availability did not explain a significant portion of the variation in its abundance. The abundance of R. pumilio decreased along a decreasing gradient of landscape connectivity, whereas estimates of food availability remained stable (epiphytes) or even increased (Piper) along the same gradient. It is possible that changes in light and microclimate factors after the creation of the fragments locally increased the density of light-demanding pioneer Piper. Newly created edges are characterized by greater light penetration due to increased tree mortality and foliage drop induced by the physiological stress of moisture and temperature changes (Lovejoy et al. 1986; Kapos 1989; Malcolm 1994; Ferreira & Laurance 1997; Laurance et al. 1998, 2002) and it appears that Piper flourishes under these conditions.

The sensitivity of R. pumilio to the loss of landscape connectivity, despite the maintenance of food resource availability, could stem from the incompatibility of their foraging strategy with the obligation to cross expanses of matrix devoid of food sources. The foraging strategy, i.e. the manner in which bats move across landscapes to search for and exploit food resources, can be roughly decomposed into two components: search flights and commuting flights (Fleming et al. 1977; Heithaus & Fleming 1978; Henry 2005; Thies et al. 2006). While search flights are devoted to finding food items within foraging areas, commuting flights refer to longer straightforward flights conducted by bats among several foraging areas. To find their widely scattered food items, understorey fruit bats mostly rely on search flights and less frequently on longer commuting flights (Fleming et al. 1977; Heithaus & Fleming 1978; Bonaccorso & Gush 1987; Henry 2005; Thies et al. 2006). As a consequence, they might be reluctant to, or fail to, efficiently exploit patchily distributed food resources that impose frequent commuting flights over an unexploited matrix.

The foraging strategy of R. pumilio could even be considered as an extreme search strategy because these bats use almost exclusively search flights and therefore exploit a single small foraging area (Henry & Kalko 2007). Most of the surveyed fragments (0·8–7·5 ha) are smaller in size than their foraging area (3·5–14·1 ha; Henry & Kalko 2007). This might force individuals to split their foraging area into smaller ones distributed over two or several contiguous fragments, resulting in regular disruptions of search flights and thus lower their foraging efficiency.

Nevertheless, our data suggest that although R. pumilio was strongly affected by the loss of landscape connectivity, it could subsist at low densities in poorly connected habitats away from the mainland and could be a resident in the fragmented area. As a pre-requisite, we got direct evidence from telemetry that individuals can cross narrow habitat disruptions between fragments when these fragments are smaller in size than their home range. Most importantly, habitat remoteness, which was directly calculated from the proximal connectivity index, did not influence the abundance of R. pumilio. This indicates that the presence of R. pumilio in rather remote sites does not depend on the proximity of mainland, and suggests that local reproductive recruitment may occur in fragmented areas. Accordingly, on many occasions we found juveniles and reproductive (gestating or lactating) females in small fragments. Reproductive activity does not seem precluded by the loss of landscape connectivity for R. pumilio.

r. pumilio as a model species

R. pumilio was a successful model to build a connectivity index applicable to other species of the same guild, namely Carollia and Sturnira species, for which search flights are also thought to be an important component of foraging behaviour. However, the equation linking connectivity index to abundance (Fig. 5) should be regarded as a very rough predictor because it is based on scarce capture data combining several species. More capture data would be welcomed to refine the model, but this seems hardly feasible in this context where shrub-frugivorous bats have apparently deserted fragments during the transition from recent to older fragmentation periods. This general decline in abundance, also depicted by GLM analyses (Table 2) may be related to the pervasive decrease of Piper as a food resource over the study area (Fig. 4). A progressive closure of canopy foliage within edges could have resulted in the disappearance of many light-demanding Piper plants during the 5 years separating the two plant surveys. Many stems of dead young Piper were found in the later period.

We suggest that enlarging the spatial scale of the study would provide better insights about the relation between landscape connectivity and the abundance of Carollia and Sturnira species. At least, Carollia perspicillata exploits larger home ranges than R. pumilio and may regularly perform longer commuting flights among several (two to three) foraging areas located 500–1500 m apart (Heithaus & Fleming 1978). Thus, C. perspicillata is likely to perceive landscape connectivity in a different way, or at a different spatial scale. Accordingly, Gorresen & Willig (2004) found significant effects of landscape structure on the abundance of C. perspicillata within focus windows of 1–5 km radii, which is 2·5–12 times the spatial scale used for R. pumilio. We can imagine that our study transposed to C. perspicillata would yield a much extended curve of flight frequency distribution (Fig. 3) with high weighting coefficients attributed to landscape units situated within a much larger radius. This would produce connectivity values much smaller than for R. pumilio in the fragmented area, and also with much less intersite variability. In other words, C. perspiscillata is likely to perceive and respond to fragmentation at an earlier degree of structural connectivity loss compared with R. pumilio. This could also explain why Carollia species eventually disappeared from fragments at Saint-Eugène while R. pumilio did not.

difficulties and alternatives

Collecting data on the foraging pattern of bats poses a technical challenge due to their high mobility. Detailed data on flight distance collected herein could even be seen as an exception within bat studies, given the unusually short displacements displayed by R. pumilio (mostly < 200 m at a time) compared with other well-studied species. Available tracking data commonly state home ranges encompassing several tens to several hundreds of hectares in a few days, making flight distances difficult to monitor. Nevertheless, an increasing number of well-documented studies with standardized tracking protocols are now available, concerning either shrub-frugivorous bats (Thies et al. 2006) or insectivorous bats (Weinbeer & Kalko 2004; Meyer, Weinbeer & Kalko 2005). Although these studies may not document flight distance between successive hanging locations with as much accuracy as in the small home-ranged R. pumilio, they can at least document flight distance observed at fixed time intervals (e.g. every 3 min, or every 10 min).

In studies where the positioning of bats is irregular in time, we propose to rely on density kernels generated by analyses of home range or foraging area (e.g. Weinbeer & Kalko 2004; Henry & Kalko 2007). Methods of probabilistic kernels produce concentric density contours delineating areas of a given probability of presence. By plotting the inverse probability of presence (1 – P) on the vertical axis, against the radius of the corresponding area on the horizontal axis, one may obtain curves similar in shape to the frequency distribution of maximal flight distance (Fig. 3).

Finally, other methods than telemetry could provide data on flight distance frequency distribution, but these could only be operational at greatest temporal and spatial scales. Capture–recapture data, when summed over many individuals in a well-sampled study area, could be treated as flight distances. Genetic analyses performed on pollen or seeds collected on captured individuals can give data on the location of visited plants, providing plants are already indexed in the vicinity. Finally, homing experiments measuring the time it takes individuals to return to their roost as a function of translocation distance, have been used in various taxa as a means to assess landscape connectivity (see the review by Bélisle 2005, p. 1993). Homing experiments could be conducted on cave-dwelling bats.

implications and future directions

Our study contrasts with the other bat surveys conducted in forest fragments surrounded by abundant second-growth vegetation characterized by very high densities of the bat plants Piper, Solanum or Vismia (Estrada et al. 1993; Brosset et al. 1996; Kalko 1998; Schulze et al. 2000; Estrada & Coates-Estrada 2001, 2002; Gorresen & Willig 2004). In these studies, Carollia and Sturnira species remained abundant in forest fragments, second growth and pastures where they are effective seed dispersers (Galindo-González, Guevara & Sosa 2000). Therefore, bat surveys in Saint-Eugène indirectly underline the important role of second growth vegetation in shaping fruit bat communities in fragmented forests. A matrix colonized by second growth vegetation may provide bats with compensatory food supplies that favour the spatial continuity of resource distribution (matrix compensation and supplementation hypotheses; Dunning, Danielson & Pulliam 1992; Norton et al. 2000). In further studies, indicators of landscape connectivity should be adapted to more complex environments including various habitat types. In the modelling calculation, the different habitat types should be assigned to different suitability values ranging between 0 and 200, depending for instance on their relative contribution to the diet of bats. Such modelling connectivity in more realistic landscape mosaics is urgently needed. Among others, it would help understanding how bat diversity patterns may be influenced by the spatial configuration of the various neighbouring habitat types, and how communities are likely to be modified in changing landscapes.

Indicators of landscape connectivity may also contribute to an understanding of ecological processes at different levels of analysis. One may, for instance, expect intraspecific variation in the perception of landscape connectivity. Breeding females have to face critical energetic, temporal and behavioural constraints, and are usually forced to reduce the spatial extent of their foraging movements (e.g. Racey & Swift 1985; Kurta et al. 1989; Charles-Dominique 1991; Henry et al. 2002). In particular, lactating R. pumilio reduce their foraging area and flight distances by 42% and 25%, respectively (Henry & Kalko 2007), and are then likely to perceive landscape connectivity in a different way. This raises questions about spatial variation of population dynamics in fragmented areas. At the interspecific level, landscape connectivity may help investigate or predict possible effects of competition relaxation in areas where perceived landscape connectivity varies greatly among competitors. Conversely, studies of the mutualistic interactions between plants and seed-dispersing bats could benefit from a better knowledge of what landscape connectivity means from dispersers’ eyes.

Acknowledgements

We are grateful to G. Dubost and C. Erard who initiated this project. We thank R. Barbault, P. Charles-Dominique, T.H. Fleming, P.M. Forget, E.K.V. Kalko, L. Granjon, J.L. Martin, N.G. Yoccoz and H.J. Young for providing valuable criticism and suggestions. Many thanks also go to all the people who participated or helped in any aspect of the fieldwork, especially P. Cerdan (Laboratoire Hydreco, barrage de Petit-Saut), S. Poirot, A.S. Hennion, A. Lyet, R. Kirsch and D. Pons. The fieldwork was supported by the ‘Electricité de France’ (Convention Muséum/EDF CQZH 1294) and the Laboratoire d’Ecologie Générale de Brunoy (MNHN-CNRS, UMR 5176, France). M.H. received a Ph.D. grant from La Fondation des Treilles.

Ancillary