Supplementary Code 1:

Data Pre-processing

 

We selected calls in Raven prior to running this code. Here we preprocessed calls from the repeated individual and site-level recordings prior to sound analysis and matrix regression. In this code and other supplementary methods, “focal individual(s)” is sometimes used to refer to repeatedly sampled individuals.

The data provided with this publication was pre-processed using the processing pipeline and code documented below.

Clean environment and install and/or load packages, including the package warbleR^[1].

# clean global environment
rm(list = ls())

###################################
# install packages from Github
# install devtools if not installed
# if (!"devtools" %in% installed.packages()[,"Package"])
# install.packages("devtools")

# library(devtools)
# devtools::install_github("maRce10/warbleR")

###################################
# install packages from CRAN in a loop
# X <- c("pbapply", "ggplot2", "data.table", "parallel")

# is_installed <- function(p) is.element(p, installed.packages()[,1])
#
# lapply(1:length(X), function(x){
#   if(!is_installed(X[x])){
#     install.packages(X[x], repos = "http://lib.stat.cmu.edu/R/CRAN")
#   }
# })

###################################
# load all packages
X <- c("warbleR", "Rraven", "tuneR", "pbapply", "ggplot2", "data.table", "parallel")

invisible(lapply(X, library, character.only = TRUE))

Repeatedly Sampled Individuals

path <- "/media/owner/MYIOPSITTA/R/Uruguay2017_MonkParakeet_CallAnalysis/Data/Focal_Individuals"

We imported Raven selection tables for focal individual calls and wrote out the selection table.

fsels <- Rraven::imp_raven(path, name.from.file = TRUE, ext.case = "upper", freq.cols = FALSE)
str(fsels)

# are all wav files in selection tables also in the working directory?
wavs <- list.files(pattern = ".wav$", ignore.case = TRUE)
unique(fsels$sound.files[!fsels$sound.files %in% wavs])

write.csv(fsels, file.path(path, "focal_individual_sels.csv"), row.names = FALSE)

Here we optimized spectrogram parameters. We visually evaluated the resulting spectrogram mosaic to choose the Fourier Transform window, window length and the partitioning of signal amplitude.

# pick a random call to generate a spectrogram mosaic with varying parameters
set.seed(89)
r <- sample(x = nrow(fsels), size = 1)
spec_param(fsels[r, ], length.out = 10, ovlp = 90, wl = c(200, 600), wn = c("hanning", "bartlett", "hamming"),  collev.min = c(-40, -100), pal = "reverse.gray.colors.2", rm.axes = TRUE, nrow = 3, ncol = 3, img.prefix = "spec_param", flim = c(0.5, 9), mar = 0.03, cex = 0.45, path = path)

We used the parameters chosen above to generate spectrograms for visual quality control. We added information to selection table spreadsheet about visual call quality, overlapping signals and any other pertintent information during visual inspection.

specreator(fsels, wn = "hanning", wl = 378, collev = seq(-53, 0, 1), flim = c(0, 10), ovlp = 90, mar = 0.01, res = 300, cex = 0.5, line = FALSE, path = path)

We manually added columns (Visual_quality_score, Overlapping_signal Overlap_section, Bandpass_overlaps_signal_freq_range) to the metadata during visual inspection of spectrograms. We read back in the metadata and added site and individual identity to the focal individual call metadata.

# read back in the csv with manually added visual quality control columns
fsels <- read.csv(file.path(path, "focal_individual_sels.csv"))
str(fsels)

# add a site variable
fsels$site <- sapply(1:nrow(fsels), function(x){
  strsplit(strsplit(as.character(fsels$sound.files[x]), split = "_")[[1]][4], split = "-")[[1]][1]
})
unique(fsels$site)

# make a column of individual IDs
fsels$indiv <- sapply(1:nrow(fsels), function(x){
  strsplit(strsplit(as.character(fsels$sound.files[x]), split = "_")[[1]][5], split = "DIV-")[[1]][2]
})

# replace long IDs for a 3 character ID per individual
fsels$indiv <- gsub("UNMARKED-BIRD", "UM", fsels$indiv)
unique(fsels$indiv)

How many calls per individual, and which individuals have < 4 calls?

table(fsels$indiv)
table(fsels$indiv)[table(fsels$indiv) < 4]

We filtered out any calls flagged for overlapping signals and individuals with < 4 calls.

fsels <- fsels[fsels$Overlapping_signal != "Y", ]

fsels <- fsels[-grep(paste(names(table(fsels$indiv)[table(fsels$indiv) < 4]), collapse = "|"), fsels$indiv), ]
str(fsels)

We tailored temporal coordinates of selections to standardize how calls were selected (a single observer, GSV, performed tailoring).

seltailor(fsels, flim = c(0, 9), mar = 0.15, osci = TRUE, auto.next = TRUE, path = path)

# read the seltailor output back in for more preprocessing
fselst <- read.csv(file.path(path, "seltailor_output.csv"), header = TRUE)
str(fselst)

We filtered out calls with SNR < 7, and decided on a frequency bandpass filter of 0.5 - 9kHz after visual inspection of spectrograms.

# select a good margin for SNR calculations
# a margin of 0.05 seconds looks good - does not overlap neighboring calls for any of the selected calls
snrspecs(fselst, osci = TRUE, snrmar = 0.05, wn = "hanning", wl = 378, flim = c(0, 10), ovlp = 90, path = path)

# calculate SNR, using bandpass determined during visual quality control above
fselst_snr <- sig2noise(X = fselst, mar = 0.05, type = 1, bp = c(0.5, 9), wl = 378, path = path)
# str(fselst_snr)

# remove calls with SNR < 7
length(which(fselst_snr$SNR < 7)) # only 2 calls will be removed
fselst_snr <- fselst_snr[-which(fselst_snr$SNR < 7), ]
# str(fselst_snr)
# nrow(fselst_snr)

# remove any calls that were flagged as low quality in visual inspections
fselst_snr <- fselst_snr[fselst_snr$Visual_quality_score != "L", ]
# str(fselst_snr)
# nrow(fselst_snr)

We saved the preprocessed metadata and calls.

# save preprocessed selection table as a csv file
write.csv(fselst_snr, file.path(path, "indiv_scale_calls_preprocessed.csv"), row.names = FALSE)

Make an extended selection table of repeatedly sampled individual calls.

fselst_snr <- read.csv(file.path(path, "indiv_scale_calls_preprocessed.csv"), header = TRUE)

cores <- parallel::detectCores() - 2
indiv_sels_tbl <- selection_table(X = fselst_snr, whole.recs = FALSE, extended = TRUE, confirm.extended = TRUE, mar = 0.1, parallel = cores, path = path)

# save the extended selection table as an RDS object
# this object will be provided in Supplementary Material as a means to share the focal individual calls and associated selection table with metadata
saveRDS(indiv_sels_tbl, file.path(path, "indiv_scale_extended_seltable.RDS"))

How to access different components of an extended selection table:

indiv_sels_tbl <- readRDS(file.path(path, "indiv_scale_extended_seltable.RDS"))
str(indiv_sels_tbl)

# access the selection table (with metadata) data frame
dim(indiv_sels_tbl) # 97 calls, 14 columns
names(indiv_sels_tbl) # column names

# access the wave objects
length(attr(indiv_sels_tbl, "wave.objects"))
attr(indiv_sels_tbl, "wave.objects")[1]

# you can write these out as wave objects in your working directory to facilitate reproducing analyses:
wavs <- attr(indiv_sels_tbl, "wave.objects")
wav_nms <- names(attr(indiv_sels_tbl, "wave.objects"))

# x <- 1 testing
invisible(lapply(1:length(wavs), function(x){
  writeWave(wavs[[x]], filename = wav_nms[x], extensible = FALSE)
}))

# access the check results data frame
attr(indiv_sels_tbl, "check.results")

We used the pre-processed focal individual calls to make call lexicons per individual.

i <- unique(fselst_snr$indiv)

invisible(lapply(1:length(i), function(x){
  catalog(fselst_snr[fselst_snr$indiv == i[x], ], wn = "hanning", wl = 378, collev = seq(-53, 0, 1), flim = c(0, 10), ovlp = 90, mar = 0.01, labels = c("sound.files", "selec"), tags = c("selec"), tag.pal = list(topo.colors, heat.colors), res = 300, img.suffix = paste("INDIV", i[x], sep = "-"), cex = 0.5, path = path)
}))

catalog2pdf(by.img.suffix = TRUE, keep.img = FALSE, path = path)

Visual inspection of catalogs revealed that monk parakeets produce consistent and distinctive calls (e.g. high acoustic similarity within individuals), such that contact call structure appeared to be characterized by individual signatures.

Higher Social Scales

We encoded duplicate recording sessions per site by adding a binary column (Y, N) to the metadata by hand prior to running this code. When sites had duplicate sessions (e.g. if recording was repeated due to poor weather or timing the first time around), we chose the recording session with the highest number and quality calls. However, we preprocessed all calls across the data set and did not remove duplicate sessions just prior to sound analysis (Supplementary Methods 2).

We imported Raven selection tables with temporal coordinates of selected calls across sites.

path <- "/media/owner/MYIOPSITTA/R/Uruguay2017_MonkParakeet_CallAnalysis/Data/Site_Recordings"

cc_sels <- Rraven::imp_raven(path = path, name.from.file = TRUE, ext.case = "upper", freq.cols = FALSE)
# str(cc_sels) 

We merged the selection table with the existing metadata for selected calls, which we generated while selecting calls in Raven to record information about call quality or social context (e.g., group size).

ccs <- read.csv(file.path(path, "SelectedContactCalls_MASTER_14_MAY_18_GSV.csv"), header = TRUE)
# str(ccs)

# make a column of unique call IDs to merge Raven selection table with existing metadata
names(ccs)
names(cc_sels)

ccs$uniq_nm <- paste(ccs$Selection_table, ccs$Selection_ID, sep = "-")
cc_sels$uniq_nm <- paste(cc_sels$selec.file, cc_sels$selec, sep = "-")

ccs <- dplyr::full_join(ccs, cc_sels, by = "uniq_nm")
# str(ccs)

write.csv(ccs, file.path(path, "higher_social_scales_calls_preprocessed.csv"), row.names = FALSE)

We added geographic coordinates and additional metadata columns to the site-level metadata.

# read in updated metadata
ccs <- read.csv(file.path(path, "higher_social_scales_calls_preprocessed.csv"), header = TRUE)

# read in geographic coordinates by site
wp <- read.csv(file.path(path, "recording_waypoints.csv"), header = TRUE)
# str(wp)

# make latitude, longitude and department vectors
# x <- 1
ccs$lat <- unlist(lapply(1:nrow(ccs), function(x){
  wp$Latitude[grep(as.character(ccs$Site[x]), wp$Site, fixed = TRUE)]
}))

ccs$lon <- unlist(lapply(1:nrow(ccs), function(x){
  wp$Longitude[grep(as.character(ccs$Site[x]), wp$Site, fixed = TRUE)]
}))

ccs$dept <- unlist(lapply(1:nrow(ccs), function(x){
  wp$Department[grep(as.character(ccs$Site[x]), wp$Site, fixed = TRUE)]
}))

# make a rec_ID column for catalog labels
ccs$rec_ID <- unlist(lapply(1:nrow(ccs), function(x){
  strsplit(as.character(ccs$Raw_File_Name[x]), split = ".WAV$")[[1]][1]
}))
# str(ccs)

# update the current metadata
write.csv(ccs, file.path(path, "higher_social_scales_calls_preprocessed.csv"), row.names = FALSE)

We made spectrograms for visual quality control filtering. During visual inspection, we added information to the metadata spreadsheet about visual call quality, overlapping signals and any other pertintent information.

specreator(ccs, wn = "hanning", wl = 378, collev = seq(-53, 0, 1), flim = c(0, 10), ovlp = 90, mar = 0.01, res = 300, cex = 0.5, line = FALSE, path = path)

We made call lexicons by site to visualize variation within and among sites, as well as to identify potential calls from repeated individuals.

# order data frame by longitude (W to E), then by latitude (N to S)
ccs <- ccs[order(ccs$lon, ccs$lat, decreasing = c(FALSE, TRUE), method = "radix"), ]

sapply(1:length(unique(ccs$General_site)), function(x){
  catalog(ccs[ccs$General_site == as.character(unique(ccs$General_site)[x]), ], flim = c(0, 9), labels = c("General_site", "rec_ID", "selec"), same.time.scale = TRUE, collev = seq(-53, 0, 1), ovlp = 90, wl = 378, wn = "hanning", res = 150, cex = 0.8, mar = 0.005, sub.legend = TRUE, orientation = "h", leg.wd = 2, img.suffix = as.character(unique(ccs$General_site)[x]), path = path)
})

# collapse jpeg image files per site into a single pdf file
# printed these lexicons to identify potential calls from repeated individuals by visual inspection
catalog2pdf(keep.img = FALSE, by.img.suffix = TRUE, path = path)

We tailored temporal coordinates to standardize how calls were selected (a single observer, GSV, performed tailoring).

# make a temporary title per spectrogram for seltailor
ccs$spec.title <- ""
# str(ccs)

# manually change the name of seltailor_output.csv to site-level_calls_preprocessed.csv
# seltailor retains all other metadata, with the tailored start and end coordinates and a column indicating whether or not the signal coordinates were tailored
seltailor(X = ccs, flim = c(0, 9), mar = 0.2, osci = TRUE, title = "spec.title", wn = "hanning", wl = 378, ovlp = 90, collev = seq(-53, 0, 1), auto_next = TRUE, path = path)

We pre-processed site-level calls for sound analysis.

# total original recording sites
length(levels(ccs$General_site)) 

## [1] 42

# total calls prior to preprocessing
nrow(ccs)

## [1] 899

We removed duplicate recording sessions.

ccs <- ccs[!ccs$Duplicate_session == "Y", ]
nrow(ccs)

## [1] 836

Summary statistics of potential repeated individual calls.

ri_sub <- ccs[which(ccs$Repeated_individual == "Y" & ccs$If_repeated_use == "N"), ]

How many calls were identified as potential repeated individuals?

nrow(ri_sub)

## [1] 19

At how many sites, and which sites?

length(table(ri_sub$General_site)[which(table(ri_sub$General_site) != 0)])

## [1] 11

table(ri_sub$General_site)[which(table(ri_sub$General_site) != 0)]

##
##    ARAZ    BAGU    BCAR    GOLF INES-04    OJOS    PAVO    PEIX PFER-01
##       2       1       6       1       1       1       1       1       1
## PFER-03    VALI
##       2       2

Over these sites, what was the mean and range number of potential repeated individual calls?

mean(table(ri_sub$General_site)[which(table(ri_sub$General_site) != 0)])

## [1] 1.727273

range(table(ri_sub$General_site)[which(table(ri_sub$General_site) != 0)])

## [1] 1 6

We removed potential repeated individual calls.

ccs <- ccs[-which(ccs$Repeated_individual == "Y" & ccs$If_repeated_use == "N"), ]
nrow(ccs) # how many calls remain?

## [1] 817

We removed calls that were visually flagged as poor quality or overlapping with other acoustic signals (that interfered with a bandpass filter of 0.5 - 9 kHz).

ccs <- ccs[-which(ccs$Visual_quality_score == "L"), ]

# some calls (10 total) had horizontal streaks that appeared to be due to amplitude saturation
ccs <- ccs[!ccs$Amplitude_saturation == "Y", ]
nrow(ccs)

## [1] 762

ccs <- ccs[-which(ccs$Overlapping_signal == "Y" & ccs$Bandpass_overlaps_signal_freq_range == "Y"), ]

Sample size across sites?

nrow(ccs)

## [1] 616

table(ccs$General_site)

##
##    1145    ARAP    ARAZ    BAGU    BCAR    CEME    CHAC    CISN    ECIL
##      13      12      15      20      13       6      12      28      17
##    ELTE    EMBR    FAGR    GOLF    HIPE    INBR INES-01 INES-02 INES-03
##      23      22       9      22       5      19      12       2      15
## INES-04 INES-05 INES-06 INES-07 INES-08    KIYU    LACA    LENA    OJOS
##       9       6       6       9      27       8       3      19      23
##    PAVO    PEIX PFER-01 PFER-02 PFER-03    PIED    PLVE    PROO    QUEB
##      25      19      34       4      19      21      11      12      16
## RIAC-01 RIAC-02    ROSA    SAUC    SEMI    VALI
##      17       8      15       6      11      23

Sites with < 5 calls?

table(ccs$General_site)[which(table(ccs$General_site) < 5)]

##
## INES-02    LACA PFER-02
##       2       3       4

We removed sites with < 5 calls (INES-02, LACA, PFER-02).

ccs <- ccs[-grep("INES-02|LACA|PFER-02", ccs$General_site), ]
nrow(ccs) # number of calls remaining

## [1] 607

We calculated SNR across site-level calls.

# check all spectrograms - can the margin of 0.05 that was used for focal individuals work for site level calls?
# it cannot, as that margin sometimes overlaps neighboring calls
# a margin of 0.01 seconds works, but there are 7 calls for which it does not work (calls delivered one after another)
snrspecs(ccs, osci = TRUE, snrmar = 0.01, wn = "hanning", wl = 378, flim = c(0, 10), ovlp = 90, path = path)

# we manually added a new column to metadata csv for whether or not SNR margin of 0.01s worked

# we calculated standard SNR (noise before and after signal) only for calls for which the SNR margin of 0.01s worked

# 7 calls for which SNR margin did not work
nrow(ccs[grep("^N$", ccs$SNR_mar_works), ])

# 5 of these calls have overlapping signals (e.g. a call from the other bird in the pair or flock) that interfered with the SNR margin at the end of the call, so we calculated SNR before the signal only
nrow(ccs[grepl("^N$", ccs$SNR_mar_works) & ccs$SNR_mar_overlap_section == "E", ])

# we calculated SNR across calls with a margin of 0.01s
df_SNR <- sig2noise(X = ccs[, grep("sound.files|^selec$|^start$|^end$", names(ccs))], mar = 0.01, type = 1, bp = c(0.5, 9), wl = 378, path = path)
nrow(df_SNR)

# we recalculated SNR for those calls for which the SNR margin of 0.01s did not work, and that were closely followed by neighboring signals. We used the sig2noise option to calulcate SNR only before the call
df_SNR2 <- sig2noise(X = ccs[grepl("^N$", ccs$SNR_mar_works) & ccs$SNR_mar_overlap_section == "E", grep("sound.files|^selec$|^start$|^end$", names(ccs))], mar = 0.01, type = 1, bp = c(0.5, 9), before = TRUE, wl = 378, path = path)
nrow(df_SNR2)

# SNR improved, as expected
df_SNR$SNR[grepl("^N$", ccs$SNR_mar_works) & ccs$SNR_mar_overlap_section == "E"]
df_SNR2$SNR

# we replaced the SNR calculation for those calls
df_SNR$SNR[grepl("^N$", ccs$SNR_mar_works) & ccs$SNR_mar_overlap_section == "E"] <- df_SNR2$SNR

# for the 2 remaining calls for which the SNR margin of 0.01s did not work, and which were closely preceded by a neighboring call, we used the SNR of the preceding call. sig2noise does not currently have the option to calculate SNR after a given signal, and the following solution ensured that the assigned SNR originated from the local call context.
df_SNR$SNR[grep("2017_07_28_ECIL-02_1182.WAV-3", ccs$uniq_nm)] <- df_SNR$SNR[grep("2017_07_28_ECIL-02_1182.WAV-2", ccs$uniq_nm)]

df_SNR$SNR[grep("2017_09_13_ELTE_1399.WAV-8", ccs$uniq_nm)] <- df_SNR$SNR[grep("2017_09_13_ELTE_1399.WAV-7", ccs$uniq_nm)] 

We removed calls with SNR < 7, the same quality filter as performed on the focal individual calls.

ccs <- data.frame(ccs, SNR = df_SNR$SNR)
str(ccs)

# how many calls with SNR < 7 and how spread out across sites?
nrow(df_SNR[df_SNR$SNR < 7, ]) # 2 calls with SNR < 7
unique(ccs$General_site[which(df_SNR$SNR < 7)]) # at FAGR only

ccs <- ccs[ccs$SNR >= 7, ]

# write out the current metadata
write.csv(ccs, file.path(path, "higher_social_scales_calls_preprocessed_final.csv"), row.names = FALSE)

Make an extended selection table of calls for higher social scales

ccs <- read.csv(file.path(path, "higher_social_scales_calls_preprocessed_final.csv"), header = TRUE)
str(ccs)

cores <- parallel::detectCores() - 2
hss_sels_tbl <- selection_table(X = ccs, whole.recs = FALSE, extended = TRUE, confirm.extended = TRUE, mar = 0.1, parallel = cores, path = path)

# write out extended selection table as an RDS object
saveRDS(hss_sels_tbl, file.path(path, "higher_social_scales_extended_seltable.RDS"))

The different components of the site-level extended selection table can be accessed as per the focal individual extended selection table (see code above), to facilitate reproducing subsequent analyses.

Summary Statistics

Individual

How many total repeatedly sampled individual calls remained after pre-processing?

nrow(fselst_snr)

## [1] 97

Calls per repeatedly sampled individual.

table(fselst_snr$indiv)

##
## AAT RAW UM1 UM2 UM3 UM4 UM5 ZW8
##  12   4  25  23   5  13   7   8

Mean number of calls per individual.

mean(table(fselst_snr$indiv))

## [1] 12.125

Range of calls per individual.

range(table(fselst_snr$indiv))

## [1]  4 25

Calls by repeatedly sampled individuals and sites.

table(fselst_snr$indiv, fselst_snr$site)

##
##       1145 CHAC EMBR
##   AAT   12    0    0
##   RAW    0    0    4
##   UM1   25    0    0
##   UM2   23    0    0
##   UM3    5    0    0
##   UM4   13    0    0
##   UM5    0    7    0
##   ZW8    0    0    8

Mean and range of SNR across calls.

mean(fselst_snr$SNR)

## [1] 16.11411

range(fselst_snr$SNR)

## [1]  7.751148 26.992037

We read back in final preprocessed set of calls for higher social scales.

Dates for recording sessions by site?

sites <- as.character(unique(ccs$General_site))

dates <- sapply(1:length(sites), function(x){

  tmp <- ccs[grep(sites[x], ccs$General_site), ]
  unique(paste(tmp$Day, tmp$Month, sep = "-"))

})

data.frame(dates, sites)

We defined pairs as groups of 2 birds for which we had calls for both birds in the pair. The pair social scale was nested within the site scale, and was represented by a subset of calls in the site-level data set.

We extracted unique overall group ID from the site-level metadata.

gid <- unique(ccs$Overall_group_ID)[!is.na(unique(ccs$Overall_group_ID))]
length(gid)

## [1] 109

We collected information about all groups of birds flying together (pairs and flocks) in the data set for summary statistics.

# filter the site-level data by all groups and generate a data frame that can be easily parsed for summary stats
gid_size <- lapply(1:length(gid), function(x){

  pat <- paste("^", gid[x], "$", sep = "")

  st <- unique(ccs$General_site[grep(pat, ccs$Overall_group_ID)])
  gi <- unique(ccs$Flock_ID[grep(pat, ccs$Overall_group_ID)])
  pid <- unique(ccs$Pair_ID[grep(pat, ccs$Overall_group_ID)])
  gs <- unique(ccs$Group_size[grep(pat, ccs$Overall_group_ID)])
  cn <- length(ccs$Group_size[grep(pat, ccs$Overall_group_ID)])

  return(data.frame(site = st, overall_group_id = gid[x], flock_ID = gi, pair_ID = pid, group_size = gs, call_num = cn))
})
gid_size <- data.table::rbindlist(gid_size)

# filter by groups for which we had > 1 call (e.g. remove groups for which only a single call remained after preprocessing)
gidf <- droplevels(gid_size[gid_size$call_num > 1, ])
# View(gidf)

Pair

pairs <- droplevels(gidf[!is.na(gidf$pair_ID), ])

# save the pair IDs for Supplementary Methods 3
saveRDS(pairs$pair_ID, file.path(path, "pair_IDs.RDS"))

Mean and range of pairs across sites?

range(table(pairs$site)) 

## [1] 1 5

mean(table(pairs$site)) 

## [1] 2.095238

How many pairs?

nrow(pairs)

## [1] 44

How many calls in the site-level data set represent this social scale?

sum(pairs$call_num)

## [1] 88

sum(pairs$call_num)/nrow(ccs)

## [1] 0.1454545

How many sites had pairs?

length(table(pairs$site))

## [1] 21

How many pairs by site?

table(pairs$site)

##
##    1145    ARAP    ARAZ    CISN    ECIL    ELTE    EMBR    INBR INES-01
##       1       1       2       4       1       5       1       3       2
## INES-03 INES-04 INES-08    LENA    OJOS    PAVO    PEIX PFER-01    PIED
##       3       2       4       1       1       1       1       2       1
## RIAC-01    ROSA    SAUC
##       3       3       2

Mean and range of pairs across sites?

range(table(pairs$site)) 

## [1] 1 5

mean(table(pairs$site)) 

## [1] 2.095238

How many and which sites have > 2 pairs?

length(table(pairs$site)[which(table(pairs$site) > 1)])

## [1] 12

table(pairs$site)[which(table(pairs$site) > 1)]

##
##    ARAZ    CISN    ELTE    INBR INES-01 INES-03 INES-04 INES-08 PFER-01
##       2       4       5       3       2       3       2       4       2
## RIAC-01    ROSA    SAUC
##       3       3       2

Flock

We defined flocks as groups of 3 or more birds flying together for which we had 2 or more calls. This social scale was nested within the site scale, and represented by a subset of calls in the site-level data set.

How many flocks total?

nrow(gid_size[gid_size$group_size > 2, ])

## [1] 36

How many flocks remain (e.g. flocks for which we had more than 1 call)?

flocks <- droplevels(gidf[gidf$group_size > 2, ])
nrow(flocks)

## [1] 29

# save the flock IDs for Supplementary Methods 3
saveRDS(flocks$flock_ID, file.path(path, "flock_IDs.RDS"))

Mean and range of flock size across the data set?

mean(flocks$group_size)

## [1] 5.275862

range(flocks$group_size)

## [1]  3 20

How many calls from the site-level data set correspond to these flocks?

sum(flocks$call_num)

## [1] 77

sum(flocks$call_num)/nrow(ccs)

## [1] 0.1272727

How many sites had flocks?

length(table(flocks$site))

## [1] 22

How many flocks by site?

table(flocks$site)

##
##    ARAP    ARAZ    CHAC    CISN    ECIL    ELTE    FAGR    HIPE    INBR
##       2       2       1       1       2       2       1       1       1
## INES-03 INES-05 INES-06 INES-07 INES-08    KIYU    OJOS    PAVO    PEIX
##       1       1       1       1       1       1       1       2       1
##    PIED    QUEB RIAC-02    ROSA
##       2       1       2       1

Mean and range of flocks across sites?

range(table(flocks$site)) 

## [1] 1 2

mean(table(flocks$site)) 

## [1] 1.318182

How many and which sites have > 2 flocks?

length(table(flocks$site)[which(table(flocks$site) > 1)])

## [1] 7

table(flocks$site)[which(table(flocks$site) > 1)]

##
##    ARAP    ARAZ    ECIL    ELTE    PAVO    PIED RIAC-02
##       2       2       2       2       2       2       2

Site

How many total calls were retained for subsequent sound analysis at this social scale?

nrow(ccs)

## [1] 605

How many sites represented?

length(table(ccs$General_site))

## [1] 39

How many calls by site?

table(ccs$General_site)

##
##    1145    ARAP    ARAZ    BAGU    BCAR    CEME    CHAC    CISN    ECIL
##      13      12      15      20      13       6      12      28      17
##    ELTE    EMBR    FAGR    GOLF    HIPE    INBR INES-01 INES-03 INES-04
##      23      22       7      22       5      19      12      15       9
## INES-05 INES-06 INES-07 INES-08    KIYU    LENA    OJOS    PAVO    PEIX
##       6       6       9      27       8      19      23      25      19
## PFER-01 PFER-03    PIED    PLVE    PROO    QUEB RIAC-01 RIAC-02    ROSA
##      34      19      21      11      12      16      17       8      15
##    SAUC    SEMI    VALI
##       6      11      23

Mean and range of calls by site?

mean(table(ccs$General_site))

## [1] 15.51282

range(table(ccs$General_site))

## [1]  5 34

How many calls by department? Uruguay is sub-divided into 19 departments, and we sampled across 7 departments (representaive of a regional scale for matrix regression with geographic distance, see Supplementary Methods 3).

table(ccs$dept)

##
##  Canelones    Colonia  Maldonado Montevideo      Rocha      Salto
##         19        275         89         99         46         12
##   San Jose
##         65

Mean and range SNR across calls?

mean(ccs$SNR)

## [1] 19.56147

range(ccs$SNR)

## [1]  7.795863 40.149548

Sites - Colonia

Colonia was the westernmost department of our geographic transect, where we sampled most intensively.

How many total calls were retained for subsequent sound analysis?

Colonia <- droplevels(ccs[grep("Colonia", ccs$dept), ])
nrow(Colonia)

## [1] 275

nrow(Colonia)/nrow(ccs) # proportion of all calls

## [1] 0.4545455

How many sites are represented?

length(table(Colonia$General_site))

## [1] 18

How many calls by site?

table(Colonia$General_site)

##
##    1145    CHAC    EMBR INES-01 INES-03 INES-04 INES-05 INES-06 INES-07
##      13      12      22      12      15       9       6       6       9
## INES-08    LENA PFER-01 PFER-03    PIED RIAC-01 RIAC-02    ROSA    SEMI
##      27      19      34      19      21      17       8      15      11

Mean and range of calls by site?

mean(table(Colonia$General_site))

## [1] 15.27778

range(table(Colonia$General_site))

## [1]  6 34

Mean and range SNR across calls?

mean(Colonia$SNR)

## [1] 19.54524

range(Colonia$SNR)

## [1]  9.222555 35.116192

Check out Supplementary Methods 2 for sound analysis. The extended selection tables for the individual and higher social scales contain the selected contact calls used for analyses in Supplementary Methods 2.

References

1. Araya‐Salas, M., & Smith‐Vidaurre, G. 2017. warbleR: An R package to streamline analysis of animal acoustic signals. Methods in Ecology and Evolution 8(2), 184-191.