Plot observed against simulated burn-severity classes

Plot observed against simulated burn-severity classes

Usage

plot_calibration_severity(
  stats,
  weight_by = c("events", "area"),
  classes = c(3, 5)
)

Arguments

stats

A run_calibration_validation() summary carrying $observed$primary$severity_dist.

weight_by

"events" (one vote per fire) or "area" (weight each fire by the area it burned). See Weighting.

classes

3 to collapse to low/medium/high before plotting – the three categories a thresholded observed reference usually distinguishes – or 5 for the raw Dynamic Fire classes.

Value

A ggplot object.

Weighting

weight_by is the important argument. The objective's severity term counts fire EVENTS: it bins each event's mean severity and gives every fire one vote, whatever its size (weight_by = "events", the default, so the figure matches what was scored). An observed severity reference is almost always a share of burned AREA – so many pixels or hectares in each class – and where severity rises with fire size the two summaries of the same simulation can differ sharply, and can even disagree in sign about which way the model is off. weight_by = "area" weights each event by the area it burned, which is the like-for-like comparison against an area-weighted reference.

Both options still summarise each fire by its mean severity, so neither recovers the within-fire variation that a per-pixel reference retains.

See also

Other calibration plots: calibration_events(), calibration_plot_palette(), plot_calibration_area_by_fuel(), plot_calibration_convergence(), plot_calibration_fire_counts(), plot_calibration_fire_sizes(), plot_calibration_loss(), plot_calibration_severity_by_size()