One of the most important parts of a scientific study is how you convey the findings to an audience. Yes, scientists write reports to describe what the data says and interpret why the results are important. But many people, myself included, so straight to the figures, tables, and images to “see” what is going on.
How we plot our findings can make the difference between an impactful study and a study that is too confusing to get any information out of it.
I have been writing up our massive climate study, using Figure 2 as my template to display the major results. However, as I have been preparing a poster for the CERF conference, I realized: there may be a better way to plot these data.
I previously wrote a post on how I created the plot featured below (Figure 2). Do not get me wrong, I still like this multi-panel plot …. and these plots will likely find themselves in the Supplemental Information. However, sometimes simplicity is best.
The extreme climate index highlighted here is Frost Days, or the annual count when the daily maximum temperature is below 0°C.
This is what Figure 2 is displaying as a series of 5 separate plots (A-E).
A) All of the Frost Days time series used in this study are displayed here. The black and gray lines are the HadEX2 and GHNCdex gridded data. The blue (northern stations) and red (southern stations) lines display all 18 of the individual NCDC-Daily weather stations. This plot shows the temporal and spatial variability in Frost Days, as well as the general downward direction, suggesting that we have observed fewer Frost Days.
B) A boxplot for all data sets, arranged from north to south. This plot highlights that the southernmost stations, in general, experience fewer Frost Days, which makes perfect sense!
C) This moving-average plot aggregates the Northern and Southern stations into one time series. This plot shows the 21-year average pattern, allowing us to see any trends in the data. Sure enough, Frost Days are significantly decreasing in all these data sets.
D&E) These two plots are probability density functions (PDFs). PDFs allow you to visualize variability as well as shifts in the 10th and 90th percentiles. The wider the PDF, the greater the spread in the data. Loosely speaking, a wider PDF mean we have a greater variety of possible Frost Day occurrences in a given year.
This plot just took a page to describe! It is too busy!
As this study has progressed, new results have altered the focal points of this study.
For example, I have removed the GHCNdex data. This shorter time series always agrees with the HadEX2. This makes senses since the HadEX2 incorporates all the data in the GHCNdex. So, we can simply say “the GHCNdex agrees with the patterns we saw in the HadEX2” and remove the line to de-clutter our images.
Additionally, in the manuscript, we aggregated the North and South stations. I realized that panel A in Figure 2 is too busy. It is impossible to pick out the individual stations, and it simply makes more sense to display the data as we used it.
With these changes, we have re-created Figure 2 to be Figure 3, which trims down our 5 panel plot into 2 panels. The data, patterns, and trends are all the same, but hopefully this new plot is easier to understand and less overwhelming.
What do you think?