Lawrence Wilson
Originally intended as a way of forecasting the accuracy of deterministic
forecasts, ensembles are increasingly used to forecast the probability of
occurrence of meteorological events of interest, especially high impact
weather events. One way of assessing probability forecasts is to use the
relative operating characteristic curve (ROC) to assess the ability of the
forecast system to discriminate situations preceding an event of interest
from those preceding the non-occurrence of the event. The ROC has become
very popular as a tool to verify ensemble forecasts. Related to the ROC
is the "potential value" curve developed by Richardson (2000) to evaluate
the value of probability forecasts to users who are able to define their
cost-loss matrix.
The presentation will include a discussion of some of the theory behind
both the ROC and the cost-loss matrix, and will be illustrated with
verification results from our own and other ensemble forecast systems.