compute_scores: Compute forecast score

In jonasbhend/VOLCprediction: analyse influence of volcanic eruptions on decadal predictability

Description

This function computes the residual variance in the forecast interval. This is the variance unexplained by the prediction in case the forecasting system has exactly zero skill. In case the forecasting system has skill, only a fraction of the residual variance is computed. Bounding assumptions on how skill plays out with an eruption in the forecast interval are computed. These bounding assumptions are: i) skill is not influenced by the eruption (optimistic), ii) all skill is lost after an eruption (pessimistic).

Usage

compute_score(x, n = 10, erup.i = NULL, skill = NULL)

comp_score(x, erup.i = NULL, n = 10, clim = 10, skill = 0)

Arguments

x	data vector, time series
n	length of prediction interval (defaults to 10 years)
erup.i	indices in x with eruptions. This is the time the volcano erupts. The assumption is that the eruption will not affect this timestep, but will start affecting from the following timestep (make sure to use the correct index for monthly or seasonal data).
skill	vector of skills of forecast (see details)
clim	number of years leading the forecast used as baseline

Details

The old function compute_score used a skill definition in fraction of standard deviation. Otherwise the two functions are identical.

Value

array of forecast scores

Examples

## set up a time series of pseudo-observations and plot
nn <- 200
## dates of virtual eruptions
erup.i <- sort(ceiling(runif(10, min=0, max=nn)))
## indices of effect of volcanoes
iafter <- outer(erup.i, 1:3, '+')
iafter <- iafter[iafter <= length(xx)]
xx <- rnorm(nn)
xx[iafter] <- rnorm(length(iafter), sd=2)
plot(xx, type='l', xlab='time', ylab='temperature')
abline(v=erup.i, lty=3)

## compute the scores
fscore <- comp_score(xx, n=10, clim=1, erup.i=erup.i, skill=seq(0.05,0.95,0.1))

jonasbhend/VOLCprediction documentation built on May 19, 2019, 7:27 p.m.