What’s all that racket! Soundscapes, phenology, and biodiversity in estuaries

we compared the results of the random forest models applied to the two data sets (i.e. one data set that included biological sounds, physical sounds, and anthropogenic noise, and the second data set that included only biological sounds). If categorical variables were significant, we applied Dunnett-Tukey-Kramer pairwise multiple comparison tests adjusted for unequal vari- ances and unequal sample sizes with 95% confidence levels using package “DTK” in R to deter- mine whether group means were significantly different from each other [50]. To identify significant changes in high, low, and broadband SPLs values during the spring- time, we applied a Bayesian change point analysis using package “bcp” in R [51, 52] to the data- set that excluded all physical sounds and anthropogenic noise. Bayesian change point analysis calculates the posterior probability (PP) for any given point in the time series that has an abrupt change. We defined a probability as significant when the change between the following and previous data point was � 50% [8]. For each year, we reported the mean spring water tem- perature, date and day of year of the first significant PP, day anomaly (i.e. day of first signifi- cant PP for each year minus the six year average of days with the first PP), and the value of the first significant PP at each station. We defined spring as the time between the astronomical vernal equinox and the summer solstice. We tested for normality by investigating the distribution of the residuals, and center and

dispersion of monthly averages of species richness, or Shannon-Wiener diversity index, and species abundance. We performed Pearson’s correlations between monthly averages of species richness (or Shannon-Wiener diversity index and species abundance) and monthly averages of high, low, and broadband frequency SPLs for all stations combined. In addition, using linear regression, we tested the relationship between monthly averages of temperature as the inde- pendent variable and corresponding monthly averages of species richness (or Shannon-Wie- ner diversity index and species abundance) as the dependent variable. Correlations and regressions were performed in MATLAB using two-sided hypothesis tests with a significance level of 0.05. Supporting information S1 Fig. Time series of broadband frequency Sound Pressure Levels (SPLs) from 2013 until 2018. Heat maps represent temporal and spatial patterns of broadband (i.e. 1–40,000 Hz) fre- quency SPLs reflecting all physical sounds, biological sounds, and anthropogenic noise at sta- tions (A) 9M, (B) 14M, and (C) 37M in the May River. Time is shown between noon and noon of the next day. Gaps in data = white, temperature = black line, and daylight hours = dotted line. Green dots indicate first posterior probability (PP) � 0.5 detected during springtime. At station (C) 37M first PP was not calculated for spring 2017 due to missing acoustic data. (TIF) S2 Fig. Time series of high frequency Sound Pressure Levels (SPLs) from 2013 until 2018 with physical sounds and anthropogenic noise files removed. Heat maps represent temporal

and spatial patterns of high (i.e. 7000–40,000 Hz) frequency SPLs reflecting snapping shrimp acoustic activity at stations (A) 9M, (B) 14M, and (C) 37M in the May River. Time is shown between noon and noon of the next day. Gaps in data = gray, files with physical sounds and anthropogenic noise removed = white, temperature = black line, and daylight hours = dotted line. Green dots indicate first posterior probability (PP) � 0.5 detected during springtime. At station (C) 37M first PP was not calculated for spring 2017 due to missing acoustic data. (TIF) S3 Fig. Time series of low frequency Sound Pressure Levels (SPLs) from 2013 until 2018 with physical sounds and anthropogenic noise files removed. Heat maps represent temporal PLOS ONE Estuarine soundscapes and biodiversity PLOS ONE | https://doi.org/10.1371/journal.pone.0236874 September 3, 2020 12 / 18