Sheraton Palace Hotel, San Francisco, California
11 December 1997

Chuck Hakkarinen proposed that we hold a follow-on meeting similar this one, perhaps in conjunction with the St. Louis GEWEX meeting this summer. John Roads suggested that the 1998 Spring AGU Meeting would be more appropriate.

Overviews of model results for Experiment 1a (15 May-15 July 1988)

Gutowski presented results from the ISU runs for Experiment 1a using RegCM2 in comparison to the NCEP reanalysis results. The mean 500 mb heights were generally well reproduced with a slight southward shift through most of the domain. The 5880 m contour does not extend into the southern part of the domain for the simulated results as it does in the NCEP reanalysis. There is too much rainfall over the mountainous western U.S. The region of low precipitation over the Midwest has values higher than observed and does not extend far enough west. Precipitable water tends to be too low in the intermountain west and the results do not capture the intrusion of moist air from the Gulf into the Dakotas-Minnesota region that appears in the reanalysis. Roads pointed out that the reanalysis fields are not truly "observations" and may not depict the precipitable water accurately. The NCEP Eta reanalysis may provide a better representation. Hakkarinen asked whether the spatial and temporal resolution of the model analysis should match that of the observations. The model does not capture the nocturnal fraction of precipitation over the central U.S.; it should also be noted that none of the various reanalyses capture this. Characteristics of the low-level jet (LLJ) over the central U.S. are in general agreement between the model and the reanalysis.

John Roads discussed results from the Regional Spectral Model (RSM). The RSM reproduces the general features of the 500 mb height although is slightly more noisy than the reanalysis. This model gives some large values of precipitation over the mountainous West, as well as some probably spurious local maxima near the boundaries. Precipitable water broadly corresponds to the reanalysis but does not have the intrusion in the western Great Plains. There is some tendency for a nocturnal precipitation maximum over the central U.S. although the pattern is noisy. It was pointed out that one of the emails to the PIRCS group contained an erroneous specification for the time period to be used in computing the nocturnal precipitation fraction (the correct period being 00 UTC to 12 UTC). The nocturnal LLJ is stronger than in the reanalysis. The LLJ is more intense in the 1993 RSM results than in the 1988 RSM results.

Jens Hesselbjerg Christensen found that the predicted 500 mb height pattern using HIRHAM is slightly more zonal than the reanalysis. The precipitation is high over the mountainous West and has a minimum in Texas that does not appear in the observations. As with the other models there is some excessive precipitation near the boundaries. The precipitable water is somewhat too low over the West but there is some indication of the observed moist intrusion into the Great Plains. Consistent with the other models shown here, the nocturnal precipitation fraction looks noisy. Jens briefly discussed some results with different soil moistures and found that surface temperature was sensitive to relatively small changes of soil moisture. This indicates that model simulations of surface temperature should be carefully evaluated, both for their intrinsic interest and as secondary indications of the realism of the soil moisture. The predicted LLJ is broadly similar to the reanalysis.

Song-You Hong showed results from another implementation of the RSM. The mean 500 mb height agrees very closely with the reanalysis, probably in part because some of the long-wavelength components are included in the model interior. The precipitation minimum over the Midwest again is not captured and there is too much precipitation in the northern U.S. The reanalysis also is in general agreement and there is a broad indication of the moist intrusion into the central U.S., though it is farther to the east than in the reanalysis. As in the other models shown at this meeting the nocturnal precipitation fraction is noisy. The magnitude of the LLJ seems realistic.

Glen Liston showed results from a preliminary RAMS run. The amplitude of the 500 mb ridge is very slightly lower (15 m) than in the reanalysis. This model reproduces the broad features of the observed precipitation including an indication of a local precipitation minimum in the Great Lakes region. The predicted precipitable water has a strong gradient east of the foothills of the Rocky Mountains and includes a moist intrusion in the central U.S. that is somewhat farther east than in the reanalysis. The LLJ is similar to that in the reanalysis.

John McGregor discussed results using DARLAM. He found close agreement with the reanalysis 500 mb height although with a center of higher 500 mb heights over the south- central U.S. McGregor noted that a region of higher 500 mb heights could also be seen in the corresponding ECMWF analyses. The predicted precipitation tends to be too low with too little precipitation over the south-central U.S. and some indication of a minimum over the Great Lakes. As was also found with other models there was some high precipitation over the mountainous West, and the question arose as to the adequacy of precipitation measurements over this region. Precipitable water patterns resembled the reanalysis with a moist tongue over the Great Plains in nearly the same location as the reanalysis. The nocturnal precipitation fraction is noisy. The LLJ is stronger than in the reanalysis (about 10 m/s versus 7 m/s). McGregor showed some additional fields including time series of domain average MSLP.

Daniel Caya briefly showed some results from the Canadian RCM simulations. The 500 mb height is slightly lower amplitude than the reanalysis. These preliminary results show too much precipitation because of a problem that seems to be present in their convection scheme.

Discussion

There was extensive discussion of the overall features of the model results as well as the adequacy of observational data for comparison to the model results. Wilby asked about the general tendency of the models shown here to overpredict the precipitation. Christensen mentioned that the 1988 simulation is likely to be very sensitive to initial conditions, especially for soil moisture. He suggested that the reanalysis soil moisture is probably too high. Roads pointed out that all of the PIRCS models are starting from the same soil moisture which is highly uncertain. Small noted that in RegCM2 there is no re- evaporation of precipitation in the semi-explicit scheme. It was also pointed out that the Experiment 1a period was characterized by generally low observed precipitation over the central U.S., so that a relatively small absolute error resulting from failure to capture one or a few events (or model prediction of a small amount of spurious precipitation) can represent a large fractional error that may not be meaningful.

Laprise asked if anyone has evaluated whether the regional domain is tracking the large-scale dynamics. Laprise and Caya presented examples of large discrepancies from observations in the day-by-day evolution of the model atmosphere, especially with large domain and weak flow. Liston suggested that we report some time series of domain-averaged (or subdomain-averaged) quantities to detect episodes when the models depart from the observations. A suggestion was made that precipitation should be stratified according to various criteria, such as by strong or weak forcing, or RMS error in upper-level heights. A long discussion ensued with Hakkarinen suggesting that we evaluate what variables the models can and cannot predict well. Laprise suggested we concentrate on moisture as a "value added" field. Should we evaluate the regional models against the reanalysis, or against the GCMs?

It was pointed out that 60-day runs are not really "climate" and that in particular the results can be sensitive to whether a few specific events are properly reproduced. There was some discussion of doing a long term (e.g., 5-year) run as an eventual follow-on experiment in order to reach toward more climate-like time scales.

The models all have problems with simulating the observed secondary maximum of precipitation over Texas. There was discussion as to whether this is this due to one or a few events, or part of the seasonal climatology. Careful post-analyses of model results are necessary to get a better idea of how the models are behaving. Sensitivity to initial conditions, especially for soil moisture, has to be evaluated. Christensen asked whether the group should run the models for an ensemble or suite of initial conditions. Caya suggested we evaluate inter- and intra-model sensitivity to initial soil moisture. There was a consensus that we should first evaluate the basic simulations but that individual groups are encouraged to perform these types of sensitivity runs.

In this vein Liston pointed out that we need concrete ideas on where to go next with the present simulations; people can do additional sensitivity studies on their own. Are we saving the right data sets? What about observational data sets? Should we define standards for evaluating models? Roads responded that we do not need to define specific standards as to what is an acceptable simulation, but rather that it is more useful to evaluate the differences between models. Liston also noted that we need to decide on what data to use for evaluating the model results. Mearns recommended that we consider using several observational data sets. Wilby pointed out that statistical downscalers have put together some thorough analysis of observed data for regions of the U.S. that could be used as a template. This includes NCDC data that have been subjected to detailed quality control. Although the data are not extensive they are of high quality. We could also look at various statistical diagnostics such as persistence of dry periods. Roads suggested that it is critical to get such data into the PIRCS archive.

Data format for archive

Gutowski showed a "straw man" proposal for the archive data format. McGregor asked whether cloud cover should be archived to cooperate with GEWEX. Perhaps it should be an optional field. Small asked whether this could be included in terms of the short-wave and long-wave cloud forcing. A suggestion was made to check with Dave Randall to see what variables would be useful. Modelers will use IEEE754 floating-point format for the fields that are transmitted to the archive, with the understanding that they may need to re- transmit the data if there are incompatibilities.

Extended discussion ensued as to the vertical coordinate on which the data should be supplied, with some suggesting pressure coordinates and others saying the data should be provided on the model's vertical levels to minimize smoothing. The consensus evolved that if the vertical coordinate is relatively straightforward (such as a sigma-pressure coordinate) the model levels should be used, but for models having more complicated vertical coordinates the data should be interpreted to pressure levels. Variables may be horizontally interpolated to a common grid when staggered grids are used. The PIRCS group at ISU will provide a sample of how the data are to be organized.

Discussion followed as to the preparation of publications describing the PIRCS results. The group agreed to provide data by 1 March with a goal of having an early draft of results prepared by the time of the 1998 AGU Spring Meeting in late May.