Patterns of forecast error and analysis uncertainty in medium-range forecast skill dropouts

A continuing problem in numerical weather forecasting is the occurrence of intermittent but significant “dropouts” in forecast skill. These forecast dropouts occur in all operational models to varying extents and can be seen, for example, in time series of 5-day anomaly correlation (AC) scores for 500hPa height error. In such cases, the AC on a particular day (or sequence of several days) may drop by 0.1 or more below the monthly average AC. Forecast skill dropouts are likely caused by a combination of initial condition error (in some cases related to observation data inadequacy) and error created by deficiencies in forecast model dynamics and physics. In this study we examine forecast skill dropouts using a multi-year (2004-2010) database of analyses and forecasts produced by ECMWF, GFS and NOGAPS. The primary research objectives of this study are to: a) identify the geographic distribution of forecast dropouts over the six-year study period, b) describe the types of flow regimes that are frequently associated with forecast dropouts, and c) correlate forecast dropouts with patterns of analysis uncertainty (defined here as differences between analyses produced by the various forecast centers). This information will provide a framework in which to develop solutions to actually mitigate forecast dropouts in operational systems. Specifically the study is intended to identify specific regions that may require observational enhancement or improved procedures to assimilate existing observations. It is hypothesized that medium-range forecast dropouts in mid-latitudes occur preferentially with large-scale flow-regime changes. The forecast dropouts may thus represent an interval of low skill during which the forecast/assimilation system is not adjusting rapidly enough to the flow-regime transition. In this presentation we will summarize results from the first phase of this project.