A method is developed whereby meteorological events with similar characteristics are identified using a rules-based algorithm. The events may be identified using criteria from either observed or prognosed fields, depending on the coverage and reliability of the observations. Inadequate measurements often preclude the accurate definition of events in the observed fields. However, the full coverage of the model forecast fields allows for estimates of the conditional forecast-observational matches given that the event was predicted.

When a point meets the definition of an event, all other adjacent points meeting the definition are collected as a contiguous object. Meteorological information within a specified area surrounding and including the object is then translated to a relative grid whose center is defined by the center of the object in the model forecast field. If the observations are incomplete, the forecasts are templated based on the valid observations. Event occurrences are also recorded on the native geographical grids to investigate their distribution in space and time.

When a number of events are composited, attributes of the forecast and observed occurrence distributions can be compared with one another. For instance, if a consistent spatial error exists, the centers of maximum occurrence will be displaced. Random errors show up as a broadening of one distribution with respect to the other. The climatology of the events in both the model and the observations can be estimated using the results on the geographical grids. For the case of incomplete observations, comparing the full distribution of the forecasts with the templated distribution provides valuable information regarding sampling error.

This method is demonstrated by comparing SSM/I observations to model forecasts of oceanic wind events. The results indicate that some events, like the Mistral, are highly predictable. In the case of the mistral, strong correlations exist between the distributions out to 54 hours.