ccrlm: Climate Conserving Recalibration (as Regression)

In jonasbhend/biascorrection: Bias Correct Ensemble Forecasts

Climate Conserving Recalibration (as Regression)

Description

Climate conserving recalibration with correction factors for each lead time without smoothing (application to daily series not recommended).

Usage

ccrlm(fcst, obs, pred = NULL, fcst.out = fcst, pred.out = pred, ...)

Arguments

fcst	n x m x k array of n lead times, m forecasts, of k ensemble members
obs	n x m matrix of veryfing observations
pred	m vector, n x m matrix, or list of m vectors or n x m matrices with additional predictors
fcst.out	array of forecast values to which bias correction should be applied (defaults to fcst)
pred.out	n vector, n x m matrix, or list of n vectors or n x m matrices with additional predictors
...	additional arguments for compatibility with other bias correction methods

Examples

nveri <- 50
ncal <- 30
nn <- nveri + ncal
ical <- seq(1, ncal)
iveri <- seq(ncal + 1, nn)
nens <- 15
signal <- rnorm(nn, sd=5)
sdfrac <- 2
fcst <- array(rnorm(nn*nens, sd=sdfrac), c(1, nn, nens)) + 
  0.6 * signal + seq(-3,3,length=nn) 
obs <- array(rnorm(nn, sd=1), c(1, nn)) + signal
fcst.deb <- biascorrection:::ccrlm(fcst[,ical,,drop=FALSE], 
                                   obs[,ical,drop=FALSE],
                                   fcst.out=fcst)
fcst.deb2 <- biascorrection:::ccrlm(fcst[,ical,,drop=FALSE], 
                                   obs[,ical,drop=FALSE],
                                   fcst.out=fcst, 
                                   pred=seq(1,ncal), 
                                   pred.out=seq(1,nn))
f.rmse <- sqrt(apply((obs[,ical, drop=FALSE] - 
            apply(fcst.deb[,ical,,drop=FALSE], 2, mean))**2, 1, mean))
f.rmse2 <- sqrt(apply((obs[,ical,drop=FALSE] - 
            apply(fcst.deb2[,ical,,drop=FALSE], 2, mean))**2, 1, mean))
f.sd <- sqrt(mean(apply(fcst.deb[,ical,,drop=FALSE], 1:2, sd)**2))
f.sd2 <- sqrt(mean(apply(fcst.deb2[,ical,,drop=FALSE], 1:2, sd)**2))
f.sd / f.rmse ## should be one
f.sd2 / f.rmse2 ## should be one
## spread to error with out-of-sample validation
f.rmse <- sqrt(apply((obs[,iveri, drop=FALSE] - 
            apply(fcst.deb[,iveri,,drop=FALSE], 2, mean))**2, 1, mean))
f.rmse2 <- sqrt(apply((obs[,iveri,drop=FALSE] - 
            apply(fcst.deb2[,iveri,,drop=FALSE], 2, mean))**2, 1, mean))
f.sd <- sqrt(mean(apply(fcst.deb[,iveri,,drop=FALSE], 1:2, sd)**2))
f.sd2 <- sqrt(mean(apply(fcst.deb2[,iveri,,drop=FALSE], 1:2, sd)**2))
f.sd / f.rmse ## should be close to one
f.sd2 / f.rmse2 ## can be close to one

jonasbhend/biascorrection documentation built on Nov. 11, 2023, 1:16 a.m.