scoring (version 0.6)

WeatherProbs: Three-category weather forecasts

Description

Probabilistic forecasts from the U.S. National Oceanic and Atmospheric Administration, concerning below/near/above average temperatures and below/near/above median precipitation.

Usage

data("WeatherProbs")

Format

A data frame with 8976 observations on the following 11 variables.

stn

Station World Meteorological Organization (WMO) number

made

Forecast issuance date

valid

Center of forecast valid period

tblw

Probability of below normal temperatures

tnrm

Probability of near normal temperatures

tabv

Probability of above normal temperatures

tcat

Realized temperature category (1=below, 2=near, 3=above)

pblw

Probability of below median precipitation

pnrm

Probability of near median precipitation

pabv

Probability of above median precipitation

pcat

Realized precipitation category (1=below, 2=near, 3=above)

Details

The forecasts are valid for a period of 6 to 10 days from the date that the forecast was made. The forecasts were supplied every weekday during April, 2009, and they specifically predict the average temperature or total precipitation for the entire valid period.

References

See http://www.cpc.ncep.noaa.gov/products/archives/short_range/README.6-10day.txt for more details on the data.

For an application of similar data (different dates, same source), see:

Wilks, D. S. (in press). The calibration simplex: A generalization of the reliability diagram for 3-category probability forecasts. Weather and Forecasting.

Examples

# NOT RUN {
data("WeatherProbs")

## Brier score for temperature forecasts
## (Warning arises because some forecast rows don't sum to 1.)
res <- calcscore(tcat ~ tblw + tnrm + tabv, data=WeatherProbs,
                 bounds=c(0,1))

## Ordered Brier score for temperature forecasts
res2 <- calcscore(tcat ~ tblw + tnrm + tabv, data=WeatherProbs,
                  bounds=c(0,1), ordered=TRUE)

## Spherical score for temperature forecasts
res3 <- calcscore(tcat ~ tblw + tnrm + tabv, data=WeatherProbs,
                  fam="sph", bounds=c(0,1))

## Average scores by station
avgbrier <- with(WeatherProbs, tapply(res, stn, mean))
avgobrier <- with(WeatherProbs, tapply(res2, stn, mean))
avgsph <- with(WeatherProbs, tapply(res3, stn, mean))

## Conclusions vary across Brier and ordinal Brier scores
plot(avgbrier, avgobrier, pch=20, xlab="Brier", ylab="Ordinal Brier")
# }