Dr. Darren S Proppe
Calvin College, Grand Rapids, MI
I am pleased to announce that Chad Apol, an undergraduate biology student at Calvin College, has been hired to conduct full-time research this summer on the impacts of noise on detectability in acoustic recordings. Chad has spent the semester learning the techniques of bioacoustical analysis and is currently becoming proficient in Kaleidoscope. He has just begun a course in field natural history that will prepare him to identify the birds he will be seeing and hearing this summer. We have familiarized ourselves with the SM4 recordings units, and we have purchased a noise-making sleep machine that will be used to introduce noise to our recordings on a very limited spatial scale. We are in the process of finalizing field sites for our experiments. We will use at least 10 abandoned oil pads located in Northern Michigan. They provide both a forested and open ecosystem for our work. We intend to test varying levels of white, pink, and Brownian noise; comparing Kaleidoscope's ability to detect and appropriately cluster bird vocalizations in comparison to quiet controls. Recording will begin at the end of May and continue throughout the summer. No data yet, but we are itching to get started. More soon!
Recorders are in the field! We are up to 16 complete trials, each containing one microphone exposed to noise playback, and a control microphone that is unexposed. Noise playback comes in the form of white, pink, and brownian played at 40, 50, 60, and 70 dBA. We have trained Kaleidoscope to detect five species commonly found in our recordings: blue jay (Cyanocitta cristata), ovenbird (Seiurus aurocapilla), red-eyed vireo (Vireo olivaceus), Eastern wood-pewee (Contopus virens), and black-throated green warbler (Setophaga virens). The first results are coming in, but it's too early to describe any patterns. We've also decided to add a human detection component. Chad will be visually detecting vocalizations in a subset of our recordings to compare human detection rates in varying noise levels to the capabilities of Kaleidoscope. Stay tuned...
Kaleidoscope analysis is well underway, with nearly half of the data from our sites having been compiled. We are already seeing significant trends for a number of parameters. One of the expected, yet interesting, preliminary results is that Kaleidoscope software has been more successful in correctly detecting vocalizations from a control SM4 unit compared to a SM4 unit subjected to noise input (see Figure). We will be completing data analysis soon and are looking forward to reporting additional trends related to noise level and noise type.
Human development can introduce significant amounts of noise pollution into the environment, often greatly exceeding the amplitude of natural ambient noise. Anthropogenic noise has been shown to negatively impact the reproduction of certain bird species (Kight et al 2012), change the vocalizations and behavior of others (Francis et al 2011), and decrease the detectability of biotic vocalizations in birds (Leonard et al. 2015) and humans (Koper et al. 2016). The detectability of biotic vocalizations is an integral aspect of avian population and community research, which often consists of surveys that locate birds through the identification of songs and calls. The use of passive acoustic recorders, such as those produced by Wildlife Acoustics, has increased dramatically in recent years, enhancing our ability to collect large acoustic datasets on avian vocal behavior. However, an increasing number of acoustic studies now occur in urban and suburban areas where anthropogenic noise is prevalent. Although increased noise levels would be expected to mask vocalizations and reduce their detectability, the extent to which this impacts acoustic detection in passive acoustic recorders is relatively unknown. Further, noise varies in frequency and amplitude, and minimal information is available on how these nuances impact detectability. We tested whether increasing the amplitude of three different types of ambient noise impacted the detectability of vocalizations in five bird species.
We placed two SM4 passive acoustic recorders (Wildlife Acoustics, Inc.) in 20 remote hardwood forests in Northern Michigan, USA. Apple earbuds were placed on one microphone of one SM4 unit, broadcasting noise tracks in 5 minute increments. Tracks included a control with no noise, and three different types of noise which vary in their spectral characteristics (brown, pink and white). Each noise type was played at amplitude 40, 50, 60 and 70 dB(A). The opposing microphone on the same unit was used as a within unit control, and a microphone on a second unit that was placed 3m away along the same azimuth was used as a between unit control. The results from the two controls did not differ for any treatment, therefore, only the between unit control was retained. Kaleidoscope Pro detection software was pre-trained using commercially available field recordings of red-eyed vireos, blue jays, black-throated green warblers, ovenbirds, and wood-pewees because these species were common at our sites.
The number of detections made by Kaleidoscope was recorded for each site, track, and species - false detections were visually inspected and removed. Statistics were carried out in program R (V3.3.3). There was a significant overall difference in the mean number of detections between the control and noise treatments at amplitudes greater than 50 dB (Figure 1), with the control microphone detecting significantly more vocalizations than the noise treatment during the same timeframe. A poisson regression model was fitted to determine whether the impacts on detection differed by noise type (Figure 2). While detection decreased with amplitude for all noise types, each was impacted differently, with white noise being least impacted and pink being most impacted. Each species was also impacted differently (Figure 3a & 3b), although none were exempt from the masking effects of noise. The red-eyed vireo is graphed separately because the number of correct detections was substantially higher than the other four species, likely because of its propensity to vocalizing continuously.
Our results reveal that ambient noise levels â¥ 50 dB can significantly impact the detectability of bird vocalizations, while ambient noise levels â¥ 70 dB may eliminate almost all detections. Although not significant, a drop in detection rate is also visible at 40 dB. Our models show that white noise, which spreads acoustic energy across all frequencies equally, impacted vocal detection less than brownian or pink noise, which concentrate more energy in the lower frequencies. This may be due to the uniform background produced by white noise, which enables easier detection of energy bursts, such as is found in bird song. However, anthropogenic noise sources tend to be concentrated in the lower frequencies, more similar to pink or brownian noise. The impacts of noise varied somewhat by species, but none were exempt from the masking effects of noise. Further work is needed to determine whether noise filters, or visual screen counts can improve the results from noise-impacted data collected from passive acoustic recorders. Nonetheless, our results suggest that caution is needed when using passive acoustic recorders in noisy areas, especially if comparisons are to be made with quiet regions.
Figure 1: Mean of correct detections from noise and control treatments recorded over the same timeframe. Error bars represent standard error.
Figure 2: Predictions from the fitted model of correct detections for three different noise frequencies (brownian, pink, and white) with increasing amplitude of noise playback.
Figure 3: Predictions from the fitted model of correct detections for five different bird species with increasing amplitude of noise playback. Species inlclude' A) blue jay (BLJA), black-throated green warbler (BTNW), ovenbird (OVEN), Eastern wood-pewee (EAWP), and B) red-eyed vireo (REVI)
Kight, C. R., Saha, M. S., & Swaddle, J. P. (2012). Anthropogenic noise is associated with reductions in the productivity of breeding Eastern Bluebirds (Sialia sialis). Ecological Applications, 22(7), 1989-1996.
Koper, N., Leston, L., Baker, T. M., Curry, C., & Rosa, P. (2016). Effects of ambient noise on detectability and localization of avian songs and tones by observers in grasslands. Ecology and evolution, 6(1), 245-255.
Leonard, M. L., Horn, A. G., Oswald, K. N., & McIntyre, E. (2015). Effect of ambient noise on parent-offspring interactions in tree swallows. Animal behaviour, 109, 1-7.
Francis, C. D., Ortega, C. P., & Cruz, A. (2011). Vocal frequency change reflects different responses to anthropogenic noise in two suboscine tyrant flycatchers. Proceedings of the Royal Society B, 278(1714), 2025-2031.