C-POD data workshop: Time series

In this topic we will explore temporal data in R in order to visualize the temporal aspect of porpoise distribution.

1. Dealing with time variables in R

As discussed previously, a variable is recognized as time in the POSIxct data type, for example by applying lubridate::parse_date_time().

A first step is to plot the value of interest against time.

ggplot(data = poddata_day, aes(x = Time, y = Dpm)) + 
  theme_bw() +
  theme(axis.text = element_text(size = 16),
        axis.title.x = element_blank()) +
  geom_point() + geom_line()

More relevant is plotting the time series for each zone seperately. This can be done with facet_wrap() or by making a list of plots.

ggplot(data = poddata_day, aes(x = Time, y = Dpm)) + 
  theme_bw() +
  theme(axis.text = element_text(size = 16),
        axis.title.x = element_blank()) +
  geom_point() + geom_line() + facet_wrap(~Zone, nrow = length(unique(poddata$Station)))

lapply(unique(poddata_day$Zone), function(x){
  ggplot(data = poddata_day[poddata_day$Zone == x,], aes(x = Time, y = Dpm)) + 
    theme_bw() +
    theme(axis.text = element_text(size = 16),
          axis.title.x = element_blank()) +
    geom_point() + geom_line() + ggtitle(x)
})

[[1]]

[[2]]

[[3]]

[[4]]

[[5]]

[[6]]

[[7]]

[[8]]

[[9]]

[[10]]

Same can be done for different deployments.

lapply(unique(poddata_day$Deployment_fk), function(x){
  ggplot(data = poddata_day[poddata_day$Deployment_fk == x,], aes(x = Time, y = Dpm)) + 
    theme_bw() +
    theme(axis.text = element_text(size = 16),
          axis.title.x = element_blank()) +
    geom_point() + geom_line() + ggtitle(x)
})

2. Autocorrelation

R provides a very easy function to investigate the autocorrelation in your data: acf.

poddata_day <- arrange(poddata_day, Zone, Time)
acf(poddata_day$Dpm)

Again, we are interested in the autocorrelation of each Zone seperately.

par(mfrow=c(2,2)) # see four plots at a time
lapply(unique(poddata_day$Zone), function(x) {
  data <- acf(poddata_day[poddata_day$Zone == x,]$Dpm, plot = F)
  plot(data, main = x)
})

[[1]]
NULL

[[2]]
NULL

[[3]]
NULL

[[4]]
NULL

[[5]]
NULL

[[6]]
NULL

[[7]]
NULL

[[8]]
NULL

[[9]]
NULL

[[10]]
NULL

We can also take these values and combine them in one plot.

lacf <- lapply(unique(poddata_day$Zone), function(x){
  o <- filter(poddata_day, Zone == x)
  oa <- acf(o$Dpm)
  dfo <- data.frame(acf = oa$acf[,,1], lag = oa$lag[,,1], Zone = x)
})
library(plyr) # Now we use the package plyr to make a dataframe out of our list
dacf <- ldply(lacf, rbind)

ggplot(dacf) + geom_path(aes(x=lag, y=acf, group=Zone))

3. Smoothing

The goal of smoothing is generally to aid visual interpretation of a time series. Let’s try on two subsets of our data.

test1 <- filter(poddata_day, Deployment_fk == 2578)
test2 <- filter(poddata_day, Deployment_fk == 2585)

Applying geom_smooth() plots a loess smoother (weighted regression) on the series.

ggplot(data = test1, aes(x = Time, y = Dpm)) + 
  geom_point() + geom_line() + geom_smooth() + ggtitle(2578)

ggplot(data = test2, aes(x = Time, y = Dpm)) + 
  geom_point() + geom_line() + geom_smooth() + ggtitle(2585)

Another way to smooth data, is to calculate a moving average.

acf(test1$Dpm) # First, we check the autocorrelation to choose our window size.

library(pastecs)

Loading required package: boot

Attaching package: <U+393C><U+3E31>pastecs<U+393C><U+3E32>

The following objects are masked from <U+393C><U+3E31>package:dplyr<U+393C><U+3E32>:

    first, last

movavg <- decaverage(test1$Dpm, order = 3) # By choosing an order of 3, a window size of 7 will be obtained (3 values to the left, 3 to the right).
movavg # movavg is a list

 Call:
 decaverage(x = test1$Dpm, order = 3)

Components
[1] "filtered"  "residuals"

plot(movavg)

Now in ggplot:

test1$decavg <- data.frame(movavg$series)$filtered

ggplot(test1, aes(x = Time, y = Dpm)) + geom_point() + geom_line() + 
  geom_path(aes(y = decavg), size= 1.2, colour = "red")

movavg <- decaverage(test1$Dpm, order = 3, times = 5)
test1$decavg <- data.frame(movavg$series)$filtered

ggplot(test1, aes(x = Time, y = Dpm)) + geom_point() + geom_line() + 
  geom_path(aes(y = decavg), size= 1.2, colour = "red")

Now, we can do the same for the other subset.

acf(test2$Dpm)

movavg <- decaverage(test2$Dpm, order = 1, times = 5)
test2$decavg <- data.frame(movavg$series)$filtered

ggplot(test2, aes(x = Time, y = Dpm)) + geom_point() + geom_line() + 
  geom_path(aes(y = decavg), size= 1.2, colour = "red")

Discussion: which smoothing technique is preferred? Is smoothing necessary?