verification (version 1.42)

crps: Continuous Ranked Probability Score

Description

Calculates the crps for a forecast made in terms of a normal probability distribution and an observation expressed in terms of a continuous variable.

Usage

crps(obs, pred, ...)

Value

crps

Continous ranked probability scores

CRPS

Mean of crps

ign

Ignorance score

IGN

Mean of the ignorance score

Arguments

obs

A vector of observations.

pred

A vector or matrix of the mean and standard deviation of a normal distribution. If the vector has a length of 2, it is assumed that these values represent the mean and standard deviation of the normal distribution that will be used for all forecasts.

...

Optional arguments

Author

Matt Pocernich

References

Gneiting, T., Westveld, A., Raferty, A. and Goldman, T, 2004: Calibrated Probabilistic Forecasting Using Ensemble Model Output Statistics and Minimum CRPS Estimation. Technical Report no. 449, Department of Statistics, University of Washington. [ Available online at http://www.stat.washington.edu/www/research/reports/ ]

Examples

#  probabilistic/ binary example


x <- runif(100) ## simulated observation.
crps(x, c(0,1))

## simulated forecast in which mean and sd differs for each forecast.
frcs<- data.frame( runif(100, -2, 2), runif(100, 1, 3 ) )
crps(x, frcs)