Time Series and Trends

library(ggopenair)
library(ggplot2)
library(dplyr)
#> 
#> Attaching package: 'dplyr'
#> The following objects are masked from 'package:stats':
#> 
#>     filter, lag
#> The following objects are masked from 'package:base':
#> 
#>     intersect, setdiff, setequal, union


# not needed - for comparisons
library(openair)

Time Plots

Unlike in openair, ggopenair does not have a dedicated time_plot() function. This is because creating time series in ggplot2 is already so simple. The “legacy” timePlot() looks like this:

timePlot(marylebone, pollutant = "nox", avg.time = "month")

The legacy openair time plot

The equivalent in ggplot2 is shown below.

plt <-
  marylebone %>%
  time_average(avg_time = "month") %>%
  ggplot(aes(x = date, y = nox)) +
  geom_line() +
  theme_bw() +
  labs(x = NULL, y = quickText("nox"))
#> Warning: Returning more (or less) than 1 row per `summarise()` group was deprecated in
#> dplyr 1.1.0.
#> ℹ Please use `reframe()` instead.
#> ℹ When switching from `summarise()` to `reframe()`, remember that `reframe()`
#>   always returns an ungrouped data frame and adjust accordingly.
#> ℹ The deprecated feature was likely used in the ggopenair package.
#>   Please report the issue to the authors.
#> This warning is displayed once every 8 hours.
#> Call `lifecycle::last_lifecycle_warnings()` to see where this warning was
#> generated.

While there are more lines of code to achieve the same thing, a ggplot2 plot is much more customisable. ggopenair even contains tools to help with common issues. For example, we can add limit values using the scale_y_limitval() function.

plt +
  scale_y_limitval(c(150, 350), "red", c("Getting\nHigher", "Too high!"), c(2, 1))

There is also no equivalent for smoothPlot() as that is already well served by geom_smooth().

plt + geom_smooth()
#> `geom_smooth()` using method = 'loess' and formula = 'y ~ x'

Temporal Variations

trend_variation() works much the same way as timeVariation(), but returns a patchwork assemblage. These can be treated similarly to a ggplot2 object, but the ampersand (&) symbol is used to style all plots together.

trend_variation(marylebone, c("no2", "pm2.5"), normalise = TRUE) &
  theme_bw() &
  scale_fill_discrete(
    aesthetics = c("colour", "fill"),
    labels = scales::label_parse()(c("NO[2]", "PM[2.5]"))
  )

The return argument can be used to obtain specific panels. This could be useful if not all panels are meaninful for your data, or you want to customise individual panels.

hour_panel <- trend_variation(marylebone, "o3", return = "hour")

avg <- mean(hour_panel$data$Mean)

hour_panel +
  theme_classic() +
  scale_color_manual(values = "hotpink", aesthetics = c("colour", "fill")) +
  scale_y_limitval(marker_values = avg, marker_labels = paste0("Mean (", round(avg, 2), ")")) +
  theme(legend.position = "none")

Time Proportion Plots

trend_prop() behaves almost identically to timeProp(), albeit with fewer arguments as more can be controlled using labs(), scales_*_*(), and so on.

marylebone %>%
  filter(format(date, "%Y") == 2019) %>%
  trend_prop(
    pollutant = "pm10",
    avg_time = "3 day",
    proportion = "wd"
  ) +
  theme_bw() +
  scale_fill_brewer(palette = "Dark2")
#> Warning: 432 missing wind direction line(s) removed

proportion behaves similarly to facet (type in the original openair). If a column isn’t present in the data set (or it is numeric), it will use openair::cutdata() to parse it and work out how it can be used to cut the data. The recommendation, however, is to get your categories in order before you use proportion — then you have full control over and understanding of the output. For example, in the below plot, we use cut_wd() to pre-cut the wind directions into bins — in this case, fewer bins by setting resolution to “low”.

marylebone %>%
  mutate(wd = cut_wd(wd, "low")) %>%
  filter(
    format(date, "%Y") == 2019,
    !is.na(wd)
  ) %>%
  trend_prop(
    pollutant = "pm10",
    avg_time = "3 day",
    proportion = "wd"
  ) +
  theme_bw() +
  scale_fill_brewer(palette = "Dark2")

Trend Heat Maps

Much like with the time proportion plots, trend_level() is very similar to trendLevel().

plt <-
  trend_level(marylebone, "nox", "hour", "month", "year") +
  scale_opencolours_c() +
  theme_minimal() +
  labs(fill = quickText("Mean NOx"), y = NULL)

plt

A minor exception is that, when possible, axes are automatically parsed as numeric/integer rather than always being factors. This avoids label overlap, and allows users to use continuous scales for further transformations and annotations. For example, here each day is split into four, 6-hour segments.

plt +
  scale_x_continuous(breaks = seq(0, 24, 6)) +
  geom_vline(xintercept = c(6, 12, 18))

Calendar Plots

Calendar plots once again behave similarly, but the annotations are dealt with using bespoke functions. These make it easier to customise or layer the annotations on top of one another.

marylebone |>
  openair::selectByDate(year = 2019) |>
  trend_calendar("nox") +
  annotate_calendar_text("date") +
  annotate_calendar_wd(colour = "black") +
  scale_opencolours_c()

An extra feature of trend_calendar() is that multiple years can now be easily plotted. Instead of a normal calendar, a year-month matrix is produced.

library(scales)
marylebone |>
  trend_calendar("nox") +
  scale_opencolours_c(limits = c(0, 600), oob = squish)

TheilSen