A meta-analysis of the influence of anthropogenic noise on terrestrial wildlife communication strategies
Data files
Mar 26, 2021 version files 27.08 KB
-
DryadScript.R
-
README.txt
-
soundbaser2.csv
Abstract
Methods
Literature Search Strategy and Inclusion Criteria
We searched the peer-reviewed scientific literature to synthesize information regarding noise pollution impacts on wildlife acoustic communication and to assess research gaps and biases. We restricted the search to terrestrial systems because general approaches to noise pollution risk assessment and recommendations for noise mitigation already exist for some coastal and marine systems (Southall et al. 2007). Perhaps more importantly, a vast body of research conducted to date on marine wildlife has yielded valuable knowledge such as species-specific spectral sensitivity, critical impact thresholds, and mitigation effectiveness which can be drawn upon to advance general theory and research and to develop further regulatory guidelines (Erbe et al. 2016). Finally, the physics of sound transmission differ between water and air, affecting both how sound is perceived by organisms and potential mitigation strategies (Würsig et al. 2000, Shannon et al. 2015). We used Web of Science (search conducted 4/5/2018) to search for studies investigating the impact of noise pollution on wildlife modulation of call frequency, rate, duration, and amplitude (see Table 2 for specific search terms). We assessed these multiple communication response variables even though they may be related because each response may have different ecological and/or evolutionary implications. An initial search produced 815 studies. After implementing all inclusion criteria (see below), our search resulted in 181 data points from 32 studies representing six continents (Table 3).
We used the “Analyze Results” feature in Web of Science to filter out irrelevant disciplines (e.g., Audiology, Speech Pathology, nexcluded = 347). After compiling remaining results into a database, we removed duplicate studies (nexcluded = 5) and studies determined to be topically irrelevant based on reading of all titles (nexcluded = 117). We excluded studies broadcasting white noise as a treatment, as we were interested in responses to spectral characteristics that more closely match environmental noise pollution (i.e., loud, low-frequency sounds, nexcluded = 3). However, we retained one study that explicitly manipulated the characteristics of white noise to approximate low-frequency traffic sounds. We excluded studies conducted in a laboratory setting, as we were only interested in responses of free-living wildlife to noises experienced in their natural habitat (nexcluded = 5). After detailed screening of article texts, we removed studies that did not assess effects of noise pollution on the above focal response variables and studies with analysis methods or reporting that precluded us from extracting a relevant effect size (nexcluded = 59).
For remaining studies, we extracted the location, focal taxa, response variable, sound source, and study design. We also extracted means, sample sizes, and standard deviations of response variables for studies assessing categorical predictor variables (e.g., call characteristics at quiet and noisy sites), or values of Pearson’s r for studies assessing continuous predictor variables (e.g., response characteristics over a gradient of decibel levels). In studies with multiple treatments, we used the two extreme ends of the environmental sound spectrum for analysis. For example, if a study tested call rates in “quiet”, “moderate”, and “loud” environments, we compared responses between “quiet” and “loud” sites. Sound sources included airplane (n = 2), construction (n = 6), energy development (n = 17), roadway (n = 52), urban (n = 101), and white noise (n = 3). We also distinguished study designs as event-based (n = 41) versus continuous (n = 140). Event-based study designs evaluated instantaneous signal flexibility in the presence of anthropogenic sound (e.g., a grasshopper calling more loudly during an airplane overflight compared to normal conditions, Fig. 2). Continuous study designs, on the other hand, evaluated differences in acoustic properties between populations in loud and quiet environments (e.g., communication characteristics of red-winged blackbirds, (Agelaius phoenicus), in rural versus urban environments; Fig. 2). Following our literature search, we incorporated a specific search for bat studies, as they were underrepresented in our initial search and we felt that they are good models for the study of anthropogenic sound impacts due to their reliance on acoustic information for both communication and foraging.
Analysis
To assess potential biases in the noise pollution literature, we assessed observed versus expected proportions of studies using Pearson’s χ2 tests. We conducted these tests to analyze numbers of studies for each response variable, sound source, focal taxa, continent, and study design; in each case we tested a null hypothesis that an equal proportion of studies have been conducted for each category (e.g., 50% of studies each for event-based and continuous study designs). To control the Type I error rate, we employed a Holm’s Sequential Bonferroni correction.
We conducted a meta-analysis to assess wildlife responses to noise pollution using the metafor package (Viechtbauer, 2010) in the R statistical environment (version 3.4.1, R Core Team 2017). We ran mixed-effects meta regression models with study design (event-based versus continuous), and taxa as fixed effects and study ID as a random effect.
When possible, we calculated Hedge’s g for each study that used a categorical noise treatment. When studies evaluated responses to noise along a continuous gradient, we calculated Hedge’s g using Pearson’s r. To evaluate overall effect of each response variable (Minimum Frequency, Maximum Frequency, Peak Frequency, Duration, Rate, and Amplitude), as well as the effect of study type and taxa, we evaluated overlap of 95% confidence intervals with zero. After conducting analyses, we constructed Q-Q plots to visually assess model fit.