Methods to evaluate probabilistic and ensemble forecasts
Yuejian Zhu and Zoltan Toth
SAIC at EMC/NCEP
Abstract:
The introduction of probabilistic weather, water, and climate forecasts
is one of the strategic goals of the NWS. Ensemble forecasting offers a
practical tool to generate objective guidance for the issuance of such forecasts.
The verification of probabilistic and ensemble forecasts requires methods
different from those used to evaluate single value or categorical forecasts.
This presentation will review the main attributes of probabilistic forecasts,
resolution and reliability. Different verification scores, such as
the Brier Skill Score, Ranked Probability Skill Score, the Relative Operating
Charactersitics, etc. will be introduced and evaluated as to how they measure
the main attributes of forecasts. In addition, the relationship between
verification and statistical postprocessing of probabilistic forecasts
will also be discussed.