Return to search

Evaluation of a probabilistic quantitative precipitation forecasting experiment

Forecasts of the likelihood of occurrence of various
amounts of precipitation are very important, since
excessive precipitation amounts over relatively short time
periods can have adverse effects on public safety and
economic efficiency, As a result, forecasters at the
National Weather Service Forecast Office in San Antonio,
Texas were asked to formulate subjective probabilistic
quantitative precipitation forecasts on an experimental
basis beginning in February 1981. This study describes
methods of evaluating probability forecasts of this
ordinal variable and presents some results of the first
year of the experiment.
Scalar and vector evaluation procedures are
described. In the case of scalar evaluation, the
inclusion of a no-skill line and a no-correlation line on
reliability diagrams is helpful in representing the skill,
reliability, and resolution qeometrically in two-state
situations. Geometrical interpretations of attributes of
forecasts can also be accomplished in three-state
situations based on vector evaluation procedures. A skill
score for subsample forecasts is shown to be useful in
identifying systematic errors made by forecasters or
forecast systems. A beta model is developed to obtain a
forecaster's predictive distributions (i.e., the
distribution of use of probability values).
The experimental results show that the skill of the
subjective forecasts is generally higher than the skill of
objective guidance forecasts for measurable precipitation
(i.e., precipitation amounts exceeding a threshold of 0.01
inches), but that the opposite is true for threshold
associated with larger precipitation amounts. This result
is due primarily to the forecaster's tendency to
over forecast for the events associated with higher
precipitation thresholds. The tendency to over forecast is
most pronounced in the nighttime forecasts and in the
forecasts for drier stations. The MCS objective guidance
forecasts, on the other hand, are quite reliable for both
periods and all stations. The vector evaluation approach
indicates that the degree of overforecasting is quite high
for bimodal forecasts and that the skill contribution from
bimodal forecasts is negative in many cases. / Graduation date: 1983
Date24 June 1982
CreatorsHsu, Wu-ron
ContributorsMurphy, Allan H.
Source SetsOregon State University
Detected LanguageEnglish

Page generated in 0.0015 seconds