Model initialization errors create uncertainty in numerical weather prediction forecasts. Forecasters must be capable of rapidly identifying weather systems, which may not be well forecast by the model, mainly due to data initialization processes. One means to assess the model's ability to forecast a feature is to use a lagged averaged forecast approach (LAF). The LAF approach provides a means to visualize trends and difference in model runs valid at the same time. Areas of significant differences should be suspect and may provide insights into systems which are not properly initialized by the model.

LAF techniques were applied to two cases involving winter storms. The first storm was a major East Coast storm of 25 January 2000, which impacted the major cities along the eastern seaboard. The upper-level short wave associated with this event was poorly initialized in both the operational NCEP Eta and AVN models. Distinct run-to-run differences could be seen in successive forecasts. Additionally, similar errors were seen in both models, suggesting an initialization error. Successive model forecasts showed a trend toward a deeper and more westerly storm track. Additionally, the LAF and dispersion showed distinct patterns suggesting an error in the track and intensity of the surface cyclone. Similar errors were found at 500 hPa.

The second case involved the 30 December 2000 snowstorm. This was a non-event for areas south and west of the main cyclone center. In this case, the coarse sea-surface temperature data in the operational NCEP Eta model analyzed the baroclinic zone too close to the coast. This contributed to a deeper cyclone, which tracked too far west, resulting in excessive inland precipitation. The LAF products showed large run-to-run inconsistencies in the Eta suggesting a low confidence forecast. The AVN LAF products showed more consistent trends and smaller run-to-run differences.

This paper will show how using LAF techniques, forecasters can identify potential problems in model forecasts due to data initialization errors.