Anthropogenic noise decreases activity and calling behavior in wild mice

Study design

Our field work was conducted from June to August 2016, 2017, and 2018 at 25 focal areas (defined below) at three different sites across the Highlands-Cashiers Land Trust Property (Brushy Face; 35°054′N, 83°188′W), Turtle Pond in the Nantahala National Forest (35°234′N, 83°559′W), and Highlands Biological Station (35°054′N, 83°188′W), in the Southern Appalachian Mountains in Macon County, North Carolina, USA. All sites were > 610 m above sea level (ASL). At each site, we determined resident deer mice and their main areas of activity which we call ‘focal areas’ through standard capture-mark-recapture live trapping. We then used radio telemetry, a foraging tray, full-spectrum audio recording, and a thermal imaging camera to measure behaviors. Our study initially focused only on deer mice, but woodland jumping mice were common at our sites. We were able to measure and report on some of the behavioral responses for both mouse species because woodland jumping mice are easily identified on video based on their jumping gait (see Supplement 1a and Supplement 1b for videos). We used radio telemetry to collect individual activity data from deer mice. Foraging trays were used to collect foraging data and assess risk perception from deer mice and woodland jumping mice, combined. The audio recording was used to collect vocalization data from deer mice and woodland jumping mice, combined and separately. We also used thermal video to determine which species first entered the focal area, first visited the foraging tray, and how long each species stayed in the focal area and at the foraging tray.

All focal areas were in close proximity (less than 20m) to a creek in dense rhododendron (Rhododendron catawbiense) riparian habitat within steep mountain topography. We used capture-mark-recapture live trapping to determine which deer mice were resident in focal areas. The preferred habitat type for deer mice and woodland jumping mice in this region is old growth hemlock forest (Tsuga canadensis) with a dense rhododendron understory. Golden mice (Ochrotomys nuttalli), although not part of our study, were present in the three study sites and were more likely to be captured away from the creek with mountain laurel (Kalmia latifolia) as the preferred understory (unpublished data). Due to microhabitat selection by deer mice and golden mice in this region, we were able to specifically target sections of our study sites to focus on deer mice.

Trap stations across all focal areas were set with 10 meter (m) spacing between stations. At each study site, we were either able to set a grid or transect based on the steepness of the topography and narrowness of the riparian zone. At the Brushy Face study site we were able to set up a grid configuration (50 m × 200 m). Due to the steepness at Turtle Pond and Highlands Biological station we set up transects. At Turtle Pond, trap stations were set along a 650m transect, and at Highlands Biological Station trap stations were set along a 1000m transect. At each trap station, we set two Sherman traps (7.6 cm × 8.9 cm × 22.9 cm) baited with a rolled oat and sunflower seed mixture. On cold nights we also added a small amount of pillow stuffing for bedding. Traps were set an hour before sunset and checked at midnight and reset, and then checked again and closed two hours before sunrise. Upon capture of any mouse species, we recorded age, sex, reproductive condition, and mass (for details see Petric & Kalcounis-Rueppell, 2013)). All mice captured for the first time were marked with a unique numbered ear tag (Monel #5). The majority of mouse captures were of the deer mouse (Peromyscus maniculatus; 49% of individuals captured). We also captured the woodland jumping mouse (Napaeozapus insignis; 28% of individuals captured) and the golden mouse (Ochrotomys nuttalli; 23% of individuals captured). Any species captured that was not a mouse was released immediately.

We identified resident deer mice as adult individuals captured two or more times at the same trap station or the surrounding five trap stations in any direction over five trapping nights. We then selected a resident deer mouse and outfitted that deer mouse with a 0.55 g M1450 mouse style transmitter (Advanced Telemetry System, ATS) as in Petric & Kalcounis-Rueppell (2013). We outfitted and followed one resident deer mouse at a time to determine its focal area. We could only follow one resident deer mouse in its focal area at a time at any given site because of equipment and logistics. Using handheld radio telemetry, we determined the main area of activity for the resident deer mouse, and this was its focal area (approximately 3 m × 5.5 m area across all resident mice). To measure the total time the radio-tagged deer mouse spent in the focal area, we deployed automated radio telemetry equipment (R4500S DCC receiver/datalogger; ATS, Isanti, MN, USA) that continually scanned for the radio signal of the radio-tagged deer mouse and allowed us to measure the time (number of minutes) the deer mouse was in the focal area. The datalogger was connected to an antenna in the center of the focal area and was programmed to detect and record the transmitter frequency of the mouse through the night. The following morning, the datalogger was removed from the field and brought to the lab to download the data and recharge the battery.

In the center of the focal area, we also set up a single seed tray to examine foraging behavior. Each tray had 20 whole sunflower seeds mixed into five cups of sand in a plastic plant saucer (30.48-cm diameter, 5.08-cm deep). The seed/sand mixture was protected from the rain with a clear, 40.64-cm diameter plate that was supported by four stakes and suspended eight cm above the mixture. The following morning, we used a mesh strainer to separate the sand from the leftover seeds/husks. The intact seeds were collected and counted. We also assessed and classified leftover husks into three categories: (1) zero husk pieces, (2) one to five husk pieces and (3) >10 husk pieces remaining after each night. A zero in leftover husks showed that mice collected the seeds but did not take the time to consume the seeds at the foraging tray, indicating that risk perception is high, whereas >10 leftover husks showed that mice took the time to consume the seeds at the foraging tray, indicating that risk perception is low. We could not determine which mouse species had consumed the seeds, so species are combined for this measure.

We recorded the focal area, through the night, using video recordings with a thermal lens (Photon 320 14.25 mm; FLIR/Core & Indigo, Wilsonville, OR, USA). The lens was mounted on a 2-meter tripod to capture the entire focal area in the field of view. The lens was connected to a JVC Everio HDD camcorder and powered by an external dry cell car battery connected to an inverter. Although not the initial focus of this study, we also captured jumping mice who were active in the area. From the video, we were able to identify the species of the mouse, a deer mouse or a woodland jumping mouse, based on gait differences (Supplement 1a and Supplement 1b).

To measure mouse calling in the focal area, we used 12 ultrasonic microphones (Emkay FG Series; Avisoft Bioacoustics, Berlin, Germany) connected to an UltraSoundGate system 1216H (Avisoft Bioacoustics, Berlin, Germany), which was connected to a small laptop (DELL Latitude E6230). The microphones were arranged in a 3x4-meter grid configuration approximately 1 − 2 − m apart with the grid centered in the focal area. All microphones were calibrated prior to deployment. Using Avisoft RECORDER Software, the system was set to record when sonic and ultrasonic sounds were detected by the microphone(s). Recording settings were as in Briggs & Kalcounis-Rueppell (2011) and Petric & Kalcounis-Rueppell (2013). Microphones were triggered and a .wav file was recorded when sounds were detected. Each morning, files were downloaded. All files recorded were examined, and all mouse recording were analyzed and counted, using Avisoft SAS Lab Pro (Avisoft Bioacoustics). Using the time of the recording of the call we matched the vocalization with the time of the activity on the video footage (as in Briggs & Kalcounis-Rueppell, 2011; Petric & Kalcounis-Rueppell, 2013), allowing us to assign the vocalization to species (either a deer mouse or a woodland jumping mouse).

We have described calls produced under laboratory conditions by adult Peromyscus from seven different species, including calls from Peromyscus maniculatus (Kalcounis-Rueppell, Petric & Marler, 2018b). We have also described calls produced by Peromyscus under free living, natural conditions in the wild. In the wild and laboratory, Peromyscus spp. most commonly produce sustained vocalizations (SVs) as summarized in Kalcounis-Rueppell, Pultorak & Marler (2018a) and as reported in Petric & Kalcounis-Rueppell (2013), Briggs & Kalcounis-Rueppell (2011), Timonin, Kalcounis-Rueppell & Marler (2018), Petric, Kalcounis-Rueppell & Marler (2022). The SVs we record are loud and used for conspecific contact and travel 3 to 7m in the wild (Timonin, Kalcounis-Rueppell & Marler, 2018). Our equipment setup was similar to the setup in our previous studies and involved the same equipment and approach to microphone placement (on the ground with a small tripod with a rain guard) and with cables carefully placed and tested prior to each experiment.

Broadcasting noise experiment

We used a “before-during” experimental design where the “before” sampling served as our control. For the “before” behaviors, we recorded the resident mice for three consecutive nights (nights one, two, and three) without any noise manipulations in the focal area. We consider these first three nights without any noise broadcasts as our control. At the same focal area, on nights four, five, and six, we introduced noise manipulation via a broadcast speaker and recorded the resident mice. At each focal area, the noise manipulation was only one of two types: either anthropogenic noise or natural noise. We consider these last three nights with noise broadcast as our treatment and there were two treatment types, anthropogenic noise and natural noise. We recorded for three consecutive nights before broadcasting and for three consecutive nights during broadcasting to make sure that there was no habituation to our equipment, which we would have observed as a night effect in our analysis. Our goal was to capture natural variation in the behaviors we measured, and we were more likely to do that across multiple nights. The broadcasting treatment type (anthropogenic noise or natural noise) was randomly assigned to each focal area. Our anthropogenic noise was pre-recorded from a diesel automatic standby generator (ranging from 1-42 kHz). Our natural noise was pre-recorded from a free-flowing creek in our study area (ranging from 1-25 kHz). We chose these particular sounds because they were present in the study area; streams represent a natural noise that would be regularly encountered by mice in the soundscape whereas the generator would not be regularly encountered but represents a loud anthropogenic noise that is possible within the soundscape.

Anthropogenic and natural noise was recorded using a Zoom H2 Handy Portable Stereo Recorder. The recorder was mounted on a mini tripod or a box between 30 cm–40 cm off the ground 1m from the source. We recorded noise continuously from sunset to sunrise for three nights in .wav format at a sampling rate of 96 kHz/24-bit (see Supplement 2 for spectrograms of exemplar recordings). We used these noise recordings to create files for our noise broadcast experiments. We randomly selected four 15-minute exemplars for each noise type (anthropogenic or natural) from across the three nights and created a single 1-hour sound file by randomly stringing the four 15-minute exemplars together using Avisoft SAS Lab Pro.

For our broadcast experiments, we set up an AT-100 wide bandwidth speaker (Binary Vocal Technology LLC. Tucson, AZ, USA). The speaker was set 2m from the edge of the focal area, and 40 cm off the ground. The 1hr sound file was then broadcasted on a continuous loop for eight hours per night, during nights four, five, and six, at a volume of approximately 80-90 dBA at 1-meter following Oliveira et al. (2009). Each night we adjusted our amplitude to match the source amplitude from our recordings at 1m from the source. Prior to the start of the experiment, we measured amplitude for five minutes at a distance of 1m from the speaker using a calibrated Avisoft USG Electret Ultrasound Microphone and matched amplitude by adjusting the volume on the speaker. The speaker was connected to a small laptop (DELL Latitude E6230) and operated using G’Tools version 1.6 PLAY’R ultrasonic generation software (Binary Vocal Technology LLC, Germantown, MD, USA).

We measured seven response variables to our experiment as follows and for each variable, we calculated the differences between treatment nights (nights four, five, and six) and control nights (nights one, two, and three). We calculated differences, as opposed to using raw data, so that we could examine the effects of noise independent of absolute levels of activity. The difference was calculated within each focal area in the following way: night four minus night one; night five minus night two; night six minus night three. The difference was a measure of the effect of noise allowing for an effect of night, with a zero-difference indicating no effect of noise. We measured the following variables: 

(1) Difference in total time spent in the focal area by deer mice—number of minutes the radio-collared deer mouse spent in the focal area from sunset to sunrise from radio telemetry data.

(2) Difference in total seeds consumed by both species—2a) number of seeds consumed in the foraging tray and 2b) the number of husks leftover at the foraging tray.

(3) Difference in latency to enter the focal area by either species—from video recordings, the number of minutes post-sunset until the first mouse (deer mouse or a woodland jumping mouse) appeared in the focal area.

(4) Difference in time spent in the focal area on the first visit by either species—from video recordings, the number of seconds the first mouse (deer mouse or a woodland jumping mouse) spent in the focal area after first entering the focal area.

(5) Difference in latency to start foraging at the tray by either species—from video recordings, number of minutes post-sunset until the first mouse (deer mouse or a woodland jumping mouse) entered a foraging tray.

(6) Difference in time spent foraging on first appearance at the tray by either species—from video recordings, the number of seconds the first mouse (deer mouse or a woodland jumping mouse) spent in the foraging tray.

(7) Calling behavior of each species—number of calls produced by each species of mouse through audio recordings identified to species (deer mouse or a woodland jumping mouse) through video.

All field work was approved through the following permissions: The University of North Carolina Institutional Animal Care and Use Commitee (IACUC) 16-002, NC Wildlife Resource commission 17-ES00336 and 17-SC00162, North Carolina Park Services 2720, Highlands Biological Station IACUC 16-08, and Highlands-Cashiers Land Trust.

Statistical analyses

Our statistical analyses and data visualization were conducted in R software version 3.2.2 (R Core Team, 2018). We used an alpha level of p < 0.05 for all analyses. Data for all dependent variables were checked for normality using a Shapiro–Wilk test. We used the package MASS to fit Generalized Linear Mixed Models (GLMM) with focal area as the random term, noise type as the fixed term, and night as a covariate to the following variables: difference in total time spent in the focal area, difference in total food consumed, difference in number of husks leftover, difference in total calls produced, difference in latency to enter the focal area, difference in time spent in the focal area on the first visit, difference in latency to start foraging, and difference in time spent foraging. Although we were interested in the effect of noise, and noise type, on behaviors, we were also interested in night effect to determine if mice were habituated to our equipment. Thus, we used night as a sample unit and we account for the repeated samples across nights by including focal area ID in the GLMM as the random term. Due to our small sample size and non-overlapping distribution among dependent variables, we were only able to run one dependent variable at a time. In our constructed generalized linear models, natural noise served as the reference level for noise effects and night four served as the reference for night effects. The generalized linear models were fitted with either a Poisson or Gaussian distribution and reported as GLMM Estimate ±SE. The Poisson distribution does not allow for any negative values therefore, the data were transformed by using the scaling method (adding the absolute smallest number to every value in the array). All GLMM models are listed in Supplement 3. All data are represented using box plots to show skewness and outliers.

We were also interested in whether noise would influence which species was first to arrive at the focal area and feeding tray. We used the Chi-squared test of Independence to examine the relationship between the first species to appear (deer mouse or woodland jumping mouse) and the presence of noise (“before” or “during” the noise broadcast).

Sample size

We tagged 96 deer mice, 54 woodland jumping mice, and 46 golden mice over 464 captures from June to August 2016, 2017, and 2018 at 25 different focal areas across the three years. The value of 464 captures included a large number of recaptures that facilitated our identification of residents. We additionally captured and immediately released the following species: southern flying squirrel (Glaucomys volans), red backed vole (Myodes gapperi), masked shrew (Sorex cinereus) and Northern short-tailed shrew (Blarina brevicauda). We radio-collared and broadcasted sounds in 21 focal areas with radio-collared deer mice (anthropogenic noise=13, natural noise=8). Of our 21 radio-collared mice, 13 were male and 8 were female. We broadcasted noise to an additional 4 focal areas where we were not able to radio-collar the resident deer mouse because we ran out of radio collars, but where we had a focal area with mice (anthropogenic noise=2, natural noise=2). Thus, for all variables that did not require individually radio-tagged deer mice, we have a total of 25 focal areas (anthropogenic noise=15, natural noise=10). The thermal imaging equipment was set-up in all 25 focal areas (anthropogenic noise=15, natural noise=10) but due to the intensive, real time, analysis of thermal video we were only able to analyze a random subset of 10 focal areas (anthropogenic noise=5, natural noise=5), and only the first mouse appearing in the focal area and at the foraging tray. The foraging tray was set-up at all 25 focal areas (anthropogenic noise=15, natural noise=10). Due to raccoon disturbance, we did not collect data from one night at two different anthropogenic broadcasting focal areas. Our final foraging dataset consisted of 148 recording nights from 25 focal areas (anthropogenic noise=15, natural noise=10).

Difference in total time spent at the focal area by deer mice

In the anthropogenic noise focal areas, deer mice spent an average of 64.7 fewer minutes/night during broadcast of anthropogenic noise than during the nights with no broadcast of noise. In the natural noise focal areas, mice spent an average of 5.9 more minutes/night during the broadcasting of natural noise than during the nights with no broadcast of noise (Table 1). The difference in total time spent by deer mice at the focal area was significantly different between anthropogenic and natural noise focal areas (Anthropogenic-GLMM Estimate 0.21 ± 0.08, df = 19, p = 0.0138; Fig. 1; Table 1). Difference in time spent in the focal area was not influenced by night (night five GLMM Estimate −0.05 ± 0.10, df = 40, p = 0.6108; night six GLMM Estimate 0.03 ± 0.09, df = 40, p = 0.7186).

Difference in total seeds consumed by both species

In the anthropogenic noise focal areas, mice consumed an average of 4.7% fewer seeds during the broadcast of anthropogenic noise (Table 2). In the natural noise focal areas, mice consumed an average of 2.5% more seeds during the broadcast of natural noise (Table 2). The difference in the number of seeds consumed did not large significantly differ between natural and anthropogenic noise broadcasts (Anthropogenic-GLMM Estimate 0.07 ± 0.047, df = 23, p = 0.1389) and there was no night effect (night five GLMM Estimate 0.01 ±0.06, df = 46, p = 0.8852; night six GLMM Estimate −0.08  ± 0.06, df = 46, p = 0.1578). However, there was a difference in the number of husks left by mice when anthropogenic noise was broadcast compared to when natural noise was broadcast (Anthropogenic-GLMM Estimate 3.22 ± 0.51, df = 23, p = 0.0000). Mice left husks in the tray during the control nights, and during the broadcast of natural noise, but not during the broadcasts of anthropogenic noise (Fig. 2). There was no night effect for the difference in the number of husks leftover (night five GLMM Estimate −0.10 ± 0.20, df = 46, p = 0.6163; night six GLMM Estimate −0.20 ± 0.20, df = 46, p = 0.2201).

Difference in latency to enter the focal area and time spent in the focal area on the first visit by both species

In the anthropogenic noise focal areas, the average latency for mice to enter the focal area was 18.4 min longer than the control during broadcasting of anthropogenic noise (Table 3). In the natural noise focal areas, the average latency for mice to enter the focal area was 16 min longer than the control during broadcasting of natural noise (Table 3). The difference in latency to enter the focal area between anthropogenic and natural noise focal areas was not significantly different (Anthropogenic-GLMM Estimate −0.02 ± 0.20, df = 8, p = 0.9211), nor was there a night effect (night five GLMM Estimate 0.04 ± 0.23, df = 18, p = 0.8522; night six GLMM Estimate −0.21 ± 0.24, df = 18, p = 0.4014). However, in the anthropogenic noise focal areas but not the natural noise focal areas, the first mouse to appear on the control nights was more likely to be a deer mouse, whereas, during the broadcasting nights the first mouse to appear was more likely to be a woodland jumping mouse (anthropogenic noise χ² =9.5, df = 1, p =0.0179; natural noise χ² =0.14, df = 1, p =0.3524).

In the anthropogenic noise focal areas, the average length of the first bout of mouse activity was 53 s shorter during broadcasting of anthropogenic noise than the control (Table 3). In the natural noise focal areas, the average length of the first bout of mouse activity was 97.7 s longer during broadcasting of natural noise than the control (Table 3). The difference in length of the first bout of mouse activity in the focal area between anthropogenic and natural noise focal areas was not significantly different (Anthropogenic-GLMM Estimate 0.13 ± 0.14, df = 8, p = 0.3580), nor was there a night effect (night five GLMM Estimate 0.16 ± 0.16, df = 18, p = 0.3139; night six GLMM Estimate 0.12 ± 0.16, df = 18, p = 0.4623).

Difference in latency to start foraging and time spent foraging by both species

In the anthropogenic noise focal areas, the average latency for a mouse to start foraging was 62 min longer during broadcasting of anthropogenic noise than the control (Table 3). In the natural noise focal areas, the average latency for a mouse to start foraging was 17.7 min shorter during broadcasting of natural noise than the control (Table 3). The difference in latency to start foraging between anthropogenic and natural noise focal areas was significantly different (Anthropogenic-GLMM Estimate −0.71 ± 0.24, df = 8, p = 0.0183; Fig. 3). There was no effect of night on difference in latency to start foraging (night five GLMM Estimate −0.17 ± 0.23, df = 18, p = 0.4942; night six GLMM Estimate −0.19 ± 0.24, df = 18, p = 0.4325). Furthermore, in the anthropogenic noise focal areas but not the natural noise focal areas, the first mouse to start foraging on control nights was more likely to be a deer mouse, whereas, during the broadcasting treatment nights the first mouse to start foraging was more likely to be a woodland jumping mouse (anthropogenic noise χ² =9.5, df = 1, p =0.0179; natural noise χ² =0.60, df = 1, p = 0.2193).

In the anthropogenic noise focal areas, the average time a mouse spent foraging at the tray was 200.6 s shorter during broadcasting of anthropogenic noise (Table 3). In the natural noise focal areas, the average time a mouse spent foraging at the tray was 22.2 s longer during broadcasting of natural noise (Table 3). There was a trend in the difference in time spent foraging between anthropogenic and natural noise focal areas (Anthropogenic-GLMM Estimate 222.73 ± 97.29, df = 8, p = 0.0553; Fig. 4). There was no effect of night for the difference in time spent foraging at the tray (night five GLMM Estimate −110.30 ± 121.72, df = 18, p = 0.3768; night six GLMM Estimate −43.50 ± 121.72, df = 18, p = 0.7250).

Vocalization production

We recorded 353 mouse calls and assigned 204 calls to deer mice and 149 calls to woodland jumping mice. All calls recorded from deer mice were of the SV type (Kalcounis-Rueppell, Pultorak & Marler, 2018a; Fig. 5A). All calls recorded from woodland jumping mice were of two types (Figs. 5B and 5C) and these are the first reports of ultrasonic vocalizations from this species. Spectral and temporal characteristics of both the deer mouse and the woodland jumping mouse calls will be described in a separate publication but we present measures from 15 randomly selected calls from each species in Supplement 4 to describe basic temporal and spectral parameters of the calls.

We only recorded six calls during the broadcasting treatment nights (anthropogenic noise=2, deer mouse=1, woodland jumping mouse=1; natural noise=4, deer mouse=3 woodland jumping mouse=1) and the remaining 347 were recorded during the control nights before broadcasting (anthropogenic noise=300, deer mouse=161, woodland jumping mouse=133, natural noise=53, deer mouse=39, woodland jumping mouse=14). At one of the anthropogenic noise sites we recorded 183 calls in a single before-broadcasting night, and there was no significant difference with or without the outlier, therefore, we include the outlier.

There was no difference in the number of calls produced between before and during broadcasting nights (Anthropogenic - GLMM Estimate 0.75 ± 0.73, p = 0.3190) and there was no night effect (night five GLMM Estimate 1.74 ± 0.42, p = 0.0860; night six GLMM Estimate −0.12 ± 0.54, p = 0.8181).