Fewer bat passes are detected during small, commercial drone flights

Site information

The study was conducted at the Kenauk Institute, an environmental research site, in western Quebec in July 2018, 2019 and 2020. All surveys occurred between 21h30 and 00h00 at night, with location and time of day randomized for each date of testing. Testing did not occur during inclement weather (rain or winds above 10 km/h). In 2018, during an initial field season, we surveyed bat populations using a traditional method (transect-based surveys) to determine which species were present. Six transects lasting 1.5 h each were laid out, and surveyed three times per season; three transects were located in open-canopy areas, and three were located in rugged, closed-canopy areas. Every 200 m, a flag marked a sampling point where we completed a 2-min static inventory using an Anabat SD2 (Titley Scientific, Columbia, MO). In this pilot study used to develop the main study, we observed all eight species known in Quebec, including the eastern red bat (Lasiurus borealis; 0.005 passes detected per minute in open-canopy habitat; 0.001 in closed-canopy), hoary bat (Lasiurus cinereus; 0.002 passes detected per minute in open-canopy habitat; 0.006 in closed-canopy) and tri-coloured bats (Perimyotis subflavus; 0.026 passes detected per minute in open-canopy habitat; none in closed-canopy). Species in the Eptesicus fuscus/Lasionycteris noctivagans acoustic complex were the most abundant (0.075 passes detected per minute in open-canopy habitat; 0.018 in closed-canopy) followed by Myotis species (Myotis leibii, Myotis septentrionalis, Myotis lucifugus; 0.075 passes detected per minute in open-canopy habitat; none in closed-canopy). Due to small sample sizes per species and because manual identification using spectrographic analyses can be unreliable for the differentiation of some bat species²², we pooled several bat species that had similar spectrograms into complexes. We pooled the big brown bat (Eptesicus fuscus) and the silver-haired bat (Lasionycteris noctivagans), and the Myotis species: little brown bat (Myotis lucifugus), northern long-eared Myotis (M. septentrionalis), and eastern small-footed bat (M. leibii)²². Therefore, these species are grouped together in analyses to minimize identification errors²². The big brown bat and silver-haired bat form the EPNO complex whereas the Myotis species form the MYSP complex. We identified to species the hoary bat (LACI), red bat (LABO), and tri-coloured bat (PESU)²². We identified bat passes visually using the output from the Anabat in the Anabat Insight software^17,23.

Detection efficiency

Because total bat passes per minute were seven times higher in open-canopy habitats than in closed-canopy habitats, in 2019 we focused our surveying efforts in relatively open habitats. The Anabat (420 g) is too large to attach to a drone, thus in 2019 and 2020, we used Echometer Touch bat detectors (20 g; Wildlife Acoustics, Maynard, MA), commercially available and inexpensive detectors, attached to iPod 7 s (88 g; Apple Inc., Cupertino, CA). We do not directly compare between surveys done with the Anabat and the Echometer Touch, but merely used the 2018 Anabat surveys as a guide for expected bat species and distributions in 2019 and 2020. The UAV used was a commercially available Phantom 4 quadcopter from DJI (1.3 kg, DJI Technology Co. Inc., Shenzhen, China). To reduce sound interference from the drone, which could reduce the detection range of the instrument, we placed a 2-in. Sonoflat acoustic foam (Auralex, Indianapolis, IN) divider between the recorder and the drone, as recommended by past studies^19,21 (Fig. 1).

In both 2019 and 2020, we surveyed in three phases: (1) a 5-min recording from the ground without UAV; (2) a 5-min recording while the detector was attached to the UAV using zip ties and carabiners and while the UAV was manually flown in a 10–15 m diameter circle at canopy height (5––10 m above the pilot), depending on the survey site; and (3), identically to Phase 1, a 5-min recording taken from the ground without UAV (Fig. 1). The ground recorder, used sparsely in 2019 and consistently in 2020, was 1 m above the ground during phase 2. Based on surveys in 2018, seven sites were identified as having higher relative activity and were repeatedly monitored in 2019 and 2020 for bat activity. Of the seven study sites, five were located next to bodies of water and four were located near buildings; all were located in open areas. Open spaces and bodies of water are preferred hunting grounds for most bat species¹⁸, and make for an easier and safer drone flight. An additional bat detector (Echometer Touch 2, Wildlife Acoustics, Maynard USA) was used on the ground during Phase 2 to simultaneously monitor bat passes from the air and from the ground, to indicate whether bats were present but not detected due to UAV noise interference. In 2020, ten surveys were conducted with Echometer Touch 2 recorders on (1) the UAV, (2) on the ground, and (3) at a control site > 1 km from the current site. Control sites were only used in 2020. Because different bat detectors, as well as different classification software, detect and identify bats at different rates, we do not directly compare among different detectors or software^24,25. In 2019, we used the Kaleidoscope software to identify bats automatically. We removed false identifications manually. In 2020, we used the Kaleidoscope software to identify all bats automatically. We also identified all passes visually and blind to the classification from Kaleidoscope. By classifying all bats using both software and visual identification, we aimed to determine whether our results were robust to identification technique.

Data were collected beyond Phase 1 if the site had a bat density above three passes per 5 min (2019: N = 24 without ground detector; N = 5 with ground detector; 2020: N = 10 with ground detector; all sample sizes refer to experiments that included Phases 2 and 3). If insufficient bat activity was recorded at a given site after a 5-min period, data collection moved on to the next site, and data from that site was excluded from any analyses. Phase 1 was done to ensure there was an established bat presence, and to maximize sampling. The length of each phase was extended to 10 min if two passes were detected by the 5-min mark of Phase 1, allowing for the collection of more data, while maintaining the time proportions of each phase. While this process, necessary logistically to obtain a sufficient sample size, could lead to more bats detected during Phase 1, there should be no impact on Phase 3 compared to Phase 2, and thus, we used Tukey tests to examine Phase 3 relative to Phase 2, as well as Phase 1 compared with both other phases²⁶.

Each drone flight was performed by two field technicians: a pilot and an assistant. The UAV pilot held a basic operations pilot certificate for a small remotely-piloted aircraft system, visual line-of-sight (certificate number PC1917023611) in accordance with federal regulations enforced by Transport Canada. The assistant held the bat detector during Phases 1 and 3. During Phase 2, the assistant acted as the drone’s elevated launching and landing pad as the additional equipment obstructing the UAV’s landing gear. For take-off, they held the UAV upright above their head and gradually let go as the UAV gained altitude. For landing, the pilot gradually decreased the altitude of the drone until the landing gear was safely grasped by the assistant, who then held the UAV above their head until the propellers stopped moving. All methods were carried out in accordance with the guidelines of the Canadian Council for Animal Care. All experimental protocols were approved by McGill University animal care committee under protocol 2015-7599 and complied with the ARRIVE guidelines for animals.

Statistical analyses were conducted using R 3.6.0 base package²⁶. Generalized linear models (glm, Poisson distribution) were performed to determine the effect of phase (i.e., 1, 2, and 3) and detector location (detector on the UAV or on the ground) on the total number of bat passes. Tukey tests were then used to determine what phases and locations were significantly different from one another. To assess interspecific variation in detectability, the difference between the mean detection rate for Phase 1 and 3 and the detection rate in Phase 2 were calculated by species for each survey. A glm was then performed on the difference in detectability by species ([Average of Phases 1 and 3 − Average of Phase 2]–Species). Species were divided into four categories: MYSP (Myotis species complex), EPNO (big brown bat/silver-haired bat complex), LABO (eastern red bat), and LACI (hoary bat). No tri-coloured bats were detected, and are therefore absent from analyses. Detection phases were also divided into four categories in relation to the UAV flight: Phase 1 (pre-flight), Phase 2 from UAV-based detection (during flight), Phase 2 from ground-based detection (ground), and Phase 3 (post-flight).

Detection capacity

To estimate the degree to which technological limitations affected the results gathered during the first experiment, a second experiment was conducted to estimate the impact of propeller-noise interference on the range of the bat detector. An Audio Generator SGA-8200 (Circuit-Test, Burnaby, Canada), connected to an Ultra Sound Advice S55/6 amplifier and loudspeaker (Ultra Sound Advice, London, UK) set to broadcast a 40 kHz sine wave at 40 dB SPL_A @ 1 m, the highest dB setting, was used to replicate the high amplitude ultrasound reached by most bat species during their echolocation calls²². The Echometer Touch bat detector was moved away from the speaker along a measuring tape until the ultrasonic frequency could no longer be detected by the microphone. The procedure was then repeated with the detector attached to the flying UAV. As ambient sound perception cannot be evaluated when the microphone is attached to the UAV, the spectrogram on the Echometer Touch cellphone app (Wildlife Acoustics) connected to the detector was recorded with the screen video recording feature of the iPod 7 (Apple). These recordings were taken as the drone and bat detector were flown slowly along the ground to three distances (10 m, 15 m, 20 m) away from the ultrasound generator to better approximate the detection range. The videos were later visually assessed qualitatively by estimating the distance at which the signal from the speaker could no longer be distinguished from the noise interference of the drone.

To quantify the spectral overlap of the drone with echolocation pulses, a spectral analysis of three 15 s recordings were performed using Avisoft SASLab Pro 4.40 (Avisoft Bioacoustics, Berlin Germany). These recordings included the drone flying, the drone motors running without propellers attached, and the ambient noise from the same location and time (control). Recordings were saved as 16 bit WAV files sampled at 256 kc/s and were normalized to 90% in SASLab Pro prior to parameterization. Spectrographs of those normalized recordings were generated using a Fast Fourier Transform length of 512 points, with a frame size of 100% and 75% overlap of Hann windows. This achieved a frequency resolution of 500 Hz and temporal resolution of 0.5 ms. Frequencies where noise was concentrated are evident from these spectrographs, but were confirmed by generating Logarithmic Power Spectra from each recording using Hann windowing achieving frequency resolution of 0.061 Hz. Noise is described at frequencies where the relative sound pressure level exceeded − 80 dB in those Power Spectra.

Source: Ecology - nature.com