Gutowski compared results for total precipitation (convective and stable) from seven of the PIRCS models to observed precipitation from the Higgins data set. The models generally are able to replicate wet and dry spells in the observations, but the amounts may be underpredicted or overpredicted. A simple "consensus" (arithmetic mean of the participating models) tracks the observed precipitation rather well. Liston brought up the point that the Higgins data are on a relatively coarse resolution and that we should generate a higher-resolution analysis for comparison to the models. The Iowa State group noted that they had already begun to develop such analyses. The Higgins data were used because of the short lead time for this meeting (models results having been provided as little as two days before).

Evaluation of model performance for different types of events can give a perspective on how the models handle different types of forcing (e.g., large-scale versus convective or mesoscale). Differences between convective and stratiform precipitation in the models can also give information on how the models respond to different types of forcing. To this end, Takle described the synoptic characteristics of specific precipitation episodes during PIRCS 1a. The early part of the period was dominated by large-scale synoptic events; all models captured the timing of these. A dry period around Julian day 159 also was represented by all models as was a convective event around day 160. Days 165-180 had small convective events popping up across the upper Mississippi Basin region, with all models capturing something, though timing may not always be precise. Pielke noted that the models differ on whether precipitation is represented as convective or stratiform. All models give convective precipitation dominating on days 165-180, though Pielke mentioned that all miss the dry period at day 187. A handout was distributed with some of these results.

The two versions of RSM differ in the relative representation of convective and stratiform precipitation for several episodes. The Scripps version of the RSM includes the physics from the operational, global spectral model, while the NCEP version has modifications to the physics introduced by Hong. Liston commented that all the models apparently capture the lateral boundary forcing as indicated by the timing of specific precipitation episodes. Hakkarinen pointed out that the models perform somewhat more poorly toward the end of the period, and several in the audience mentioned that this may point to the implications of soil moisture feedbacks.

Gutowski showed maps of cumulative precipitation from four models. All have a dry area in the lower Mississippi River basin that does not appear in the observations. The nocturnal fraction in most of the models exhibits strong spatial gradients that may not be physically realistic, and the models underrepresent the nocturnal maximum in the central U.S. Hong pointed out that convection schemes generally tend to initiate precipitation during the daytime. RAMS seems to give the most realistic nocturnal maximum in the central U.S., though its precipitation tends to be too low. Wilby asked whether the spatial patterns should be looked at separately for convective or stratiform episodes. Liston mentioned that to evaluate the daytime versus nocturnal precipitation there are several factors that must be accounted for; e.g., there must a significant amount of precipitation and not just a few isolated events. Hakkarinen mentioned that the RAMS precipitation appears to most strongly aligned with the underlying terrain and coastlines.

2.2 Low-level jets

Arritt presented analyses of the low-level jet in several of the models based on the criteria of Bonner (1968, Mon. Wea. Rev.). In 1988 the profilers had not yet been deployed so there is not a good source of observed data. Therefore the NCEP reanalysis is being used for preliminary evaluations. Comparison with profiler observations shows that the reanalysis itself has biases, including too many LLJs (Bonner's criterion 1) in the southern Great Plains and too few in the northern Great Plains. Incidence of strong LLJs (Bonner's criterion 3) is severely underrepresented in the reanalysis, perhaps because the reanalysis terrain is too coarse. It is noteworthy that mesoscale models tend to produce more LLJs than the reanalysis. For 1988 the region of high LLJ frequency tends to be to north of the climatological distribution, as expected in a dry year.

MM5 gives a pattern of LLJ frequency that is similar to the reanalysis, but with generally higher incidence of LLJs. DARLAM does not extend quite as far north in maximum LLJ frequency and tends to get more strong LLJs. RAMS has a relatively high incidence of LLJs and a tendency for spatial patterns of LLJ occurrence to follow the Bermuda high. Several models show a Y-shaped or "branched" distribution with swaths of high LLJ incidence to the west along the foothills of the Rocky Mountains and to the east along the periphery of the Bermuda high. RAMS output was provided on pressure levels and results may be different if data were used on the native levels of the model. The NCEP version of the RSM gives a somewhat lower frequency of LLJs than the other models, extending rather far north (again, as might be expected in a dry year).

2.3 Additional diagnostics

The group as a whole will begin working on papers that report PIRCS. The first will be a general description of the results and procedures, including some examples of analysis products but not in-depth analysis. Shortly following would be a "science paper" with more detailed analysis focusing perhaps on how precipitation is coupled (or not) with mesoscale dynamics. There is a need to show what is the added value of regional models. For example, GCMs can produce low-level jets but still do not properly represent the diurnal cycle of precipitation. Roads mentioned that little attention has previously been given to diurnal variations in the reanalysis.

Liston questioned whether we are ready to release the data sets to others and that we should take time to make sure that the models are at a credible stage. Takle mentioned WGNE and CLIVAR are looking to PIRCS to show added value from regional models. Gutowski noted that PIRCS has ongoing cooperation with MERCURE and that we should proceed in a timely manner in order to share information with them. Hakkarinen mentioned the push for regional impacts for the IPCC TAR and the U.S. national assessment of the state of science of impacts of climate change. New GCM runs also will be done with intent of serving as drivers for regional models. Roads says we need to push for the first paper but for the second we need more of a critical mass of participants, encouraging others to collaborate with us. Roads mentioned Fall AGU as a target date for another meeting.