Ranked Probability Score
Calculates the ranked probability score (rps) and ranked probability skill score (rpss) for probabilistic forecasts of ordered events.
rps(obs, pred, baseline=NULL)
obs |
A vector of observed outcomes. These values correspond to columns of prediction probabilities. |
pred |
A matrix of probabilities for each outcome occurring. Each column represents a category of prediction. |
baseline |
If NULL (default) the probability based on the sample data of each event to occur. Alternatively, a vector the same length of the as the number categories can be entered. |
rps |
Ranked probability scores |
rpss |
Ranked probability skill score. Uses baseline or sample climatology as a references score. |
rps.clim |
Ranked probability score for baseline forecast. |
Perhaps the format of the data is best understood in the context of an example. Consider a probability of precipitation forecast of "none", "light" or "heavy". This could be [0.5, 0.3, 0.2]. If heavy rain occurred, the observed value would be 3, indicating event summarized in the third column occurred.
The RPS value is scaled to a [0,1 ] interval by dividing by (number of categories -1 . There is a discrepancy in the way this is explained in Wilks (2005) and the WWRF web page.
Matt Pocernich
WWRP/WGNE Joint Working Group on Verification - Forecast Verification - Issues, Methods and FAQ http://www.cawcr.gov.au/projects/verification/verif_web_page.html#RPS
Wilks, D. S. (2005) Statistical Methods in the Atmospheric Sciences Chapter 7, San Diego: Academic Press.
### Example from Wilks, note without a baseline and only one ### forecast, the rpss and ss are not too meaningfull. rps( obs = c(1), pred = matrix(c(0.2, 0.5, 0.3), nrow = 1))
Please choose more modern alternatives, such as Google Chrome or Mozilla Firefox.