CRPS {SeaVal}

Continuous Ranked Probability Score

Description

Taking CRPSs of ensemble forecasts stored in long data tables:

Usage

CRPS(
  dt,
  f,
  o = "obs",
  by = by_cols_ens_fc_score(),
  pool = "year",
  mem = "member",
  dim.check = TRUE,
  ens_size_correction = FALSE
)

Arguments

dt

Data table containing predictions and observations.

f

column name of the forecasts. May not be called 'f'

o

column name of the observations.

by

column names of grouping variables, all of which need to be columns in dt. Default is to group by all instances of month, season, lon, lat, system and lead_time that are columns in dt.

pool

column name(s) for the variable(s) over which is averaged. Typically just 'year'.

mem

Name of the column identifying the ensemble member.

dim.check

Logical. If True, a simple test whether the dimensions match up is conducted: The data table should only have one row for each level of c(by,pool,mem)

ens_size_correction

logical. If TRUE, the CRPS is corrected for sample size (see Ferro et al. 2008: 'On the effect of ensemble size on the discrete and continuous ranked probability scores'). This is slower, but you should do it if you compare ensembles of different size.

Value

A data table with the scores

Examples

dt = data.table(fc = 1:4,obs = c(4,4,7,7),member = c(1,2,1,2),year = c(1999,1999,2000,2000))
CRPS(dt,f = 'fc')