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

These SPL values were plotted versus date and time at each acoustic station (i.e. 9M, 14M, and 37M) with corresponding water temperature and daylight hours. We performed invertebrate and fish sampling one to two times per month in the May River between 2016 and 2018 using a haul seine (i.e. seine width = 9.1 m, height = 1.2 m, and mesh diameter = 3 mm) and block nets (i.e. additional stationary seine nets to stop animals from escaping) at two to four locations per passive acoustic station (Fig 5). This sampling equated to six to twelve seines per month selected randomly from a list of sites. Seine sites included tidal pools (i.e. shallow pools of water created on the low tide), intertidal creeks (i.e. small secondary or tertiary creeks feeding from the main river accessible on the low tide), and sides of the river (i.e. stations located along the bank of the primary river). Before each seine, we recorded envi- ronmental parameters (i.e. water temperature, salinity, pH, and dissolved oxygen) using a YSI 556 Handheld Multiparameter Instrument (YSI Inc./Xylem Inc., Yellow Springs, OH, USA), and then we seined the area between the block nets. We measured the length and width of each seine for an area calculation. We transferred the catch into a live well, quantified the abundance of each species, and then released all the organisms at the original sampling loca- tion. Mortality of the organisms was low during the wintertime (< 5%) and moderate during summertime (~5 to 50%). We calculated average species richness per month based on the

number of species in each seine standardized by the seine area. In addition, we calculated the average Shannon-Wiener diversity index divided by the area per month using package “vegan” in R version 3.4.2 (R Core Team, 2012) [39]. We calculated mean species abundance per month based on the abundance of all the species in each seine standardized by the seine area. This work was conducted under South Carolina Department of Natural Resources permit numbers 5135 and 5136 and IACUC protocol 2233-101181-022217. Statistical analyses To assess the significance of specific factors in explaining variations in SPLs, we used package “Boruta”, a wrapper algorithm based on the random forest algorithm in R [40–49]. Random forest models are non-parametric and do not require formal distribution assumptions. We used “permutation importance” rather than the default “mean decrease in impurity impor- tance” to assess the importance of factors included in the model, since permutation impor- tance is less biased of continuous and categorical variables with many levels [41–44]. In the final model, we set the specific parameters to: p = 0.01, mtry = 2, ntree = 200, nodesize = 5, and set.seed = 42 [45–47]. We included location, year, lunar phase, tide, day/night, temperature, day length, and rainfall as factors. We used four categories to differentiate the lunar and tidal cycle following methods previously described, and we used National Oceanic and Atmo- spheric Administration (NOAA) weather stations located close to the May River to obtain

rainfall data for each day (S6 Table) [28]. Before applying the model, we tested the data for col- linearity. Temperature, rainfall, and day length exhibited multi-collinearity that could bias the Boruta feature selection algorithm [47]. Hence, we created three different models for high, low, and broadband frequency SPL. In the first set of models, we included temperature as a factor, in the second set, we used day length, and in the third set, we used rainfall. Then, we compared the R2 for each model that used a different factor, and we reported the models that best explained the variability of the data. We removed files that contained physical sounds and anthropogenic noise, and we followed the same approach to test variable importance. Then, Fig 5. Map of the May River, SC, USA. Locations of stations 9M, 14M, and 37M that were acoustically monitored from February 2013 to December 2018 (blue) and seining stations monitored from May 2016 to December 2018 (yellow). (Inset) Location of the May River (black) in reference to the USA coast. https://doi.org/10.1371/journal.pone.0236874.g005 PLOS ONE Estuarine soundscapes and biodiversity PLOS ONE | https://doi.org/10.1371/journal.pone.0236874 September 3, 2020 11 / 18