Skip to contents

Continuous Rank Probability Score (CRPS) for an ensemble.

Source: R/ens_crps.R
ens_crps.Rd

The CRPS and its decomposition are computed as columns in a harp_list, or harp_ens_grid_df object. Typically the scores are aggregated over lead time, but other grouping variables cam be chosen.

Usage

ens_crps(
  .fcst,
  parameter,
  groupings = "lead_time",
  num_ref_members = NA,
  keep_full_output = FALSE,
  show_progress = TRUE,
  ...
)

# S3 method for harp_ens_point_df
ens_crps(
  .fcst,
  parameter,
  groupings = "lead_time",
  num_ref_members = NA,
  keep_full_output = FALSE,
  show_progress = TRUE,
  fcst_model = NULL,
  ...
)

Arguments

.fcst

A harp_df or harp_list object with tables that have a column for observations, or a single forecast table.

parameter

The name of the column for the observations data. Can be the column name, quoted, or unquoted. If a variable it should be embraced - i.e. wrapped in {{}}.

groupings

The groups for which to compute the scores. See group_by for more information of how grouping works.

num_ref_members

For "fair" scores, the score is scaled to be valid for this number of ensemble members. Set to NA (the default) to not modify the score.

show_progress

Logical - whether to show progress bars. Defaults to TRUE.

...

Reserved for methods.

fcst_model

The name of the forecast model to use in the fcst_model column of the output. If the function is dispatched on a harp_list object, the names of the harp_list are automatically used.

Value

An object of the same format as the inputs but with data grouped for the groupings column(s) and columns for crps, crps_pot

and crps_rel.