ar_preddistr: Predictive Moments from an AR Ensemble

In JuGross/ensAR: Autoregressive postprocessing methods for ensemble forecasts

Description

Computes the predictive mean and predictive standard deviation based on a model designed to handle an AR modified ensemble by ar_ensemble.

Usage

ar_preddistr(ar_ens, train = 30)

Arguments

ar_ens	A list generated by ar_ensemble.
train	The length of the training period.

Details

The predictive mean mu is the usual mean of the AR modified ensemble members (data frame forecast from ar_ens). The predictive standard deviation sd is the weighted mean of two standard deviations. The first one is the square root of the mean of AR variances (data frame variance from ar_ens). The second one is the sample standard deviation (with denominator n, not n-1) of the AR modified ensemble members. The weights are chosen in order to minimize the average CRPS computed from a rolling training period of length train and assuming a predictive Gaussian distribution.

Value

A data frame containing

• obs: the observation

• mu: predictive mean

• sd: predictive standard deviation

• w: the weight corresponding to the first employed standard deviation

and additional columns as given by additional when invoking ar_ensemble.

Author(s)

J. Gross, A. Moeller.

Examples

mod <- ar_ensemble(ens = Magdeburg[1:(90 + 30 + 1), -c(57,58)],
    obs_col = 6, mem_col = 7:56)
ar_preddistr(mod) # data frame of one row

JuGross/ensAR documentation built on May 10, 2019, 8:23 a.m.