Materials and methods

Study site and design

We surveyed bats in three different sites in a closed-canopy oil palm plantation using bat point counts, mist nets, and automated ultrasound recorders. Our sampling sites are inside the Humusindo Makmur Sejati (01.95° S and 103.25° E, 47 ± 11 m a.s.l.) company estate, near Bungku village in the lowlands of Jambi province, Sumatra, Indonesia. We set the center of each site within 10 m of a stream (2-4 m wide) and an unpaved road (4-5 m wide) in order to maximise potential species detections, as it is widely known that bats use trails and streams for commuting and hunting (Voigt & Kingston, 2016). The sites were bordering the same river and separated by at least 600 m to allow independent captures. We sampled all three sites simultaneously with rotating methods on three consecutive nights with one field team, and we repeated this twice, obtaining three sampling nights for each method and site in total (Fig 1). The surveys occurred during nine nights from 21 to 31 May 2019. Due to our selection of sites with identical surrounding habitats, the temporally rotating design of the methods, the simultaneous comparison of methods in equivalent sites, and the short sampling period, any effects of weather conditions, moon phase, and fluctuating food resource availability was minimized and should not bias our results.

Bat point counts

We conducted bat point counts for one hour per night in each site. We used a sampling rig with sensors for ultrasound, thermal, and infrared waves; we provide full technical details of the rig and the observation workflow in (K. Darras et al., 2021). Each point count scanned a 120° field of view, directed either towards the road, river, or oil palm. We determined the detection area of the thermal scope and the full-spectrum microphone - fitted with a horn to amplify sounds from the front - with a chirp emitter at 40 kHz (K. Darras et al., 2021). The first three point counts took place in the first hour after the survey started, and the next three point counts happened in the second hour. Between the two hours, the bat point counts team assisted bat extraction at the mist netting site that was permanently attended by a third person. Thus, it was not possible to sample bats with point counts for the same duration as with the other methods.

Mist netting

We mist-netted bats for 4 hours per night for a total of 576 net-hours (m✕h) in each site. Mist netting was our reference trapping method for sampling bats: we did not use harp traps as they were ineffective in previous assessments in oil palm plantations in our region (Darras, unpublished data). We opened four 3.2 m high × 12 m long nets 1 to 1.5 m above the ground for four hours starting at sunset (Ultra Thin Series M Mist Net, 20mm mesh, Ecotone). Nets were installed in presumed flight ways, delimited a quadrilateral, and their position relative to the river, oil palm, or road was approximately the same in all plots for each survey set. Most of the below-canopy flying space was covered with our nets. Mist nets were checked every 15 minutes from sunset to two hours afterwards, then every 30 minutes until four hours after sunset. Captured bats were kept in tissue pouches until the nets were closed. Bat morphology was measured to identify bats according to (Huang et al., 2014) directly in the field. There were no particular regulations or ethical guidelines for research on live bats in Indonesia at the time of the study, but we wore protective equipment (masks and gloves) to handle them.

Ultrasound recordings

We made continuous ultrasoundscape recordings (i.e., without triggers) with sound recorders for 4 hours per night in each site. One recorder (SM2Bat+, Wildlife Acoustics) were set up with one microphone (Parus open-source model, (K. Darras et al., 2018)) parallel to the ground, sampling at 384 kHz at 2 m height, starting at sunset and lasting four hours in each site. We measured the ultrasound detection space covered by the recorders (K. Darras et al., 2016): similarly to bat point counts, we pointed an ultrasound calibrator (Wildlife Acoustics) to the recorder, and recorded its 40 kHz chirps emitted from 2 m height at distances of 4, 8, 16, and 32 m, in three directions (to the river, the road, and the oil palm plantation) to derive the site’s sound transmission profiles (Fig S3).

Data analysis

Species identification

Ultrasound recordings from bat point counts and automated ultrasound recorders were uploaded on the open-source platform BioSounds (K. Darras et al., 2020) to annotate the spectrograms with identitfied bat detections; we included both acoustic as well as thermal-only detections (detections without ultrasound that were vocally mentioned by the observer) (K. Darras et al., 2021). All bat calls were identified using our reference collection of bat calls obtained from captured bats (Chiroptera reference collection in BioSounds (K. Darras et al., 2020)) and literature data (Hughes et al., 2011; Kingston, 2013; Zhu et al., 2012). We distinguished broadband frequency-modulated (BFM), constant-frequency (CF), and frequency-modulated, quasi-constant frequency (FM-QCF) calls. We measured calls for each bat call type within each recording, but only if the bat pass was recorded clearly (to avoid biased call parameters from distant calls), using the three strongest, not saturated calls: We measured peak frequency (Fmax, frequency with maximum energy), start frequency, end frequency, call duration, and inter-call interval (from the start of one call to the onset of the next). Frequency-modulated quasi-constant frequency (FM-QCF) calls were split in three sonotypes based on their end frequency: around 33 kHz, between 38 and 42 kHz, and around 48 kHz.
We matched ultrasound recordings and near-infrared photographs from bat point counts to their respective detections using a conservative workflow that discarded uncertain matches (K. Darras et al., 2021). We used data from one additional, incomplete survey night during which our infrared lamp power supply failed to aid with the taxonomic identification. Bat species identification usually relies on direct, external and internal body part measurements and categorical features of caught specimens. However, absolute measurements from pictures are inaccurate due to our large depth of field of approximately four meters: a 10 cm bat at a distance of 8 m would appear as large as a 15 cm bat at 12 m. Thus, we rely on categorical features as well as relative measurements of external, readily recognisable body parts - as they do not depend on the bats’ distance - for identifying bat point count detections. We used pixel-measuring software tools on photos where the measured body parts were parallel to the camera focus plane to avoid underestimates. We only used near-infrared images to identify thermal-only detections. We confirmed or determined the identity of bat sonotypes in bulk by using clear near-infrared pictures of selected detections. We devised a new identification key for South-East Asian bats found in oil palm based on (Huang et al., 2014) to determine bat identity from near-infrared pictures and ultrasound calls (Box 1).

Rarefaction and extrapolation sampling curves

We compared the species richness sampling effectiveness of bat point counts against mist netting and automated ultrasonic recording by comparing the size of the species pools sampled by each method using rarefaction and extrapolation sampling curves (Chao et al., 2014). We pooled all three sampling sites to represent our oil palm plantation’s bat community. We calculated a conservative estimate of each species’ abundance in sound recordings by usingthe maximum number of simultaneously recorded bats per night. We computed taxon presence/absence at each sampling hour and method as well as abundance matrices for each method, by summing the species abundances over the three sites. We only used the taxonomic identities yielded by each sampling method, independently of the insights gained from the other methods. We generated raw incidence as well as abundance-based rarefaction and extrapolation sampling curves and compared the number of species at a 95% sampling coverage for a robust estimation of diversity (Hsieh et al., 2016). We assessed the statistical significance of differences in species richness between sampling methods with 83% confidence intervals (Krzywinski & Altman, 2013).

Acoustic and thermal detection ranges

Thermal detection ranges of bat point counts were obtained for each direction by measuring the maximal distance at which the hand of a field assistant was detectable in the thermal scope. Acoustic detection ranges for point counts (with ultrasonic horn) and automated ultrasound recordings (without horn) were obtained for each direction from the intersection of the chirp emitter’s sound level curve against distance with the ambient sound level (K. Darras et al., 2016). Sampling spaces of mist nets are not determinable per se, but we assumed that they cover at least the inner area delimited by their border.