Introduction
Yesterday Richard James posted about “hythergraphs”, which he’d seen on Toolik Field Station’s web site.
Hythergraphs show monthly weather parameters for an entire year, plotting temperature against precipitation (or other paired climate variables) against each other for each month of the year, drawing a line from month to month. When contrasting one climate record against another (historic vs. contemporary, one station against another), the differences stand out.
I was curious to see how easy it would be to produce one with R and ggplot.
Data
I’ll produce hythergraphs, one that compares Fairbanks Airport data against the data collected at our station on Goldstream Creek for the period of record for our station (2011‒2022) and one that compares the Fairbanks Airport station data from 1951‒2000 against data from 2001‒2022 (similar to what Richard did).
I’m using the following R packages:
library(tidyverse)
library(RPostgres)
library(lubridate)
library(scales)
I’ll skip the part where I pull the data from the GHCND database. What we need is a table of observations that look like this. We’ve got a categorical column (station_name), a date column, and the two climate variables we’re going to plot:
# A tibble: 30,072 × 4 station_name dte PRCP TAVG <chr> <date> <dbl> <dbl> 1 GOLDSTREAM CREEK 2011-04-01 0 -17.5 2 GOLDSTREAM CREEK 2011-04-02 0 -15.6 3 GOLDSTREAM CREEK 2011-04-03 0 -8.1 4 GOLDSTREAM CREEK 2011-04-04 0 -5 5 GOLDSTREAM CREEK 2011-04-05 0 -5 6 GOLDSTREAM CREEK 2011-04-06 0.5 -3.9 7 GOLDSTREAM CREEK 2011-04-07 0 -8.3 8 GOLDSTREAM CREEK 2011-04-08 2 -5.85 9 GOLDSTREAM CREEK 2011-04-09 0.5 -1.65 10 GOLDSTREAM CREEK 2011-04-10 0 -4.45 # ℹ 30,062 more rows
From that raw data, we’ll aggregate to year and month, calculating the montly precipitation sum and mean average temperature, then aggregate to station and month, calculating the mean monthly precipitation and temperature.
The final step adds the necessary aesthetics to produce the plot using ggplot. We’ll draw the monthly scatterplot values using the first letter of the month, calculated using month_label = substring(month.name[month], 1, 1) below. To draw the lines from one month to the next we use geom_segement and calculate the ends of each segment by setting xend and yend to the next row’s value from the table.
One flaw in this approach is that there’s no line between December and January because there is no “next” value in the data frame. This could be fixed by seperately finding the January positions, then passing those to lead() as the default value (which is normally NA).
airport_goldstream <- pivot |>
filter(dte >= "2010-04-01") |>
# get monthly precip total, mean temp
mutate(
year = year(dte),
month = month(dte)
) |>
group_by(station_name, year, month) |>
summarize(
sum_prcp_in = sum(PRCP, na.rm = TRUE) / 25.4,
mean_tavg_f = mean(TAVG, na.rm = TRUE) * 9 / 5.0 + 32,
.groups = "drop"
) |>
# get monthy means for each station
group_by(station_name, month) |>
summarize(
mean_prcp_in = mean(sum_prcp_in),
mean_tavg_f = mean(mean_tavg_f),
.groups = "drop"
) |>
# add month label, line segment ends
arrange(station_name, month) |>
group_by(station_name) |>
mutate(
month_label = substring(month.name[month], 1, 1),
xend = lead(mean_prcp_in),
yend = lead(mean_tavg_f)
)
Here’s what that data frame looks like:
# A tibble: 24 × 7 # Groups: station_name [2] station_name month mean_prcp_in mean_tavg_f month_label xend yend <chr> <dbl> <dbl> <dbl> <chr> <dbl> <dbl> 1 FAIRBANKS INTL AP 1 0.635 -6.84 J 0.988 -0.213 2 FAIRBANKS INTL AP 2 0.988 -0.213 F 0.635 11.5 3 FAIRBANKS INTL AP 3 0.635 11.5 M 0.498 33.1 4 FAIRBANKS INTL AP 4 0.498 33.1 A 0.670 51.2 5 FAIRBANKS INTL AP 5 0.670 51.2 M 1.79 61.3 6 FAIRBANKS INTL AP 6 1.79 61.3 J 2.41 63.1 7 FAIRBANKS INTL AP 7 2.41 63.1 J 2.59 57.9 8 FAIRBANKS INTL AP 8 2.59 57.9 A 1.66 46.5 9 FAIRBANKS INTL AP 9 1.66 46.5 S 1.04 29.5 10 FAIRBANKS INTL AP 10 1.04 29.5 O 1.16 5.21 # ℹ 14 more rows
Plots
Here’s the code to produce the plot. The month labels are displayed using geom_label, and the lines between months are generated from geom_segment.
airport_v_gsc <- ggplot(
data = airport_goldstream,
aes(x = mean_prcp_in, y = mean_tavg_f, color = station_name)
) +
theme_bw() +
geom_segment(aes(xend = xend, yend = yend, color = station_name)) +
geom_label(aes(label = month_label), show.legend = FALSE) +
scale_x_continuous(
name = "Monthly Average Precipitation (inches liquid)",
breaks = pretty_breaks(n = 10)
) +
scale_y_continuous(
name = "Monthly Average Tempearature (°F)",
breaks = pretty_breaks(n = 10)
) +
scale_color_manual(
name = "Station",
values = c("darkorange", "darkcyan")
) +
theme(
legend.position = c(0.8, 0.2),
legend.background = element_rect(
fill = "white", linetype = "solid", color = "grey80", size = 0.5
)
) +
labs(
title = "Monthly temperature and precipitation",
subtitle = "Fairbanks Airport and Goldstream Creek Stations, 2011‒2022"
)
You can see from the plot that we are consistently colder than the airport, curiously more dramatically in the summer than winter. The airport gets slighly more precipitation in winter, but our summer precipitation is significantly higher, especially in August.
The standard plot to display this information would be two bar charts with one plot showing the monthly mean temperature for each station, and a second plot showing precipitation. The advantage of such a display is that the differences would be more clear, and the bars could include standard errors (or standard deviation) that would help provide an idea of whether the differences between stations are statistically significant or not.
For example (the lines above the bars are one standard deviation above or below the mean):
In this plot of the same data, you can tell from the standard deviation lines that the precipitation differences between stations are probably not significant, but the cooler summer temperatures at Goldstrem Creek may be.
If we calculate the standard deviations of the monthly means, we can use geom_tile to draw significance boxes around each monthly value in the hytherplot as Richard suggests in his post. Here’s the ggplot geom to do that:
geom_tile(
aes(width = 2*sd_prcp_in, height = 2*sd_tavg_f, fill = station_name),
show.legend = FALSE, alpha = 0.25
) +
And the updated plot:
This clearly shows the large variation in precipitation, and if you carefully compare the boxes for a particular month, you can draw concusions similar to what is made fairly clear in the bar charts. For example, if we focus on August, you can see that the Goldstream Creek precipitation box clearly overlaps that of the airport station, but the temperature ranges do not overlap, suggesting that August temperatures are cooler at Goldstream Creek but that while precipitation is much higher, it’s not statistically significant.
Airport station, different time periods
Here’s the plot for the airport station that is similar to the plot Richard created (I used different time periods).
This plot demonstrates that while temperatures have increased in the last two decades, it’s the differences in the pattern of precipitation that stands out, with July and August precipitation much larger in the last 20 years. It’s also curious that February and April precipitation is higher, but the differences are smaller in the other winter months. This is a case where some sense of the distribution of the values would be useful.