Species of open-space bats that are relatively large, such as bats from the genus Nyctalus, are considered as high-risk species for collisions with wind turbines. However, important information on their behavior and movement ecology, such as the locations and altitudes at which they forage, is still fragmentary, while crucial for their conservation in light of the increasing threat posed by progressing wind turbine construction. We adopted two different methods of microphone array recordings and GPS-tracking capturing data from different spatio-temporal scales in order to gain a complementary understanding of the echolocation and movement ecology of Nyctalus aviator, the largest open-space bat in Japan. Based on microphone-array recordings, we found that echolocation calls during natural foraging are adapted for fast-flight in open space optimal for aerial-hawing. In addition, we attached a GPS tag that can simultaneously monitor feeding buzz occurrence and confirmed that foraging occurred at 300 m altitude and that the flight altitude in mountainous areas is consistent with the turbine conflict zone. Thus, our acoustic GPS survey clearly identified N. aviator as a high-risk species in Japan.

Microphone-array recordings

Microphone array recordings were conducted on a total of 10 days in August 2020 and 2021: August 4 and 5, 2020, June 26, 27, 30, July 2, 28, 30, August 1 and 2, 2021 (total 1210 min), at Tokiwa Park in Asahikawa city, Hokkaido, Japan (N43°46'28.6" E142°21'27.1"), in the proximity of N. aviator roosts within hollows of Ulmus davidiana. The measurement was started around 21:00 in 2020 and around sunset (approximately 19:00) in 2021, and the recordings were made for about 2–3 hours for each day until the batteries ran out.

A Y-shaped microphone array, consisting of four omnidirectional microphones (FG-23329-C05; Knowles Electronics, Itasca, IL), was placed at a height of approximately 1 m above the ground with the microphone tips pointing upwards. Based on the difference in arrival times between a central and three outer microphones, the 3D locations of flying bats were reconstructed using a custom made program in Matlab (Math Works, Natick, MA, USA). The detailed procedure of recording and calculation of the 3D positions is described in Fujioka et al 2011 and Mizuguchi et al 2022. The coordinates of the bats were calculated within a range of 47 m from the microphone array where the theoretical range error was less than 40 cm (corresponding to wing length of N. aviator).

Those echolocation calls that were recorded by the central microphone with a sufficiently high signal to noise ratio were further analyzed. The end and start times of the pulses were automatically obtained based on a -15dB threshold from the peak power. Then, the inter pulse interval (IPI), which is the time between the beginning of a pulse and the end of the next pulse, the pulse duration, the minimum frequency, the bandwidth and the peak frequency were calculated using a custom-made program in Matlab, respectively. We identified feeding activity based on a characteristic sequence of echolocation calls termed “feeding buzz” from the spectrogram.    

Acoustic-GPS logging

We caught bats using mist nets close to a roosting tree (Ulmus davidiana) next to Asahikawa elementary school in eastern Asahikawa city, central Hokkaido, Japan (N43°46'10.0" E142°26'27.5") during sunset on July 29^th, 2021. Then, we carefully attached custom-made acoustic GPS data loggers (ArumoTech Corp., Kyoto, Japan, 2.4g) with a telemetry unit (PicoPip Ag337, Lotek, Canada, 0.3g) using Skin Bond (Osto-bond, Montreal Ostomy Inc., Canada) to the back of the bat. We held the bats for about 10 min to allow the glue to dry and released them on the roosting tree.

The timers were set to start GPS logging every 5 seconds from 20:00 o’clock on the second day after the attachment until batteries ran out (approximately 1 hour). It is possible to continuously record pulse emission timing with a high temporal resolution of 4 MHz sampling rate, because acoustic GPS loggers are designed to output a high voltage when the acoustic signal voltage exceeds a certain threshold.

We identified feeding buzzes from the recorded IPI patterns based on observations from the microphone array recordings. In particular, pulse sequences with at least five pulses below an IPI threshold of 20 ms and a maximum duration of 100 ms were automatically classified as a feeding buzz (attack). The automatic classification of feeding buzzes corresponded to 100% of those that we visually found in the IPI sequences. We identified the coordinates of the attack location by resampling GPS data from 5 seconds to every second using linear completion, and searching for the timing closest to the end of a feeding buzz. The flight altitude was obtained by subtracting the ground altitude (the Geospatial Information Authority of Japan, https://www.gsi.go.jp/ENGLISH/index.html) from the absolute altitude above sea level recorded by the acoustic GPS logger. We identified habitat types (river, mountain, urban) from 'topographic' and 'satellite' maps provided in MATLAB. 

Statistical Analysis

All statistical analyses were performed in the R environment for statistical computing (R Core Team 2021) and its extended packages. We modeled the flight speed, altitude as well as the presence/absence of an attack, as a function of the habitat type (river, mountain, urban area) using linear modeling, generalized liner modeling [lme4, version 1.1-28, 29] and generalized linear mixed modeling using Template Model Builder [package glmmTMB_1.1.4, 30], respectively, to examine whether the tagged bat adapted its flight and foraging behavior to the respective habitat type.

In general, we examined the quality of a model fit graphically using the functions in the DHARMa package [version 0.3.3.0, 31] We checked whether the model explained more variance than its respective null model by comparing them either via a parametric bootstrapping method [package pbkrtest_0.5.1, 32] or via a χ² test (function anova, R 2022). A χ² type-II-Wald test [function Anova, package car_3.0.12, 33] was used to check the significance of the factor within the respective model. Bonferroni correction was applied to all pairwise post-hoc comparisons between the levels the factor (function lsmeans, package emmeans_1.8.0).

In the residuals of the linear model for the flight speed, we detected heteroscedasticity which was most probably caused by the strong variance in the data and potentially also by the unbalanced number of data points between habitat categories. To correct the model, we subsampled the data by randomly selecting 76 data points from each habitat category.

The residuals of the first model for flight altitude indicated temporal autocorrelation as well as heteroscedasticity. We corrected this model by randomly subsampling the data to a number of 40 data points per habitat type and applying a glmmTMB model with a Gaussian error distribution and the factor for habitat type as the dispersion parameter.

Finally, we analyzed whether the attack probability differed between habitat types by modeling the presence and absence of an attack in a generalized model with binomial error distribution. For this model, we were able to use the full amount of data without any subsampling.

The [.csv] file is GPS coordinate data generated from the Acoustic-GPS logger, and the [.kml] file shows flight path from this [.csv] and can be opened using Google Earth. One [.xlsx] is the data of sonar emission timing (JST) measured by the acoustic-GPS data logger. The other [.xlsx] file shows all data of sound characteristics of sonar emission of the bats in this study.  The [.csv] and [.xlsx] files can be opened using Microsoft Excel.