Event-driven model skill assessment

We expect a numerical simulation to improve upon our first guess; typically some sort of climatology. If a model is not better than this initial estimate of the ocean state, the model provides no new information and serves no purpose. Very often in coastal regions we are interested in predicting events—episodes when the ocean state differs greatly from the climatology. The ability of a model to predict these events will define its usefulness, as these are the deviations from climatology that are unknown before the model is run. An idealized time-series of an event is used to investigate the effects various errors have on the prediction of model skill, a commonly used metric of a model’s ability to reproduce observations. It is shown that the choice of climatology is as important in determining model skill as other sources of model error.