quantileScore: Quantile Score

In verification: Weather Forecast Verification Utilities

Description

Calculates verification statistics for quantile forecasts.

Usage

quantileScore(obs, pred, p, breaks, ...)

Arguments

obs	Vector of observations
pred	Vector of quantile forecasts
p	Probability level of quantile forecasts [0,1].
breaks	Values used to bin the forecasts
...	Optional arguments

Details

This function calculates the quantile score and its decomposition into reliability, resolution, and uncertainty. Note that a careful binning (discretization of forecast values) is necessary to obtain good estimates of reliability and resolution (see Bentzien and Friederichs (2013) for more details).

Value

qs.orig	Quantile score for original data
qs	Quantile score for binned data
qs.baseline	Quantile score for climatology
ss	Quantile skill score
qs.reliability	Reliability part of the quantile score
qs.resolution	Resolution part of the quantile score
qs.uncert	Uncertainty part of the quantile score
y.i	Discretized forecast values – defined as the mean value of forecasts in each bin
obar.i	Conditional observed quantiles
prob.y	Number of forecast-observation pairs in each bin
obar	Climatology – unconditional sample quantile of observations
breaks	Values used to bin the forecasts
check	Difference between original quantile score and quantile score decomposition

Note

This function is used within verify.

Author(s)

Sabrina Bentzien

References

Bentzien, S. and Friederichs, P. (2013) Decomposition and graphical portrayal of the quantile score. Submitted to QJRMS.

See Also

check.func, qrel.plot

Examples

data(precip.ensemble)

#Observations are in column 3
obs <- precip.ensemble[,3]

#Forecast values of ensemble are in columns 4 to 54
eps <- precip.ensemble[,4:54]

#Quantile forecasts from ensemble
p <- 0.9
qf <- apply(eps,1,quantile,prob=p,type=8)

#generate equally populated binnng intervals
breaks <- quantile(qf,seq(0,1,length.out=11))

qs <- quantileScore(obs,qf,p,breaks)
## Not run:  qrel.plot(qs) 

verification documentation built on May 2, 2019, 1:24 a.m.