From doremire at ucar.edu  Fri Mar  2 13:09:50 2007
From: doremire at ucar.edu (Emily Doremire)
Date: Fri, 02 Mar 2007 13:09:50 -0700
Subject: [CCSM-participants] decision on convection for CAM3.5
Message-ID: <45E8848E.4010103@ucar.edu>

At our last AMWG meeting, three candidate modifications of convection
schemes were discussed and evaluated using CAM3 standalone and coupled
simulations.  These modifications were submitted by Guang Zhang at
SIO, Xiaoqing Wu at Iowa State Univ., Rich Neale and Jadwiga Richter
at NCAR.  The Wu submission was a joint work between Xiaoqing Wu and
Guang Zhang.  The Wu and Neale-Richter modification also included
cumulus momentum transport. We wish to congratulate all of them, since
the three configurations all resulted in measurable improvements to the
CAM3 as witnessed by the meeting participants, and all subsequent
analysis that we have performed.

During the meeting, we assessed the model configuration in order to
produce a descendant of CAM3 that could be used by the
Biogeochemistry WG as CAM3.5/CCSM3.5 - an intermediate version before
CAM4/CCSM4 is settled upon in a year or so.  A consensus was reached
that the two configurations with cumulus momentum transport were better,
so that we decided not to consider the "Zhang" configuration without
moment transport further.

The other two submissions by Wu/Zhang (WZ) and Neale/Richter (NR),
have many similaries, and deciding between them has proved
difficult. Mean metrics are very close, and we do not feel the
differences are decisive. For example, the annual space-time weighted
Taylor values for rmse are .903 and .895 for NR and WZ,
respectively. For precipitation, an especially important field, the
annual rmse (against Xie- Arkin) is 1.17 for NR and 1.25 for WZ. The
double ITCZ is less a problem in WZ, and that factor would probably
favor WZ overall, if only mean fields were considered. But other
metrics are also important.

For ENSO variability, the picture is also mixed. Measured against
HadSST monthly anomalies, the amplitude for NR is about 1.6x, while WZ
is about .5x. WZ does poorly on correlations with non-Pacific basins
compared to NR.  The WZ does slightly better for the spatial structure
of the correlation of Nino 3 and SST temperature anomalies just off
the coast of south america, but the NR configuration shows a much more
realistic negative correlation structure in the subtropics and
midlatitude pacific. NR nicely captures strong amplitude peaks at
around both 3 and 5 years, while WZ contains a very weak peak at only
3 years. The NR scheme also shows a more realistic seasonal cycle of
the variability in variance.  Inherently, some subjectivity is
involved in weighing these factors. We felt the presence of the 5-year
peak in NR broke what otherwise might be a tied performance overall,
and rated NR somewhat better on ENSO.

Large differences between NR and WZ are apparent in the amplitude of
diurnal and annual cycles of tropical oceanic precipitation and in
sub-seasonal precipitation variability. NR is much closer to
observations in all of these cases. The WZ configuration shows a
strong diurnal variation over both land and ocean. Over some land
areas (like the ARM SGP site) that variation matches the observations
quite closely, but there are many regions (like the Amazon) where that
signal is less realistic, and the ocean diurnal variation is
unrealistically strong in the WZ configuration everywhere.

NR and WZ both improved the original CAM in many ways and they are
similar by many metrics. If key fields like precipitation are given
high priority. WZ is more realistic with respect to the double
ITCZ. The NR configuration looks better on ENSO, and diurnal variability
over much of the globe.  The large differences in variability at time
scales annual and shorter strongly favor NR.

For these reasons, we (the AMWG co-chairs) have selected (unanimously)
the NR formulation for the interrim CAM3.5 model.

Both formulations, however still have their many deficiencies.  For
WZ, these are the excessive transient variabilities, spurious
precipitation diurnal cycle, and weak ENSO signal.  For NR, these are
large ENSO signal and surface wind errors.  We expect that both
modifications will be further improved to reduce their respective
errors for the configuration of CAM4.

Having made our decision, we note that both of these models are very
close. As was emphasized at the spring AMWG meeting, if clear
improvements over the intermediate benchmark develop, that component can
be swapped into the earth-system development later. In light of the
closeness of WZ and NR, further experimentation is encouraged if the
developers feel it can address some of the remaining deficiencies. WZ
remains a strong alternative, and we encourage its developers to
continue their efforts for future consideration in the
CAM4 configuration.

Note that we have based this evaluation on model performance. The
justifications for the modification are also important.  WZ uses an
empirical closure that has been extensively published in the
literature, albeit in only limited field circumstances and with no
theoretical or conceptual framework. The NR closure has not yet been
published. Conceptually the additional sensitivity introduced within
the underlying Zhang McFarlane formulation to dilution and the latent
heat of fusion are attractive, and we eagerly await publications in
refereed literature.

Obviously, the 1 March deadline required that we come to a decision,
which was very difficult. We thank all involved in this process and
look forward to further collaboration. We are far from being anywhere
near a definitive solution on cumulus parameterization!  We especially
thank Guang Zhang, Xiaoqing Wu, Rich Neale and Jadwiga Richter for
their laudable efforts, and all those who contributed to the
evaluations of the three modifications.  This has been a long journey
and a learning experience for us.

The AMWG co-chairs.