Sampling flying bats with thermal and near-infrared imaging and ultrasound recording: hardware and workflow for bat point counts

Study site and design

To showcase the utility of our sampling method, we sampled a bat community comprised of echolocating and non-echolocating bat species (i.e., Pteropodidae) in a tropical, agricultural ecosystem, in the lowlands of Jambi province, Sumatra, Indonesia. We surveyed bats in May 2019 in three sites in an oil palm plantation inside the PT. Humusindo Makmur Sejati company estate (01.95° S and 103.25° E, 47 ± 11 m a.s.l.), near Bungku village. Our research project (EFForTS – CRC990) has a Memorandum of Understanding with the company to allow research activities. Oil palm plantations are relatively open below the canopy and afford unobstructed lines of sight to test our sampling method. The center of each sampling site was within 10 m of a river and 10 m of an unpaved road. One field team (EY and co-author I) sampled the sites on consecutive nights and repeated this twice so that each site was sampled three times. Each night, we conducted three 10-minute bat point counts in the first hour after nautical twilight, and three 10-minute bat point counts in the second hour. Each of the three point counts was directed either towards the road, river, or oil palm plantation. Additional sampling methods were also used for a separate ecological assessment and comparison of the bat communities (Darras et al., 2021).

Hardware

Support rig. We constructed a sampling rig (Figure 1, Table 1) for mounting three different devices sensing ultrasound, thermal, and near-infrared waves. We used a ball head tripod to mount a custom-built U-shaped aluminium slat designed to attach all devices safely and point them exactly in the same direction. Threaded holes allowed to attach the infrared and thermal imaging devices, which have the most common tripod socket type (1/4" diameter, 20 threads per inch) on their main bodies. At the top ends of the slat, the infrared illuminator was attached with screws through their radiators. The ball head tripod enables smooth and rapid movement for the observer to actively track flying bats. We connected a cabled remote shutter release to the camera so that the tripod grip could be held in one hand and the shutter triggered with the other.

Figure 1. Sampling rig for bat point counts.

View from the front (A), from the observer or back (B), overview of the sampling rig (C), and near-infrared live view at night (D). 1: near-infrared illuminator (LEDs off, fitted with secondary optics); 2: thermal scope (front); 3: near-infrared camera (front); 4: ultrasonic microphone horn; 5: tripod ball head; 6: illuminator heat sink; 7: illuminator power cord; 8: near-infrared camera (back, with infrared live view); 9: thermal scope (back); 10: tripod; 11: cabled remote shutter release; 12: Automated ultrasound recorder; 13: power supply unit (inside tool box); 14: live view of trees illuminated with near-infrared light at night (trees are visible in the background due to incomplete filtering of near-infrared in most commercial cameras; this is not visible with the naked eye)

Our current mid- and near-infrared imaging setup, detailed below, was decided after trial-and-error with different devices. We tested digital rangefinders (Luna Optics) and optical night vision devices (Pulsar Challenger G2 3.5 × 56), infrared LED flashlights, small LED arrays fitted with secondary optics, and custom-adapted infrared digital cameras (Panasonic Lumix LX100). The ability to record high resolution infrared images and high thermal screen refresh rates were crucial.

Thermal imaging. We use a thermal scope used for hunting (Quantum XQ19, Pulsar) to detect flying bats. That thermal scope has an upscaled VGA screen resolution of 640 × 480 pixels, from its microbolometer’s 384 × 288 pixels resolution, which - in our experience - is the lower usable limit for detecting small bats at a distance of several meters. It also has a refresh rate of 50 frames per second, allowing smooth tracking of fast-flying bats without delay. Its relatively broad field of view of 19.5 × 14.7 degrees (horizontal × vertical) is essential for observing a large area and being able to track rapidly-moving objects. We disabled the sleep function, the digital zoom, and the automatic calibration to avoid interrupting operation.

Near-infrared imaging. We used a mirrorless interchangeable lens camera without its infrared filter for capturing near-infrared photographs. We removed the infrared filter ourselves (Darras, 2018), but some companies provide that service for private customers. The camera (Panasonic Lumix G8) was coupled with a moderate zoom, variable aperture lens (Panasonic G Vario 35-100mm f/4.0-5.6). This camera was equipped with a fast memory card to shoot 10 frames per second, and it has a large buffer allowing users to shoot extended image bursts. A fast data pipeline is essential for capturing many successive steps of the rapid bat passes, which can last several minutes.

We zoomed the lens between 35 and 50 mm, depending on the distance of the bats. We set up the camera in manual mode with a shutter speed between 1/250 s (for slow-flying bats) until 1/500 s (fast-maneuvering bats), an aperture of F4 (at 35 mm) or F5.0 (at 50 mm). We adjusted the gain with the “Auto ISO” function with a maximum ISO of 6400. We used the continuous shooting mode with the “high” burst rate setting and electronic first-curtain shutter. We used the manual focus mode, auto white balance, and saved pictures as a JPEG in “high quality”, at full resolution, with a 4:3 aspect ratio. JPEG saturation, contrast, sharpening and noise reduction were all at minimum. The exposure metering mode was set to center, the screen brightness to minimum, sensor stabilization and autofocus assist lamp were off.

We maximise the photograph exposure with artificial lights. We attached a custom-built near-infrared lamp made of four 12-LED rigid, aluminium infrared LED strips. The LEDs were outfitted with secondary optics - lenses that focused the emitted infrared beam to a more narrow cone of 14 degrees FWHM (Full Width at Half Maximum: the angular width of the beam when the intensity at the edge is half the intensity in the center of the beam). The LED strips are cooled with vertically arranged radiators attached with screws into the lens attachment holes of the LED strips. Our setup required us to use plastic rings below the screws to avoid shorting the electrical circuit. We did not use visible light or visible light flashes to illuminate bats as to avoid disturbing them, and also to avoid attracting insect prey.

The near-infrared illuminator requires a custom continuous current supply unit built by AK (Darras, 2021). The power source is a lithium-polymer 4500 mAh (66 Wh) battery, four voltage converters are each powering one LED strip with a constant 0.7 A. The power adapter is turned on with a toggle switch on the side of the toolbox; a green LED indicates when it is powered. The battery capacity theoretically lasts for 30 min of use and there is currently no indication of the battery charge. To avoid over-discharge, the battery should be recharged after 20 minutes, using a balanced charger.

Ultrasound recording. We used an open-design full-spectrum microphone, capable of recording audible sound or vocal notes, as well as ultrasound bat calls (Parus microphone, (Darras et al., 2018)). We attached the microphone to the sampling rig with rubber bands to isolate it from handling noise; it was fitted with an ultrasound horn (Wildlife acoustics) to make the microphone more directional and sensitive to the sound coming from the direction we pointed it to. The ultrasonic horn, even though it makes more directional recordings, does not entirely exclude echolocation calls from outside of the thermal scope’s field of view, especially when bats are close to the microphone. The microphone was connected to an acoustic recorder (SM2Bat+, Wildlife Acoustics) with a 3 or 5 m cable, laid on the ground below the tripod (Figure 1). It recorded mono audio (on one channel) at a 384 kHz sampling frequency in WAV format, without triggers.

Workflow

Field procedure. At each sampling site, we determined the thermal detection area by measuring the maximum detection distance in the three observation directions (oil palm, river, and road), each separated by 120 degrees. We pointed the thermal scope at the hand of co-author I facing away from the thermal scope to gauge at what maximal distance bats, that are roughly as large as a human hand, are still visible in the thermal scope. We then measured the distance of co-author I using a rangefinder (pointed at a hand-held whiteboard (A4 sheet dimensions) acting as a reflector, illuminated with a headlamp to facilitate aiming at it.

Before sampling each direction, we observed in the infrared camera where the bats’ flyways generally are. We then held the whiteboard at the most often used distance in flyways, measured its distance with the laser rangefinder, and wrote it on the whiteboard along with the site code and the direction (oil palm, road, river). The whiteboard was used to autofocus the infrared camera to that distance to clearly identify the site. We used a dedicated smartphone app to calculate the depth of field at the corresponding focal length, subject distance, and aperture (given the sensor size), making sure it was at least four meters wide.

We set up devices on the sampling rig and turned them all on, aligning the direction of the thermal scope and infrared camera precisely. We checked that the camera and ultrasound recorder times were synchronous. We started manual recordings and mentioned the name of the plot and direction vocally. Ten minute counts were measured with a timer. We used the thermal scope to scan the field of view (120 decimal degrees) in front of the observer (EY) up and down, then shifted the frame sideways, systematically, until a flying bat was detected. When a bat was detected, the observer told the assistant to switch the LED illuminator on, mentioned "bat detected", and took infrared pictures. The bat was described while tracking it, recording the flight pattern (“insectivorous” when maneuvering, “frugivorous” when not maneuvering), activity (flying, hunting, hanging, etc.), and the number of individuals. When the bat was gone, the observer said "bat gone". The observer rested for a maximum of 5 minutes until the next point count.

Processing ultrasound recordings. Ultrasound recordings were retrieved from the recorder and uploaded in ecoSound-web (Darras et al., 2020) to annotate the thermal and non-thermal bat detections inside spectrograms. Non-thermal detections are bat passes that were picked up only by the ultrasonic microphone - they were not detected by the observer because the bat was outside of the thermal scope’s field of view. We zoomed into the recording and scrolled through it until bat calls or camera shutter sounds were found to determine the bat detection timings. The start and end timings of a thermal detection were determined from the vocal mentions (“bat detected/gone”). The start and end timings of an ultrasound-only detection were determined from the start and end of the bat pass based on the visible echolocation calls; we defined bat passes as sequences of at least two calls less than one second apart. Ultrasound recording spectrograms (Fast-Fourier Transform with a window size of 1024) were screened by EY at a magnification of 860 pixels/60 s and cross-checked by co-author I. We noted the number of individuals mentioned by the observer in the annotation for thermal detections. Bat passes from the same individual that were separated by less than 10 seconds were merged into one tag. For thermal detections, we entered an ID into the annotation, thereby linking it to a table containing their vocally mentioned observation data. All detected bats were flying.

Each annotation with echolocation calls was directly assigned a sonotype or species. We compared bat calls against our own reference collection of bat calls obtained from captured bats (Chiroptera reference collection in ecoSound-web) to identify the calls to species; when inconclusive calls were found, we assigned them to sonotypes. Within each recording, we measured call parameters for each bat species and sonotype that was recorded clearly (to avoid biased call parameters from distant calls). We measured call parameters based on the three strongest, not saturated calls: call duration (duration of a single pulse), inter-call interval (time from the start of one call to the onset of the next), start frequency (frequency value at the start of the call), end frequency (frequency value at the end of the call) and peak frequency (frequency with maximum energy for the whole call).

Processing infrared images. We used infrared imagery to confirm whether detections are from insectivorous (echolocating) or frugivorous (non-echolocating) bats using visible morphological features. Using the photograph meta-data (Exif), the infrared images were automatically assigned to the corresponding detections based on their timings (see R script in (Darras, 2021). Images with ambiguous timings were manually assigned to detections. For each point count, we counted the number of images shot and the number of images containing a bat image to compute the hit rate. For each detection, we noted the number of images containing a bat. We used the latter to record the presence of diagnostic morphological features (Figure 2) into our observation table: we distinguished between small eyes (characteristic of insectivorous bats) and large eyes (characteristic of Pteropodidae); we distinguished large ears (only found in insectivorous bats such as Hipposideridae, Rhinolophidae, Megadermatidae, Nycteridae) from smaller ears (found in Pteropodidae and insectivorous bat families as well), we distinguished between tail types (A: short tail, not enclosed in reduced interfemoral membrane, as in Pteropodidae, B: tail enclosed in interfemoral membrane as in Rhinolophidae, Hipposideridae, Vespertilionidae, Miniopteridae (Srinivasulu et al., 2010)).

Figure 2. Near-infrared photographs showing diagnostic morphological features, obtained with bat point counts.

Pictures were cropped from the original.

Determining the workflow. We assessed the usefulness of the different data types obtained with bat point counts: the observed flight pattern, the near-infrared bat photographs, and the ultrasound recordings. We used infrared imagery to generate subsets of confirmed insectivorous (having large ears, B-type tails or small eyes) and frugivorous (having large eyes or A-type tails bats). Among those, we counted for how many detections each flight pattern type could be observed and the number of detections with and without echolocation calls. We computed the total hit rate (proportion of photos with bats) and the useful hit rate (proportion of photos with bats that had a visible aspect of the morphology). Based on our results, we prioritised the different data types and devised a workflow for taxonomic discrimination and identification to the lowest possible level.

Software and hardware

Workflow

Confirmed frugivorous bat detections often had no simultaneous echolocation calls (85 %) and non-maneuvering flight patterns (both 71 %). We found seven confirmed frugivorous bat detections with visible large eyes, one had a visible tail characteristic of Pteropodidae. One of these had simultaneous echolocation calls, and two had a maneuvering flight pattern. Large reflective eyes were also visible from far away, even before the bat body was clearly visible, and also when the bat was flying sideways to the observer (Darras, 2021). Frugivores’ tails were generally not observed as they were generally gliding low and not maneuvering much.

Confirmed insectivorous bat detections almost always had simultaneous ultrasound (97 %) and often had maneuvering flight patterns (78 %). We found 31 confirmed insectivorous bat detections with either small eyes (31), large ears (1), or a visible tail with a patellum (with the tail enclosed in the interfemoral membrane) (33). Only one of these detections had no simultaneous echolocation call, and eight had a non-visible flight pattern.

Based on these results, we devised the following workflow for assigning taxonomic identifications of thermal bat detections (Figure 3). When detections had infrared photographs showing clearly visible aspects of the bat morphology, they were used to determine the guild; insectivores were identified using the simultaneously recorded echolocation calls and infrared imagery was used to confirm and narrow down their identity; frugivores were identified based on the infrared imagery. We excluded detections without any near-infrared photographs or ultrasound (6% of thermal detections). For detections that had bat photos but no visible bat morphology, we considered the bat to be inside our detection range for ultrasound: When there were simultaneous echolocation calls, we deemed the bat to be insectivorous and used the calls to determine its identity; When there were no simultaneous echolocation calls, we deemed the bat to be frugivorous but could not identify it further than the family level. Using this workflow, 84% of thermal bat detections could be assigned to a guild and subsequently identified to genera or species using ultrasound and near-infrared data in a subsequent ecological study (Darras et al., 2021)

Figure 3. Workflow used for discriminating bat taxa with bat point count data.

The circle diameter scales with the number of detections. Green disks correspond to detections where species identification can theoretically be achieved; orange corresponds to detections where identification could be achieved on the family level only; red corresponds to those where no identification is possible.

Underlying data

Open Science Framework: Bat point counts - design, analysis, images & data

https://doi.org/10.17605/OSF.IO/ZHP3W (Darras, 2021)

This project contains the following underlying data:

Extended data

Open Science Framework: Bat point counts - design, analysis, images & data

https://doi.org/10.17605/OSF.IO/ZHP3W (Darras, 2021)

This project contains the following extended data:

Data are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).