For over five years, the Penn. State/NCAR Mesoscale Model (MM5) has been run at 36, 12, and 4-km grid spacing over the Northwest and verified against a dense regional observational network. More recently, a mesoscale ensemble system has been added to evaluate the potential of probabilistic guidance. This talk will use this multi-year experience to reflect upon some key issues before the mesoscale community, including:

1. What is the relative value of high resolution (deterministic) versus probabilistic (ensemble) guidance at various resolutions and time projections?

2. What are some of the major problems with current parameterization schemes?

3. What is the best approach for running limited area mesoscale models? Are large outer domains always a good idea?

4. The deficiencies of current mesoscale verification schemes and how we might approach the problem.

5. What is the role of post-processing of mesoscale model output to remove systematic errors?

6. The disappointing results of mesoscale initialization.