Plotting Model Quality Data

library(mqor)

There are three plot types built into mqor(). To demonstrate them, we first need to calculate some statistics to use as inputs.

stats_long <- summarise_mqo_stats(demo_longterm, "PM10", term = "long")

stats_short <- summarise_mqo_stats(demo_shortterm, "PM10", term = "short")

Plot Types

‘Comparison’ Bars

To visualise measurement data directly, it can be useful to start with plot_comparison_bars().

For long-term data, this draws a bar chart of both observed and modelled values, with an ‘acceptability range’ (AR) drawn for the observation bars. Modelled values meeting MQI[long] <= 1 for a particular sampling point are within the corresponding observed value’s AR.

plot_comparison_bars(stats_long)

For short-term data, plot_comparison_bars() shows RMSE & RMSU*[0] (RMSU multiplied by the square root of 1 + beta squared).

plot_comparison_bars(stats_short)

You will likely notice that both statistics objecst are fed into the same function; the function is intelligent enough to understand if the stats are for short- or long-term data, and choose appropriate statsitics to visualise.

Model Quality Bars

To actually apply the MQO, plot_mqi_bars() will create an ordered bar chart showing the MQI for each sampling point. The dashed horizontal line shows the MQI[90th]. These plots are very similar regardless of whether the short- or long-term statistics are of interest.

plot_mqi_bars(stats_long)

plot_mqi_bars(stats_short)

Model Quality Scatter

For large numbers of sites, a complementary ‘scatter’ chart can be useful as a bar chart will grow more crowded.

For long-term data, plot_mqi_scatter() shows a scatter of modelled and observed concentrations, with the AR as a shaded area.

plot_mqi_scatter(stats_longterm = stats_long)

For short-term data, plot_mqi_scatter() shows a special ‘target’ diagram. If both sets of stats are provided, the target diagram also labels MQI long.

plot_mqi_scatter(stats_shortterm = stats_short, stats_longterm = stats_long)

These plots vary somewhat for ‘mixed’ networks of both fixed and indicative sites, incorporating colour to show both acceptability ranges on one plot.

demo_longterm |>
  dplyr::mutate(
    type = c(rep("fixed", 10), rep("indicative", 5))
  ) |>
  summarise_mqo_stats("PM10") |>
  (\(x) plot_mqi_scatter(stats_longterm = x))()
#> ! term assumed to be 'long'.
#> ℹ If this is incorrect, please specify the data's term using the term argument.

demo_shortterm |>
  dplyr::mutate(
    type = c(rep("fixed", 50), rep("indicative", 25))
  ) |>
  summarise_mqo_stats("PM10") |>
  (\(x) plot_mqi_scatter(stats_shortterm = x))()
#> ! term assumed to be 'short'.
#> ℹ If this is incorrect, please specify the data's term using the term argument.

Model Quality Report

Both short- and long-term statistics can be viewed simultaneously in the MQI “report” format, which shows both the short- and long-term MQI as well as all complementary indicators. Much like the MQI scatter plots, this plot distinguishes between fixed and indicative sites.

plot_mqi_report(stats_shortterm = stats_short, stats_longterm = stats_long)

Customisation

Users have some basic control over the colours used in each plot. These arguments always start with color_ and take either R colour names or hex codes.

plot_comparison_bars(
  stats_long,
  color_mod = "#003776",
  color_obs = "#F8AE21",
  color_outline = "#404040"
)

The static plots are built using ggplot2, a highly capable and customisable plotting language. A particular strength of ggplot2 is its allowance for post-hoc customisations. If you aren’t familiar with ggplot2, please refer to https://ggplot2-book.org/ for guidance from its authors. The below code chunk demonstrates a few options available to users, including changing the plot theme, axis scales, annotations, and so on.

plot_comparison_bars(stats_short, color_outline = "white") +
  # add a new theme
  ggplot2::theme_dark() +
  ggplot2::theme(legend.position = "bottom") +
  # change labels
  ggplot2::labs(x = "Monitoring Sites", fill = "Statistic") +
  # change y scale
  ggplot2::scale_y_continuous(limits = c(0, 40)) +
  # add an annotation
  ggplot2::annotate(
    x = "S11",
    y = 20,
    geom = "text",
    color = "white",
    label = "Look\nat this!"
  ) +
  ggplot2::annotate(
    x = "S11",
    xend = "S11",
    y = 17.5,
    yend = 14,
    geom = "segment",
    color = "white",
    arrow = ggplot2::arrow(length = ggplot2::unit(0.25, "cm"))
  )
#> Scale for y is already present.
#> Adding another scale for y, which will replace the existing scale.

For complete customisation, users are naturally free to re-plot the summary statistics in any way they find useful. ggplot2 is a good candidate for this, but users can choose to use base R plotting, a different plotting library, a JavaScript wrapper like plotly or echarts4r, or even export the statistics tables for use in other software like Excel or Python.

library(ggplot2)

summarise_mqo_stats(demo_shortterm, pollutant = "PM10")$by_site |>
  ggplot(aes(x = mean_obs, y = mean_mod)) +
  geom_smooth(method = "lm") +
  geom_pointrange(aes(ymin = mean_mod - sd_mod, ymax = mean_mod + sd_mod), alpha = 0.5) +
  geom_pointrange(aes(xmin = mean_obs - sd_obs, xmax = mean_obs + sd_obs), alpha = 0.5) +
  coord_fixed() +
  theme_bw() +
  labs(
    x = "Observed Values",
    y = "Modelled Values"
  )
#> ! term assumed to be 'short'.
#> ℹ If this is incorrect, please specify the data's term using the term argument.
#> `geom_smooth()` using formula = 'y ~ x'

Dynamic Plots

Each of the plotting functions has an interactive argument. When TRUE, the JavaScript library plotly is used to produce a dynamic chart. These outputs are naturally inappropriate to include in a typical PDF report, but are better suited for web applications, Quarto or rmarkdown dynamic reports, or exploratory analysis.

plot_comparison_bars(stats_long, interactive = TRUE)

plot_mqi_bars(stats_long, interactive = TRUE)

plot_mqi_scatter(stats_longterm = stats_long, interactive = TRUE)

plot_mqi_scatter(stats_shortterm = stats_short, stats_longterm = stats_long, interactive = TRUE)

plot_mqi_report(stats_short, stats_long, interactive = TRUE)