While some tests were made concerning model physics, the real culprit was found to be an seemingly innocuous procedure in the analysis in which any superadiabatic lapse rate above lever 4 was removed. Seems reasonable, but the turbulence scheme in RUC2 was allowing a slight superadiabatic within the sigma part of the domain up to to near the top of the mixed layer, which is also reasonable. This slight adjustment of temperature over a deep layer was resulting in significant imbalances, particularly when the PBL extended above the top of the sigma domain, as it is doing more frequently now as it gets warmer. The solution was to simply comment out a 4-line do loop.

In our tests, this problem was shown to be at the heart of the late afternoon crashes that were occurring in RUC2 over the last couple of weeks. Fixes were put into the analysis program in RUC2 this morning as of the 16Z run. Besides avoiding the crashes, you should see less noise in height analyses and slightly quieter forecasts overall.

The official implementation of RUC2 is now scheduled for 31 March 1998, assuming that tests with the analysis code change run reliably between now and then and also assuming that 31 March is not a critical weather day.

Stan