WeatherProbs: Three-category weather forecasts
In scoring: Proper Scoring Rules

Description Usage Format Details Source References Examples

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

1	data("WeatherProbs")

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)

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.

Data were obtained from http://www.cpc.ncep.noaa.gov/products/archives/short_range/ (see URL in 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.

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")