This article presents the “Unbraided Ribbon Problem” in which
geom_ribbon() incorrectly fills the area between two alternating lines with two different colors. To fix the problem, we use
geom_braid() from ggbraid with
method = 'line'.
Let’s compare the temperatures of two cities in the United States: New York, New York and San Francisco, California.
library(ggplot2) library(ggbraid) library(dplyr) library(tidyr) data(temps) temps#> # A tibble: 730 × 3 #> city date avg #> <chr> <date> <dbl> #> 1 New York 2021-01-01 36.5 #> 2 New York 2021-01-02 43.5 #> 3 New York 2021-01-03 36 #> 4 New York 2021-01-04 39 #> 5 New York 2021-01-05 39 #> 6 New York 2021-01-06 37.5 #> 7 New York 2021-01-07 35.5 #> 8 New York 2021-01-08 32 #> 9 New York 2021-01-09 30.5 #> 10 New York 2021-01-10 35 #> # … with 720 more rows
New York or
date is a calendar date in the
YYYY-MM-DD format, and
avg is the average temperature recorded in degrees Fahrenheit (°F) and rounded to the nearest half degree.2
What do the daily average temperatures look like?
ggplot(temps) + geom_line(aes(x = date, y = avg, linetype = city))
We see much higher variability in temperatures in New York compared with San Francisco. This makes sense — New York is in the Northeastern US and experiences hot, humid summers and cold, occassionally snowy winters. San Francisco is on the West Coast and its Mediterranean climate means its temperature does not change much season to season.
Before we proceed further, let’s clean up the plot a bit and assign it to a variable
p so we can reuse it throughout the article.
<- ggplot() + p geom_line(aes(x = date, y = avg, linetype = city), data = temps) + scale_x_date(date_breaks = "1 month", date_labels = "%b") + scale_y_continuous( breaks = seq(20, 90, by = 10), labels = function(x, ...) format(paste(x, "°F"), ...), limits = c(18, 90) + ) guides(fill = "none") + labs( title = "Average Daily Temperatures in 2021", linetype = NULL, y = NULL, x = NULL + ) theme_minimal(base_size = 14) + theme( plot.title = element_text(size = 15), plot.title.position = "plot", legend.position = c(0.75, 1.06), legend.direction = "horizontal", legend.key.size = unit(2, "line"), legend.text = element_text(size = 12), panel.grid.major.x = element_line(size = 0.4), panel.grid.major.y = element_line(size = 0.4), panel.grid.minor.x = element_blank(), panel.grid.minor.y = element_blank() ) p
Let’s fill the area between the two lines. We can do so with
geom_ribbon() requires three aesthetics:
ymax. We can map
x as we did in
geom_line(). However, we’ll need to transform
temps to create new variables that we can map to
We can pivot
pivot_wider() from the
tidyr package, taking column names from
city and values from
avg. Call the new data frame
<- temps %>% temps_wide pivot_wider(names_from = city, values_from = avg) %>% rename(ny = `New York`, sf = `San Francisco`) temps_wide#> # A tibble: 365 × 3 #> date ny sf #> <date> <dbl> <dbl> #> 1 2021-01-01 36.5 51.5 #> 2 2021-01-02 43.5 50 #> 3 2021-01-03 36 50.5 #> 4 2021-01-04 39 54.5 #> 5 2021-01-05 39 50 #> 6 2021-01-06 37.5 50.5 #> 7 2021-01-07 35.5 53 #> 8 2021-01-08 32 52.5 #> 9 2021-01-09 30.5 52 #> 10 2021-01-10 35 50.5 #> # … with 355 more rows
Now we can add a new layer to
ymax.3 Finally, add some transparency with
alpha = 0.3.
+ p geom_ribbon( aes(x = date, ymin = ny, ymax = sf), data = temps_wide, alpha = 0.3 )
geom_ribbon() we’ve added a light grey ribbon that runs between the two lines.
On second thought… what if we used two colors for the ribbon? We could have one color when New York is hotter than San Francisco and another color when New York is colder than San Francisco.
This shouldn’t be hard to do. Map
sf > ny to
+ p geom_ribbon( aes(x = date, ymin = ny, ymax = sf, fill = sf > ny), data = temps_wide, alpha = 0.7 )
Chaos. What happened?
Is this a bug in
No, it’s not a bug. The problem is that we haven’t dealt with line intersections properly.
For example, consider rows 80-82 from
After we have pass
geom_ribbon() and map
sf > ny to
fill, we get the following:
x is the integer representation of
date, the number of days since January 1, 1970, the “Unix epoch”)
Ok, note the middle row.
ymax are equal here, so this is a point where the two lines intersect. It turns out that
geom_ribbon() requires two rows for every line intersection, one row where
FALSE and another row where
So we must insert a new row in the data, yielding the following:
We call this process braiding.
We need to braid the ribbon where the lines intersect.
And the intersection described here is not the only type that requires braiding.
There are instances where the two lines intersect between two rows in the data. In these cases, we must use a mathematical formula to determine the exact point at which the lines intersect and braid the ribbon accordingly. There are also instances where both lines are vertical at the same
x, an uncommon situation but one that produces an infinite number of intersection points and requires braiding to fix.
The functions in ggbraid take care of all the braiding for you. Simply replace
+ p geom_braid( aes(x = date, ymin = ny, ymax = sf, fill = sf > ny), data = temps_wide, alpha = 0.7 )#> `geom_braid()` using method = 'line'
There we go!
Notice the message from
geom_braid() that it is using
method = 'line'. Since we’ve drawn lines with
geom_line() we must use
method = 'line' to determine the point at which the lines intersect when the intersection occurs between two rows in the data. We can silence this message by explicity including
method = 'line' within
geom_braid() takes the data provided, performs the necessary braiding operations on it with
stat_braid(), and passes the result to
geom_ribbon() for drawing. If we’d like, we can still use
geom_ribbon() and set
stat = 'braid'.
+ p geom_ribbon( aes(x = date, ymin = ny, ymax = sf, fill = sf > ny), data = temps_wide, stat = "braid", method = "line", alpha = 0.7 )
This is the same plot as before. We’ve also silenced the message by including
method = 'line'.
Finally, it may be helpful to label the ribbon colors so it’s clear what they represent. This can happen in a legend (which we’ve turned off with the
guides(fill = "none") layer in
p). Another possibility is to provide text annotations on the plot.
<- scales::hue_pal()(2) # ggplot2 default color palette hues + p geom_braid( aes(x = date, ymin = ny, ymax = sf, fill = sf > ny), data = temps_wide, method = "line", alpha = 0.7 + ) annotate("text", x = as.Date("2021-09-10"), y = 84, size = 4, label = "NY hotter than SF", hjust = 0, color = hues) + annotate("text", x = as.Date("2021-02-20"), y = 23, size = 4, label = "NY colder than SF", hjust = 0, color = hues)
It is difficult to pull data from the NWS. It does not provide the data via an API and the data it returns through its point-and-click interface isn’t in plain text format! To make matters worse, you can only retrieve data from a city one month at a time. For San Francisco, visit weather.gov/wrh/climate?wfo=mtr and choose “San Francisco City, CA”, “Daily data for a month”, and a month from 2021; for New York, visit weather.gov/wrh/climate?wfo=okx and choose “NY-Central Park Area”, “Daily data for a month”, and a month from 2021. Copy and paste the data into spreadsheet software for further processing.↩︎
For those who use degrees Celsius: 0°C is 32°F, 10°C is 50°F, 20°C is 68°F, and 30°C is 86°F.↩︎
Why not the other way around, with
ymax? That’s fine too because the lines alternate over/under one another.↩︎