Facetting & Grids
Faceting Plots
Displaying all the relevant information on a single plot panel is ideal as it minimizes the amount of space consumed, but it is sometimes impossible to use only one plot panel and maintain a readable plot that isn’t crowded or trying to do so much. In this case, a handy approach is to facet the plot which essentially breaks up the plot into multiple panes so that they can be viewed together.
Often, this is the case when one or both axes are the same, and a separate variable or group of that variable is being shown on each plotting pane. For example here, let’s refer to data on stream chemistry:
library(tidyverse)
library(lterdatasampler)
df <- lterdatasampler::luq_streamchem
head(df)## # A tibble: 6 × 22
## sample_id sample_date gage_ht temp p_h cond cl
## <chr> <date> <dbl> <dbl> <dbl> <dbl> <dbl>
## 1 QS 1987-01-05 2.82 20 7.22 48.2 7.3
## 2 QS 1987-01-13 2.66 20 7.34 49.8 7.5
## 3 QS 1987-01-20 2.61 20 7.12 50.3 7.5
## 4 QS 1987-01-27 2.58 20 7.19 50.4 7.3
## 5 QS 1987-02-03 2.8 20 7.36 49.6 7.3
## 6 QS 1987-02-10 2.63 20 7.19 53.3 7.2
## # … with 15 more variables: no3_n <dbl>, so4_s <dbl>,
## # na <dbl>, k <dbl>, mg <dbl>, ca <dbl>, nh4_n <dbl>,
## # po4_p <dbl>, doc <dbl>, dic <dbl>, tdn <dbl>,
## # tdp <dbl>, si_o2 <dbl>, don <dbl>, tss <dbl>There are two variables we might want to compare – Sodium (Na) and Potassium (K). One option could be plotting them through time with two y-axes, but these plots are often confusing. An easier to interpret way would be to have two separate plot panels. This will require some data prep first.
One (not the only) way to go about this, is to get both of our variables of interest into one column, with a separate grouping column that denotes what the values refer to. The easiest way to do this is by re-shaping the data Let’s do that here:
df_long <- df %>%
# remove columns we don't need
dplyr::select(sample_date, na, k) %>%
tidyr::pivot_longer(
# specify which columns to join together
cols = c(na, k),
# specify what the new name of the grouping variable will be
names_to = "element"
)
df_long## # A tibble: 634 × 3
## sample_date element value
## <date> <chr> <dbl>
## 1 1987-01-05 na 4.75
## 2 1987-01-05 k 0.18
## 3 1987-01-13 na 4.81
## 4 1987-01-13 k 0.19
## 5 1987-01-20 na 5.19
## 6 1987-01-20 k 0.2
## 7 1987-01-27 na 5.08
## 8 1987-01-27 k 0.18
## 9 1987-02-03 na 4.86
## 10 1987-02-03 k 0.17
## # … with 624 more rowsPerfect! We will also want to pull out the month and year values as columns since it will be month we’ll use to group by in our plots.
library(lubridate)
df_long <- df_long %>%
dplyr::mutate(
year = lubridate::year(sample_date),
month = lubridate::month(sample_date)
)
df_long## # A tibble: 634 × 5
## sample_date element value year month
## <date> <chr> <dbl> <dbl> <dbl>
## 1 1987-01-05 na 4.75 1987 1
## 2 1987-01-05 k 0.18 1987 1
## 3 1987-01-13 na 4.81 1987 1
## 4 1987-01-13 k 0.19 1987 1
## 5 1987-01-20 na 5.19 1987 1
## 6 1987-01-20 k 0.2 1987 1
## 7 1987-01-27 na 5.08 1987 1
## 8 1987-01-27 k 0.18 1987 1
## 9 1987-02-03 na 4.86 1987 2
## 10 1987-02-03 k 0.17 1987 2
## # … with 624 more rowsNow we can do about plotting this iteratively. In this example, we’ll use a box plot (see description here) since it’s one way to show a change through time. As a quick refresher, here’s how we can interpret a boxplot:
Stream Chemicals
Stream chemistry is an important indicator of ecosystem health. In our example, we consider long term stream chemistry data from a stream in a protected forest, Luquillo Experimental Forest, in Puerto Rico. However, stream chemistry is of interest closer to home too!
PhD Candidate Rachel Giles, in Dr. Chelsea Rochman’s lab in the EEB department, is investigating how anthropogenic contaminants including road salt impact wildlife in urban streams in Toronto, ON. Stream chemistry is a valuable and important consideration of stream health both in remote intact forests, as well as in busy metropolises.
Plotting with and without Facets
Let’s begin by plotting only the bare necessities:
ggplot() +
geom_boxplot(data = df_long, mapping = aes(x = year, y = value))
Well, that didn’t quite work out. We will need to make year a factor here:
df_long = df_long %>%
dplyr::mutate(
year = as.factor(year)
)Now we’ll try again:
ggplot() +
geom_boxplot(data = df_long, mapping = aes(x = year, y = value))
That looks better. Now here’s why we might want to facet: right now our plot is displaying both sodium and potassium in the same value which doesn’t makes sense since they’re completely different scales of measurement. We could try to address this by separating them out by some aesthetic (e.g. colour), so let’s try that:
ggplot() +
geom_boxplot(data = df_long, mapping = aes(x = year, y = value,
colour = element))
Ok so now we have grouped box plots. I suppose in theory we could leave it like this, but I am not personally a fan. Let’s add in the necessary components and then think about faceting:
ggplot() +
geom_boxplot(data = df_long, mapping = aes(x = year, y = value,
colour = element)) +
labs(x = "Year", y = "Amount of Element (mg/L)") +
ggthemes::theme_base()
Nope, still heinous! Time to facet! We’ll make this on two panels by faceting using facet_wrap() and operating on the element variable. I also want the plot to display as two plots stacked on top of one another, so I can refer to the number of columns as 1 to force the facet_wrap() to do this.
ggplot() +
geom_boxplot(data = df_long, mapping = aes(x = year, y = value,
colour = element)) +
labs(x = "Year", y = "Amount of Element (mg/L)") +
ggthemes::theme_base() +
facet_wrap(~element,
# I want this to be one column with two rows
ncol = 1)
Ok, we’re getting there! However, the scales for both are not helping. Let’s fix it so the sodium can vary within it’s max & min, and the potassium can vary within it’s (different) max & min. We can do this with the scales argument in the facet_wrap.
ggplot() +
geom_boxplot(data = df_long, mapping = aes(x = year, y = value,
colour = element)) +
labs(x = "Year", y = "Amount of Element (mg/L)") +
ggthemes::theme_base() +
facet_wrap(~element,
# I want this to be one column with two rows
ncol = 1,
scales = "free")
Okay, this is looking better! Now it feels like we can actually see the spread of the data. We can still make some aesthetic changes however. Let’s change to a fill not a colour on our box plot:
ggplot() +
geom_boxplot(data = df_long, mapping = aes(x = year, y = value,
fill = element)) +
labs(x = "Year", y = "Amount of Element (mg/L)") +
ggthemes::theme_base() +
facet_wrap(~element,
# I want this to be one column with two rows
ncol = 1,
scales = "free")
Good, and now we’ll change the actual colours we’re using with scale_fill_manual():
ggplot() +
geom_boxplot(data = df_long, mapping = aes(x = year, y = value,
fill = element)) +
labs(x = "Year", y = "Amount of Element (mg/L)") +
ggthemes::theme_base() +
facet_wrap(~element,
# I want this to be one column with two rows
ncol = 1,
scales = "free") +
scale_fill_manual(values = c("#90ee90", "#F8E473"))
I would also like to change the legend to have the full names of the elements. Recall we can also use the scale_fill_manual() to perform this:
ggplot() +
geom_boxplot(data = df_long, mapping = aes(x = year, y = value,
fill = element)) +
labs(x = "Year", y = "Amount of Element (mg/L)") +
ggthemes::theme_base() +
facet_wrap(~element,
# I want this to be one column with two rows
ncol = 1,
scales = "free") +
scale_fill_manual("Element",
values = c("#90ee90", "#F8E473"),
labels = c("Potassium (K)",
"Sodium (Na)"))
Looks good! But one last thing. Somewhere along the way, there have developed some ugly labels on the top of each of the plots. We could expand them so they have the same thing as the legend. This is done via the labeller argument inside the call to facet_grid(). We pass a vector of matching values to the as_labeller() function and that goes as the argument to labeller. Yes, it’s all annoyingly complicated …
ggplot() +
geom_boxplot(data = df_long, mapping = aes(x = year, y = value,
fill = element)) +
labs(x = "Year", y = "Amount of Element (mg/L)") +
ggthemes::theme_base() +
facet_wrap(~element,
# I want this to be one column with two rows*
ncol = 1,
scales = "free",
labeller = as_labeller(c(`k` = "Potassium (K)",
`na` = "Sodium (Na)"))) +
scale_fill_manual("Element",
values = c("#90ee90", "#F8E473"),
labels = c("Potassium (K)", "Sodium (Na)"))
Okay, well there we have it!! I would personally argue that you could likely leave off either the legend or the title as you really only need one of the two, but for the sake of practice, we can leave it as is.