Skip to contents

Plot model performance metrics

Source: R/plot_assess.R
plot_assess.Rd

Visualise the model performance statistics returned by assess_aeme. Four complementary views are available via type:

  • "dot" - Cleveland dot plot, one facet per metric with a free x-axis, points coloured by model. The general-purpose default: it keeps metrics on their own scales so native-unit errors (bias, rmse, nmae) and bounded skill scores (nse, kge, d2, r) are not forced onto a common axis.

  • "heatmap" - Tile plot of model (x) against variable (y), faceted by metric. Fill is the value rescaled to 0-1 within each metric/variable and oriented so that 1 is the best-performing model; the raw value is printed in each tile.

  • "taylor" - Normalised Taylor diagram (Taylor, 2001), one point per model per variable, encoding the correlation and the ratio of modelled to observed standard deviation. The observations plot at (1, 0).

  • "target" - Target diagram (Jolliff et al., 2009), plotting normalised bias against normalised unbiased RMSD. Points inside the unit circle have a total RMSD smaller than the observed standard deviation.

Usage

plot_assess(
  aeme,
  model,
  var_sim,
  type = c("dot", "heatmap", "taylor", "target"),
  metrics = c("bias", "rmse", "nmae", "nse", "kge", "d2", "r")
)

Arguments

aeme

Aeme object.

model

character vector; models to use. One or more of "dy_cd", "glm_aed", "gotm_wet", "simstrat_aed2". Defaults to all models if not found in aeme.

var_sim

string; of variable to plot

type

character; which view to draw. One of "dot" (default), "heatmap", "taylor" or "target".

metrics

character vector; metrics to include in the "dot" and "heatmap" views. Defaults to c("bias", "rmse", "nmae", "nse", "kge", "d2", "r"). Ignored for "taylor" and "target".

Value

A ggplot2 object.

Details

"dot" and "heatmap" are built from assess_aeme output; "taylor" and "target" are computed from the paired observed/modelled values returned by get_var.

See also

assess_aeme for the underlying statistics.

Examples

if (FALSE) { # \dontrun{
  metrics <- assess_aeme(aeme = aeme, model = c("glm_aed", "gotm_wet"))
  plot_assess(aeme = aeme, model = c("glm_aed", "gotm_wet"))
  plot_assess(aeme = aeme, type = "heatmap")
  plot_assess(aeme = aeme, type = "taylor", var_sim = "HYD_temp")
} # }