How to plot a time-series scatterplot showing missing y-axis values as gaps?

Question

I am trying to create a scatterplot with date/time on the x-axis and salinity on the y-axis. However, there are some date/time points which do not have a salinity value due to equipment failure, but I still need these portions of time to show on my graph to help explain the ecological patterns I am looking at. Can anyone advise on how to keep these missing sections shown on the graph?

My current code for the data and the plot which does not show the missing values.

Edit My data has explicit missing values where data was removed due to logger errors but is listed as an 'NA' (see photo). Unfortunately I have thousands of data points collected half hourly so it is difficult to show all of the data.

Screenshot of data showing 'NA' values

OY1_AllTimes <- read_csv("~/Documents/TAMUG_Thesis/Rollover_Pass_Data/Logger/RP_LoggerData_OY1_AllTimes.csv")
summary(OY1_AllTimes)

OY1_AllTimes$Date_time<-paste(OY1_AllTimes$Date, OY1_AllTimes$Time)
summary(OY1_AllTimes$Date_time)

date_time_OY1_AllTimes<-as.POSIXct(OY1_AllTimes$Date_time, format="%m/%d/%Y %H:%M")
date_time_OY1_AllTimes
date_time2_OY1_AllTimes<-as.factor(date_time_OY1_AllTimes)
date_time2_OY1_AllTimes
summary(OY1_AllTimes)

Summary of OY1_AllTimes

p_OY1_AllTimes <- ggplot(data = OY1_AllTimes, aes(x=date_time2_OY1_AllTimes, y=Salinity)) + geom_point() + theme_classic()+
  scale_x_discrete("Date", breaks=c("0019-10-04 09:30:00", "0019-11-01 05:00:00", "0019-12-01 00:00:00", "0020-01-01 00:00:00", "0020-02-01 00:00:00",
                                    "0020-03-01 00:00:00","0020-04-01 00:00:00", "0020-05-01 00:00:00", "0020-06-01 00:00:00"),
                   labels=c("10/2019", "11/2019", "12/2019", "1/2020", "2/2020", "3/2020", "4/2020", "5/2020", "6/2020"))+ylab("Salinity")+ggtitle("OY1")
p_OY1_AllTimes

Scatterplot of OY1 without missing values

Essentially I would like to see the above scatterplot with gaps representing the periods without salinity data so that the date/time scale is continuous.

Subsample of data:

# A tibble: 50 x 5
   Site  Date    Time   Salinity Date_time       
   <chr> <chr>   <time>    <dbl> <chr>           
 1 OY1   10/4/19 09:30    NA     10/4/19 09:30:00
 2 OY1   10/4/19 10:00    NA     10/4/19 10:00:00
 3 OY1   10/4/19 10:30     0.891 10/4/19 10:30:00
 4 OY1   10/4/19 11:00     0.961 10/4/19 11:00:00
 5 OY1   10/4/19 11:30     1.02  10/4/19 11:30:00
 6 OY1   10/4/19 12:00     1.10  10/4/19 12:00:00
 7 OY1   10/4/19 12:30     1.19  10/4/19 12:30:00
 8 OY1   10/4/19 13:00     1.27  10/4/19 13:00:00
 9 OY1   10/4/19 13:30     1.33  10/4/19 13:30:00
10 OY1   10/4/19 14:00     1.42  10/4/19 14:00:00
# … with 40 more rows```

Answer 1

Here's my attempt at demonstrating this with some reproducible code that we can all run.

Here's some arbitrary fake data. What's important is that it has a big gap in the timestamps, since I removed a few hundred rows from 100:399. At this point timestamp is stored as datetime data, in the "POSIXct" variety, the most typical, and the same as your date_time_OY1_AllTimes variable.

set.seed(42)
my_fake_data <- data.frame(timestamp = as.POSIXct("2021-01-01 00") + cumsum(runif(1000, 0, 6E4)), reading = cumsum(rnorm(1000)))
my_fake_data <- my_fake_data[c(1:99, 400:1000),]

The typical thing in ggplot2 is to plot using that POSIXct value. You'll see the gap. ggplot2 maps the timestamp to the x axis, and picks the default labels for us.

ggplot(my_fake_data, aes(timestamp, reading)) +
  geom_point() 

If we want monthly labels, we can specify that and the format we want to see:

... + scale_x_datetime(date_breaks = "month", 
                       date_labels = "%b\n'%y", minor_breaks = NULL)

In your example, the timestamps have been converted to factors, which preserves their sequence, but it removes them from their context in time, so the gaps have disappeared. Here I've added discrete labels manually, but they no longer have an explicit relationship in time to my data points. I can make them say whatever I want, and they'll be wrong unless I put in some work to align them manually.

ggplot(my_fake_data, aes(as.factor(timestamp), reading)) +
  geom_point() +
  scale_x_discrete(breaks = as.factor(my_fake_data[1+100*0:7,1]),
                   labels = format(
                     seq.Date(as.Date("2021-01-01"), 
                              as.Date("2021-08-01"), by = "month"), "%b %Y"))

Answer 2

It is quite hard to tell what your data really looks like, I am ussuming you have a implicit missing data problem.

Which would mean, you have a data.frame/time series with missing observations. But the problem is, these missing values are not explicitly given as NAs. Instead these are just left out.

A time series with NAs would look like this:

1.1.2021 14:00
1.1.2021 15:00
1.1.2021 16:00
1.1.2021 17:00
1.1.2021 NA
1.1.2021 19:00

I guess your problem looks like this:

1.1.2021 14:00
1.1.2021 15:00
1.1.2021 16:00
1.1.2021 17:00
1.1.2021 19:00

So the difference is, there is no NA value for the 18:00 timestep. But, of course you do know there is a missing values (thats why it is called implicit missing value).

Assuming you have a regular spaced time series (meaning values measured in regular intervals e.g. 1h), you can use the tsibble package to convert the implicit missing values to normal missing values, where you have the NAs is the series.

Here is an easy example (as I don't have your data):

library("tsibble")

# Read in your data as tsibble
data_example <- tsibble(
  year = c(2016, 2017, 2018, 2019, 2021, 2022),
  measure = sample(1:10, size = 6),
  index = year
)

# Take a look at the data
data_example

# Use the fill_gaps function of tsibble
data_na <- fill_gaps(data_example, .full = TRUE)

# You can see now, the implicit missing year 2020 is now added as NA  
data_na

You can of course also do this for all kinds of different regular spaced time series data (15 seconds, minute, hour, month, ...). You just have to define the time step this while creating your tsibble object.

Plotting is easy now:

library("ggplot2")
ggplot(data = data_na) + geom_point( aes(year, measure))

This will give you this plot:

As you can see, as you wanted, the series is plotted, but there is just no for 2020 in the plot. If you want to put more focus to the missing data, you can also use the imputeTS package.

library("imputeTS")
ggplot_na_distribution(data_na)

This would then look like this:

This is only a small example time series, for larger time series this would look like e.g. this plot: