Blog

2.1 Monthly average

To get the monthly average temperature we will need to combine the maximum tmax and minimum tmin columns in some way. The simplest option is to just take the average of these values.

This is a little trickier than it seems, since we need to take an average of values from multiple columns, unlike most of the time in R when we are taking the average of multiple rows.

The easiest way is to do this is simply go back to first principles and remember the formula for calculating a mean: add all the values, then divide the sum by the number of values. I can do this using mutate() to create a new column called tmean.

oxford_data<-
  oxford_data %>%
    mutate(tmean=(tmin+tmax)/2)

I am also going to over-write the existing object at this point by assigning the output to be an object called oxford_data. Since step 2.2 will not be inside a continuous pipe (although in the video solution, I did use a slightly more complicated method to do everything within a single pipe. The code for that alternative is included at the end of this solution.)

2.2: Reference period average

From the updated data created in 2.1, I will filter to just the years 1971-2000 and then get the mean value.

refmean=oxford_data %>%
  filter(yyyy>=1971 & yyyy<=2000) %>%
    summarise(mean=mean(tmean))

2.3: Yearly average

Using group_by and summarise, I can now calculate mean values for each year. Remember there was one missing value in our dataset. Because of the seasonality of temperature, if we use na.rm=T when calculating the mean temperature for the year, we will definitely have a biased estimate of the mean. Therefore it is probably safer to not use this option for calculating the mean, and leaving the mean for this year as an NA.

oxford_data %>%
  group_by(yyyy) %>%
    summarise(tmean=mean(tmean))

## # A tibble: 167 x 2
##     yyyy tmean
##    <int> <dbl>
##  1  1853  8.97
##  2  1854  9.69
##  3  1855  8.83
##  4  1856  9.53
##  5  1857 10.3 
##  6  1858  9.50
##  7  1859 10.1 
##  8  1860 NA   
##  9  1861  9.76
## 10  1862  9.90
## # ... with 157 more rows

2.4: Temperature anomaly

Finally, we can calculate the temperature anomaly by subtracting the yearly averages from the reference period average:

oxford_data %>%
  group_by(yyyy) %>%
    summarise(tmean=mean(tmean)) %>%
        mutate(anomaly=tmean-refmean$mean)

## # A tibble: 167 x 3
##     yyyy tmean anomaly
##    <int> <dbl>   <dbl>
##  1  1853  8.97  -1.42 
##  2  1854  9.69  -0.699
##  3  1855  8.83  -1.56 
##  4  1856  9.53  -0.853
##  5  1857 10.3   -0.116
##  6  1858  9.50  -0.891
##  7  1859 10.1   -0.291
##  8  1860 NA     NA    
##  9  1861  9.76  -0.628
## 10  1862  9.90  -0.482
## # ... with 157 more rows

Note that, even though there is only one value in the data frame refmean, we still need to refer to the column name to access this value.

2.5 Filter data

Because of the missing value in 1860, then it probably makes sense to only plot the years we have complete data for. So let’s restrict the plot to only 1861 and onwards.

oxford_plot_data<-
  oxford_data %>%
  group_by(yyyy) %>%
    summarise(tmean=mean(tmean)) %>%
        mutate(anomaly=tmean-refmean$mean)%>%
          filter(yyyy>=1861)

I also decided to assign this into a new object called oxford_plot_data at this point. I will then use this data for my plots!

Make the initial plot

Let’s deconstruct the plot structure to understand more about how it works.

• We have year on the x axis
  
• A coloured ‘stripe’ for each year, (a bar of constant height)
  
• The colour represents the anomaly

So within ggplot2 our solution will be

• Map yyyy to the x axis
  
• Map anomaly to the fill axis
  
• The variable going onto the y axis should not be a variable at all! We want a constant, so we can just set this to be an arbitrary value. We do need to set this to be equal to some value, so R knows how high to make each bar.
  
• Use geom_col not geom_bar, because we are setting the heights of the bars, rather than using bars to summarise counts, or other statistics, from data

This gives us:

ggplot(data=oxford_plot_data,aes(x=yyyy,y=1,fill=anomaly))+
  geom_col()

Not a bad start!

Now we need to think about how to format this into the classic ‘warming bars’ style. There are a few things we need to do:

1. Change the colour palette so that low values are blue and high values are red and zero is white
   
2. Make the colour scale ‘symmetrical’. You can see from the legend that the zero value is not in the middle of scale.
   
3. Remove the small spaces between bars
   
4. Remove all the gridlines, axis labels, and backgrounds

3.1 Colour Palette

This is where we will need to use a scale_ function, to modify the way the colours are allocated.

There are a few different scale functions for creating colour gradients depending on whether you need a single colour gradient, a double colour gradient or a n-level colour gradient. In this particular case we need a “double” gradient - because we want to set negative values to go from dark blue up to white (at zero).

Then from zero we want an additional gradient to go from white to dark red as the numbers become more positive. So the function we need is scale_fill_gradient2. In this we need to set the colour values for our “low”, “mid” and “high” extremes of colour.

ggplot(data=oxford_plot_data,aes(x=yyyy,y=1,fill=anomaly))+
  geom_col()+
    scale_fill_gradient2(low = "darkblue",mid="white",high="darkred")

You can see if you do not add in the “white” mid point, and just use the single red-blue gradient, using scale_fill_gradient() then the plot does not look quite right:

ggplot(data=oxford_plot_data,aes(x=yyyy,y=1,fill=anomaly))+
  geom_col()+
    scale_fill_gradient(low = "darkblue",high="darkred")

An even better option here would be to use the new function scale_fill_fermenter()which works in a similar way to scale_fill_brewer() except with continuous variables being used to set colours instead. This is nice because the “RdBu” colour palette from colour brewer is exactly the colour palette used in the warming stripes plots. And this colour palette does already centre around a shade of white.

ggplot(data=oxford_plot_data,aes(x=yyyy,y=1,fill=anomaly))+
  geom_col()+
    scale_fill_fermenter(palette="RdBu")

3.2 Symmetrical colour scale

A little surprisingly, because of the default options within scale_fill_gradient2, we now have a symmetrical colour palette! The function defaults to a midpoint of zero, and stretches the colour assignments so that they are symmetrical. You can’t tell this from just looking at the plot, but the top level of ‘reds’ being plotted is not at the full extent of ‘darkred’, unlike the blue which does extend as far as possible. If we did want to make sure of this explicitly we could set the limits within the scale call, exactly as we might modify the limits of an x or y axis as we have seen previously.

ggplot(data=oxford_plot_data,aes(x=yyyy,y=1,fill=anomaly))+
  geom_col()+
    scale_fill_gradient2(low = "darkblue",mid="white",high="darkred",limits=c(-2,2))

The plot is identical, but you can see the darkest red in the legend is now a bit darker!

Within scale_fill_fermenter this is not automatic - so we would need to explicitly set limits, and make sure they are symmetrical.

ggplot(data=oxford_plot_data,aes(x=yyyy,y=1,fill=anomaly))+
  geom_col()+
    scale_fill_fermenter(palette="RdBu",limits=c(-1.5,1.5))

3.3 Removing spaces

It may not be obvious without zooming in, but there are small spaces between each bar. This is because by default R sets bars to fill 90% of the width available, and leaves 5% spacing between bars. However no such spaces exist in the ‘classic’ warming stripes. Thankfully this is easy to resolve by adding the option width=1 into geom_col.

ggplot(data=oxford_plot_data,aes(x=yyyy,y=1,fill=anomaly))+
  geom_col(width=1)+
    scale_fill_fermenter(palette="RdBu",limits=c(-1.5,1.5))

3.4 Removing labels, gridlines etc.

With the theme() function we can manually change individual components. We could use this to individually remove the labels, and gridlines. But there is also a built in theme which will do this for us - theme_void().

ggplot(data=oxford_plot_data,aes(x=yyyy,y=1,fill=anomaly))+
  geom_col(width=1)+
    scale_fill_fermenter(palette="RdBu",limits=c(-1.5,1.5))+
      theme_void()

We might also want to get rid of the legend, and add a title. Just to add a finishing touch

ggplot(data=oxford_plot_data,aes(x=yyyy,y=1,fill=anomaly))+
  geom_col(width=1,show.legend = FALSE)+
    scale_fill_fermenter(palette="RdBu",limits=c(-1.5,1.5))+
     theme_void()+
      labs(title="Oxford Warming Stripes",subtitle = "1861-2019")

Job done! Although I am sure there is plenty more we could keep tweaking around here!