Univ. of Kansas: "Eta model performance for warm season heavy rainfall events across the central United States"

Final Report

This project examined the Eta model forecasts in heavy precipitation cases in Kansas. The work used the 80-km grid spacing, and the major goals were to

• Evaluate the quality of Eta model forecasts of heavy precipitation in the central United States during the summer.
• Determine whether the model can reproduce features previously associated with heavy precipitation. If some features are handled by the model better than others, attempt to determine why and whether the differences can be explained by the structure of the model.
• Establish what grid spacing is needed to reproduce the features associated with heavy precipitation. At the time, the Eta model was run with an 80-km grid spacing but output was provided to NWS forecast offices on a 190-km grid through PCGRIDDS.

For the purpose of this study, heavy precipitation was defined as 2 or more inches of rainfall in 24 hours. Thirteen cases were chosen for the study, and the precipitation fields of the 24-hour forecast were compared to the gridded observed data during the same period. Results indicate that the model had difficulty with the precipitation results. Statistics indicated that there was a 41% probability that the Eta model was generating precipitation forecasts comparable to those produced by chance alone, and that the forecast had little or no skill for these events.

A further examination of a number of other Eta model fields indicated that the model did well in predicting vertical velocity and related parameters, moisture parameters, potential vorticity, and low-level jets. Interestingly, model fields of the divergence of Q-vectors and temperature advection were not generally good indicators of heavy precipitation events. Since other parameters in the Eta model appeared to be fairly good indicators of an upcoming heavy precipitation event, the precipitation parameterizations within the model do not seem to be generating precipitation at a rate corresponding with what one would expect from the other model variables. Virtually all the precipitation in these cases was produced by the cumulus parameterization which could have problems where the atmosphere is convectively (potentially) unstable.

Preliminary analysis of the PCGRIDDS fields indicates that it will represent the features associated with the heavy precipitation events with less intensity and focus.