At ECMWF several types
of statistical verification scores are computed for a number of
areas and parameters, and stored in a historical data base. Most of
them have been defined in agreement with other NWP centres and are
regularly exchanged following WMO/CBS recommendations.
The mathematics of the
must common verifications are quite simple, while the
interpretation of them is a constant source of confusion and
debate. The reason is not only that different verification methods
often give conflicting indications, sometimes it is not obvious how
to interpret the results from just one verification system: -What
looks good might be bad, what looks bad might be good
Firstly, it is
important to distinguish between the different types of forecasts
we want to verify: Direct Model Output (DMO) from NWP,
Post-Processed Products (PPP) and End-Products (EP) delivered to
the public or paying customers. Here we will mainly deal with the
first category although many of the conclusions are also valid for
PPP and EP. Secondly, for all three categories it is important to
distinguish between statistics which evaluate the accuracy, the
skill and the utility of the forecasts.
The most common
verifications measure the accuracy of the forecasts in relation to
the analyses or observations. While the accuracy is absolute, the
skill measure is relative, comparing with some reference like
persistence, climate or some alternative forecast system. Finally,
the utility aspect determines the success of a forecast system in
terms of the monetary or political use that can be made from its
predictions.
|
