The seasonal variation of the west coast is handled with the greatest fidelity. This result cuts across all the models and may be attributable to the fact the SST forcing is specified and common to all the simulations. The common SST forcing is apparently a dominant factor in determining this region's precipitation climatology.
On the space scales of the regions selected, there is little consistent evidence that points to any specific model feature as a predictor of model performance. None of the obvious candidates such as horizontal resolution, convective closure schemes or land surface schemes are reliable discriminators of a model's ability to simulate precipitation.
For one smaller sub-region centered over Arizona, chosen because of the dominance of the semiannual cycle, there is evidence that increased horizontal resolution has an effect. For this intermountain region the higher resolution models as a whole do better than the low resolution models. However, even in this case there is enough variation amongst the individual simulations as to obscure the conclusion that increased horizontal resolution is a necessary or sufficient quality to produce a reliable simulation.
The models tend to have less interannual variation than the observations with more variance being explained by the leading ( annual cycle ) PC, while the observations have a less peaked spectrum.
The models consistently overestimate the precipitation
in the spring and early summer in all regions. This might indicate a common
failing of all the convective schemes in dealing with extratropical convective
instability that is endemic to this time of year.