July 22, 1999
IUGG General Assembly, Birmingham, U.K.
Attending: Raymond Arritt, Jens
Hesselbjerg Christensen, Michael Fox-Rabinovitz, Filippo
Giorgi, Richard Jones, Jack Katzfey, Brent Lofgren, Daniel
Luethi, Bennert Machenhauer, John McGregor, John Roads,
Markku Rummukainen, Huqiang Zhang.
After introductions and a buffet dinner (courtesy of EPRI)
Arritt gave an overview of the agenda including brief plans
for analysis and publications from Experiment 1b. It was
agreed that while consistency of the regional models with
observed large-scale fields is a necessary point, the
analysis should focus mainly on other issues.
Some preliminary comments were made on possible followup
studies. Machehauer pointed out the need to address questions
of predictability, requiring true "climate" runs
(as opposed to specific episodes) or ensembles. Jones
suggested a second integration with different initial
conditions. It was generally agreed that followup studies
such as sensitivity experiments should be done as individual
efforts and not as a part of the PIRCS intercomparison.
Analysis of Experiment 1b
Discussion then turned to analysis of the 1993
case. Most
of the group felt that the analysis should focus on the
hydrologic cycle. Katzfey suggested examination of the
frequency distribution of precipitation, for which we could
get some information from the 3 hourly data. Following from
this suggestion, Roads recommended looking at extremes and
distributions. Fox-Rabinovitz suggested looking at
diagnostics related to the hydrologic cycle such as cloud
fraction, precipitable water and vertically integrated
fluxes. Katzfey pointed out the utility of satellite data
(such as ISCCP) to provide cloud optical depth and cloud
water for comparison to the models. Comparison of surface
fluxes also was recommended. Jones suggested looking at the
question of moisture recycling; i.e., the relative roles of
evaporation versus horizontal moisture flux in producing
precipitation. Roads suggested looking at the diurnal cycle
of precipitation and its association with the low-level jet.
Rummakainen pointed out the importance of critical
combinations of variables and variable extremes. It could be
instructive to see whether regional models can produce these.
Machenhauer noted that surface drying that has often been
found in climate scenarios and suggested this could be
examined.
Proposed PIRCS Experiment 2
The second major topic of discussion centered
around
possibilities for a future experiment or experiments. Roads
presented a proposal for simulating the El Nino episode of
1997-99. Impacts on Florida and California are of particular
interest. The regional models could be embedded within a
prediction instead of analysis. Ideally resolution should be
25 km in order for the models to represent terrain effects in
California, but this would almost certainly require a domain
smaller than the continental U.S. Fox-Rabinovitz asked
whether the models should use a free running land surface
scheme or prescribed soil moisture. Christensen asked what
criteria will be used for validation. The ENSO cycle is
driven by the large scale, but are there additional aspects
to be considered? How do we quantify how accurate regional
models have to be?
Rummukainen asked about the source of driving data. Roads
responded that this could be taken from analyses or from
forecasts, e.g., short term seasonal predictions. Christensen
asked whether an ensemble approach might be useful, while
Fox-Rabinovitz asked what is meant by prediction mode. Roads
responded that there are many ways of driving the regional
models. We can do "pure" prediction as well as pure
analysis, or any range between. Examples include analyses
with observed SST, or predicted SST, etc. We could use
prescribed SST and NCEP analysis, then compare with a
prediction having some skill (e.g., IRI). McGregor asked
whether an AMIP style approach could be useful. Giorgi pointed out the need to clarify the objective of
the experiment and that Roads' proposal would fit an
objective of seasonal prediction. We need to consider what is
a natural followon from Experiment 1. The group could do a
multiyear run, and look at other mesoscale effects (e.g.,
topography, western U.S., winter) in more of a natural
extension than seasonal prediction. The possibility of an
enhanced CO2 scenario was briefly discussed but it
was decided that meaningful runs would have to be too long to
be practical.
The topic of verification data for the 1993 experiment was
briefly revisited. Fox-Rabinovitz mentioned that his group
has run the 1993 experiment using their stretched grid model
in data assimilation mode with 60 km resolution over the U.S.
These results include comprehensive diagnostics and are
available for use as verification. Roads agreed that this
could be useful for the second paper and asked whether it is
it better than the NCEP reanalysis. Fox-Rabinovitz replied
that the stretched grid analyses are a bit better than the
standard (coarser resolution) GEOS (not NCEP) analyses,
according to his evaluation for summers of 1988 and 1993.
Giorgi suggested comparing the 1988 and 1993 model-predicted
precipitation to reanalysis precipitation. There was brief
discussion of the ways in which reanalysis precipitation is
affected by the "spinup" problem.
Roads said we need to ask what are people willing to do.
Roads suggested reconsidering what data should be archived as
the data volume can be a burden on some participants; for
example, whether it might be sufficient to save monthly
averages. Fox-Rabinovitz felt there is a clear need to save 6
hourly values in order to resolve the diurnal cycle and
mesoscale features. Arritt pointed out that data volume could
be reduced by saving only certain levels and pre-computing
some of the required quantities such as heights of standard
pressure surfaces. Katzfey recommended we save and examine
cloud fields (fraction, optical depth).
Stretched grid modeling
Fox-Rabinovitz presented an update on stretched grid modeling. His group is working on developing the stretched grid version of the new NASA/NCAR model. This version will be used for regional climate studies. Several groups are doing stretched grid modeling with about 3-4 groups aggressively involved. This could be a new component for PIRCS. Fox-Rabinovitz is running the model both in "simulation mode" and "assimilation mode." For the PIRCS experiment, they are running in assimilation mode with data withheld over the U.S. He will submit the 1993 results for the PIRCS comparison.
NCAR Community Regional Climate Model
In the absence of Ruby Leung, Arritt briefly discussed plans for a possible regional modeling option within the NCAR MM5, presented at a meeting on 24 June 1999 at NCAR. The modified OSU land-surface scheme (similar to the Eta model) is being added to make MM5 more suitable for extended runs. Giorgi mentioned that the relatively simple standard MM5 radiation scheme could be a problem as well as the land surface.
Geographical regions for Experiment 2
The group returned to recommendations for PIRCS
Experiment
2, focusing on the region or regions to be simulated. This
was motivated by comments by Christensen that while the
original PIRCS goals addressed aspects of interest to all,
not just U.S. participants, the project seems to have
progressed to a U.S. focus. This could discourage
participation from groups outside of the U.S. Fox-Rabinovitz
pointed out that modelers once were preoccupied with Indian
monsoon and that now ENSO over the U.S. is considered an
interesting event.
Christensen pointed out that the program needs overall
scientific objectives and must also be of benefit to
individual groups; do we learn anything new? We need to
revisit what is the fundamental reason for doing
intercomparisons and stress the new science. Giorgi felt that
a natural next step is to do multi-year perfect boundary
condition experiments and for this reason likes the 1997-99
ENSO idea.
Arritt brought up the possibility of doing the ENSO for
two regions. North America could be run as an extension of
Experiment 1 with a second region being simulated for
contrast. Giorgi asked what this region might be, and the
possibility of Brazil was mentioned. Other regions were also
mentioned. It was pointed out that there already lots of
people doing regional climate simulations for East Asia.
Africa was brought up as a possibility. After discussion, a
consensus emerged that South America would probably be the
most appropriate region to simulate in addition to (or
instead of) North America. This is partly because there are
some similarities to North America (e.g., the low-level jet
as a frequent mesoscale feature) and data from other studies
could be used for validation. Christensen reiterated the need
to keep in mind the ultimate goal of the project and that it
is hard to use separate simulations as an ensemble. We could
approach the two regions as 2a and 2b. Giorgi pointed out the
need to make contacts with other investigators and
institutions; e.g, Carlos Nobre, PACS, IAI. Roads mentioned
funding and suggested we collaborate more closely with AMIP.
Arritt volunteered to summarize the discussions into a
plan for PIRCS Experiment 2, with feedback from the rest of
the Iowa State group and from the PIRCS community at large.
Next PIRCS meeting
The topic of the next PIRCS meeting was brought up. The most likely venues are the AGU Fall Meeting in San Francisco (December 1999) or the AMS annual meeting in Long Beach (January 2000). Some felt we should meet at both places.
Copyright/Trademark Legal Notice |