Singing silver-haired bats (Lasionycteris noctivagans)

Methods from the manuscript: Study Species In Canada, silver‐haired bats are considered migratory, moving south for winter months (Naughton 2012) however, in British Columbia (BC) and parts of the northwestern U.S. (including Washington, Idaho and Montana), silver‐ haired bats are recorded year‐round, flying in winter during hibernal arousals (Schowalter et al. 1978, Falxa 2007, Lausen et al. 2022). Banding records provide evidence that at least some silver‐haired bats reside year‐round at some mines in British Columbia (Lausen et al. 2022). Study Area Our recordings spanned several U.S. states and areas of BC. We recorded bats at 23 acoustic detector sites across western U.S. and Canada, each with varying degrees of forested/rocky terrain (Table S1, available in Supporting Information; Figure 1): California (4), Colorado (1), Idaho (3), Utah (2), Montana (2), Washington (one main active monitoring area), and British Columbia (BC, 10). At the Washington site, overwintering roosts of silver‐haired bats were observed in bat boxes, under Douglas‐fir and Western red cedar tree bark, and in small crevices on building exteriors. At 20 of the stationary recording sites, we had no knowledge of whether the sites were used by silver‐ haired bats for roosting. Two of the 10 monitored sites in BC were mine sites in forested areas where silver‐haired bats hibernate and have been documented year‐round (CA‐9, CA‐10; Figure 1, Table 1, S1). The CA‐10 mine is an inaccessible and deep abandoned mine complex with many large (>16 m2) openings and a central pit roughly 100 m in diameter. The CA‐9 mine is an accessible (but gated) shallow (~50 m) mine with 2 large (>20 m2) openings immediately adjacent to each other. Silver‐haired bats use the CA‐9 mine in both summer and winter and have been radiotracked to day roosts (summer and winter) in trees surrounding the mine (Lausen et al. 2022). Mist‐net capture in winter of silver‐haired bats flying in or out of each mine also facilitated examination of bats to determine sex, signs of breeding, etc. (Lausen et al. 2022). CaptureUpon first recording these songs during the 2011 season, we were uncertain whether they were produced by big brown or silver‐haired bats, because the echolocation calls that preceded or followed song phrases were ambiguous and could be attributed to either species (Betts 1998). We therefore conducted capture inventories at both mines, mist‐netting and harp‐trapping bats flying in and out of the entrances to identify species. To understand potential reasons for winter flight and song production, we also examined genitalia to assess any signs of mating activities. Acoustic recordingWe recorded bats passively using Anabat (models SD1 and SD2, Titley Scientific, city, Australia), SM2Bat (model Plus, Wildlife Acoustics, Maynard, MA, USA), Anabat Swift (Titley Scientific, Australia), SM4Bat (Wildlife Acoustics, primarily in a lowland mixed‐conifer wooded park (Squaxin Park) in Olympia, WA (Figure 1). Bats were recorded primarily near forest edge adjacent to water features, roads, or other open nonforested landscapes in all seasons, and in particular active recordings were typically in November, when silver‐ aired bats were the only non‐Myotis bats recorded (Falxa 2007). These active recordings were associated with silver‐haired bats by observing a bat's flight pattern and foraging behaviors; during these recordings, bats producing low frequencies (low‐frequency bats) were observed flying in slow, straight lines over open areas and along canopy edges while echolocation calls and songs of silver‐haired bat were recorded. Some song‐containing files were recorded in close proximity (1–50 m) to known silver‐haired bat roosts. All passive bat detectors were programmed to trigger on bat ultrasound and record up to a maximum duration of 15 s before beginning a new sound file. Adhering to standardized acoustic terminology (Loeb et al. 2015), we refer to single file (recording) as a bat pass, and we define a call as a single burst (pulse) of sound, whether it is echolocation or social in nature. A sequence of calls is any series of pulses, whether they are echolocation or social. To differentiate pulses produced in songs versus those typical of echolocation, we refer to pulses making up a song as syllables. Different types of syllables vary in frequency and duration parameters. A distinct sequence consisting of different types of syllables which is repeated to produce a song, we refer to as a phrase. Generally, a song will consist of one or more phrases, however, if only part of a phrase was recorded, as long as the syllables present are clearly attributable to a phrase, then we refer to this as a partial song. In some cases, such as when a bat is only in the detection volume of a bat detector for a very short period of time, only partial songs (i.e., less than one complete phrase) were recorded. We processed acoustic recordings using a variety of methods, with larger datasets first being processed with either Kaleidoscope Pro (Wildlife Acoustics, MA, USA) or SonoBat (California State Polytechnic University, Humboldt, CA, USA) for auto-identification prior to manually vetting; some smaller datasets were examined without auto‐ identification as a first step. We manually vetted files using either Kaleidoscope Pro, SonoBat, AnalookW (C. Corben, hoarybat.com, MO) or Anabat Insight (Titley Scientific, Brisbane, Australia). Recordings from BC were largely zero-crossing format, and all other datasets were recorded in full spectrum format. British Columbia datasets were manually vetted using a combination of AnalookW and Anabat Insight. Washington and Mendocino County California datasets were manually reviewed using Kaleidoscope Pro. Recordings collected in Montana, Idaho, Utah, Colorado, and Lassen Volcanic National Park (California) were manually reviewed using SonoBat. Active recordings from Washington were reviewed by one of us (G. Falxa) manually to identify songs using both Sonobat3 and Kaleidoscope Pro. Files were manually analyzed in real‐time (uncompressed) mode in order to ensure visualization of a song pattern; compressed mode can obscure the diagnostic song pattern if some song pulses are of low intensity. For larger datasets that underwent automated classification, we determined that silver‐haired bat songs were often misclassified as big brown bat, Brazilian free‐tailed bat, pallid bat (Antrozous pallidus), Townsend's big‐eared bat (Corynorhinus townsendii), western red bat (Lasiurus blossevillii), hoary bat, or long‐eared myotis (Myotis evotis). Thus, files automatically classified as any of the above species were manually vetted to search for silver‐haired bat songs. To avoid reporting extraneous results, we present silver‐haired bat activity only in the context of low‐ frequency bat passes (minimum frequency <30 kilohertz [kHz], where one recorded file = pass), as silver‐haired bat songs are most likely to be confused with echolocation calls of low‐ frequency bat species. We refer to call parameters (measured in AnalookW, Anabat Insight and/or Kaleidoscope Pro) as follows: call body is the flattest part of the call (Figure 2A), with Sc being the slope of that call body in octaves per second (OPS); time between calls (TBC) is the amount of time between pulses/calls; and minimum (Fmin)/maximum (Fmax) frequencies refer to the lowest/highest frequency produced in one call/pulse. Generally, echolocation pulses of North American bats (low duty cycle) are frequency modulated (FM) and thus start high in frequency and end lower, having bandwidth greater than zero (Russo et al. 2018). If the call body is flat or nearly flat, bandwidth approaches zero and this is referred to as a quasi‐constant frequency (QCF) component of a call. The rate of change of frequencies can vary as the call is produced (i.e., the slope changes over time) and may be gradual or sudden. In the latter, this change of slope creates a bend in the call, often called a knee (Figure 2A). The slope of the call from its start at Fmax, to the knee, is measured as S1 (OPS), and the slope from the knee to the Fmin, is generally the call body and is the Sc as discussed above. We measured call duration (time from start to end of a pulse, measured in milliseconds [ms]). All parameters were measured in zero‐crossing format using AnalookW. Spectrograms in the figures are shown in True Time (x‐axis shows the real time elapsed) unless specified (x‐ axis in Compressed Time shows only the time within each pulse in real time, with the time between the pulses largely removed, for display purposes). All spectrograms display a logarithmic frequency (y axis). To assess the auto‐identification treatment of songs, we processed all full spectrum song files using both Kaleidoscope Pro (classifier version North America 5.4.0) and SonoBat (versions 4.4.0 and 4.4.5 Great Basin classifier) including the following species (Kaleidoscope Pro): big brown bat; Brazilian free‐tailed bat; California myotis, Myotis californicus; canyon bat, Parastrellus hesperus; fringed myotis, M. thysanodes; hoary bat; little brown myotis, M. lucifugus; long‐eared myotis; long‐legged myotis, M. volans; pallid bat; silver‐ haired bat; spotted bat, Euderma maculatum; Townsend's big‐eared bat; western red bat; western small‐footed myotis, M. ciliolabrum; and Yuma myotis, M. yumanensis. The Montana classifier set included eastern red bat (Lasiurus borealis) instead of western red bat, and additionally included northern myotis (M. septentrionalis). We used default settings for each software package (Kaleidoscope Pro Balanced; SonoBat 0.7 call quality, 0.9 sequence decision threshold). We examined each recording to ensure the software was triggering on the song pulses, and thus considering them in the auto‐identification process. We calculated the percentage of misclassifications.