Quantile Score
Calculates verification statistics for quantile forecasts.
quantileScore(obs, pred, p, breaks, ...)
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 |
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).
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 |
This function is used within verify.
Sabrina Bentzien
Bentzien, S. and Friederichs, P. (2013) Decomposition and graphical portrayal of the quantile score. Submitted to QJRMS.
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)
Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.