Mesoscale Verification Using Meteorological Composites

Abstract

Mesoscale models are often used to explicitly predict discrete, highly structured phenomena. Information regarding the ability of the model to predict events as coherent entities is thus a useful statement of performance. Observational constraints are a significant problem, though, as the shape, size, and intensity of any given event are often only partially known. Composite techniques offer an attractive approach because the full deterministic information about any one event need not be known. If enough quasi-random observations of a distribution of events exist, bulk properties of the distributions of forecasts and observations can be estimated. Composites are also useful in that the verification measures are based on conditional samples of events. Sample distributions contingent on event existence in either the forecasts or the observations can be compared to one another. A verification technique in which meteorological events are located and composited on a relative grid centered on each event is described herein. This technique is described and demonstrated by comparing the 27-km Naval Research Laboratory's Coupled Ocean/Atmosphere Mesoscale Prediction System (COAMPS) mistral wind forecasts to the Special Sensor Microwave Imager (SSM/I) observations for a 1-yr period. Diagnostic information regarding the forecast reliability, error type, and error spatial characteristics are derived. Also, statistics from the conditional distributions of both the observed and predicted events are compared. The difference between the two conditional biases (CBD) is found to reveal valuable information regarding the contribution of false alarms and missed forecasts to the forecast errors. The results indicate the mistral is remarkably predictable with high pattern correlations out to 66 h.