Physical Initialization for the Regional Spectral Model

1
Physical Initialization for the Regional
Spectral Model
Ana Nunes, John Roads and Masao Kanamitsu Scripps
Experimental Climate Prediction Center (ECPC)La
Jolla, California, USA anunes_at_ucsd.edu
2
PHYSICAL INITIALIZATION FOR THE REGIONAL SPECTRAL
MODEL
Although atmospheric analyses and reanalyses are
now providing physical realistic fields for many
variables, precipitation remains problematic.
Physical initialization (PI) has been proposed as
a methodology for improving precipitation and
related hydroclimatological simulation skill. For
this reason, the Scripps Experimental Climate
Prediction Center (ECPC) is now implementing a PI
procedure in the Regional Spectral Model (RSM).
We summarize here some of the improvements
obtained for climate simulations over two
distinct domains (1) U. S. and Mexico during a
Southwest-Mexico monsoon season and (2) South
America during the rainfall season of the Amazon
region and increased activity of the South
Atlantic Convergence Zone.
3
PHYSICAL INITIALIZATION FOR THE REGIONAL SPECTRAL
MODEL
The Scripps ECPC RSM, described previously by
Juang and Kanamitsu (1994) Anderson et al.
(2001) and Roads (2003), used for these
experiments had 60-km resolution and 28 levels in
the vertical. A Mercator projection was used for
the projection of the regional grid. The RSM is
a primitive equation model, with similar physics
as the driving NCEP-DOE reanalysis II (R-2)
Global Spectral Model as reported in Kanamitsu et
al. (2002).
4
PHYSICAL INITIALIZATION FOR THE REGIONAL SPECTRAL
MODEL
RSM initial and boundary conditions were obtained
from the coarser-scale R-2 reanalysis (1.875
resolution) and 28 vertical levels. Daily rain
rates were provided by the Climate Prediction
Center (CPC) precipitation analysis (see Higgins
et al., 2001) over the U. S. domain. R-2
precipitation fields were used for the rest of
the model domain, including Mexico. The CPC
precipitation was provided on a regular grid of
0.25º. SSM/I-OLR precipitation estimates were
used in the simulations over South America. The
SSM/I-OLR estimate was provided on a Gaussian
grid of 0.7º. The NOAA/NESDIS SSM/I algorithm
(Ferraro and Marks, 1995) was used to estimate
the rain rates. All rainfall fields were
bi-linearly interpolated to the regional models
grid. SST (1 degree resolution) was taken from
the PIRCS data set.
5
PHYSICAL INITIALIZATION FOR THE REGIONAL SPECTRAL
MODEL
This scheme basically adjusts the humidity
profile using the difference between the
observed and predicted rain rates as factor of
this adjustment. In order to provide consistent
temperature profiles, the cumulus and large-scale
precipitation parameterizations are then
immediately called. This methodology differs
from the used by the FSU Nested Regional Spectral
Model (Nunes and Cocke, 2003), where a modified
Kuo parameterization is the convection scheme,
however the general PI procedure follows the same
structure as shown in Fig. 1.
OBSERVED RAIN RATES TIME STEP ASSIMILATED
PI-ANALYSIS
PHYSICAL INITIALIZATION SCHEME
FORECAST
DAY -1 ANALYSIS
DAY 0 ANALYSIS
Fig. 1 - General overview of the PI procedure
considering a continuous data assimilation
system.
6
PHYSICAL INITIALIZATION FOR THE REGIONAL SPECTRAL
MODEL
The experiments were performed during the
summertime. U. S. and Mexico - July-August-Septe
mber, starting at July 1st, 1999 at 0 UTC. South
America - January, starting at January 1st, 1999
at 0 UTC. The control simulations were not
initialized. In the PI simulations, the rain
rates were updated every 24 hours, and the
moisture adjustment took place every time-step
which was 3 min. The boundary conditions were
updated every 6 hours.
7
U. S. and MexicoMonthly Accumulated
Precipitation (mm)July 1999
SAS
Verification
RAS
8
U. S. and MexicoMonthly Accumulated
Precipitation (mm)August 1999
SAS
Verification
RAS
9
U. S. and MexicoMonthly Accumulated
Precipitation (mm)September 1999
SAS
Verification
RAS
10
South AmericaMonthly Accumulated Precipitation
(mm)January 1999
SAS
Verification
RAS
11
PHYSICAL INITIALIZATION FOR THE REGIONAL SPECTRAL
MODEL
A preliminary evaluation of the PI scheme results
was based on spatial correlation coefficient (r)
and bias given by The variables x and
y represent the predicted and observed
precipitations, and the subscript i, the grid
point on the Mercator projection.
12
U. S. and MexicoSpatial Correlation
CoefficientJuly 1999
13
U. S. and MexicoSpatial Correlation
CoefficientAugust 1999
14
U. S. and MexicoSpatial Correlation
CoefficientSeptember 1999
15
U. S. and MexicoBias July 1999
SAS
RAS
16
U. S. and MexicoBias August 1999
SAS
RAS
17
U. S. and MexicoBias September 1999
SAS
RAS
18
South AmericaSpatial Correlation
CoefficientJanuary 1999
19
South AmericaBias January 1999
SAS
RAS
20
U. S. and Mexico850 hPa Temperature (K)July 1999
SAS
R-2
RAS
21
U. S. and Mexico500 hPa Temperature (K)July 1999
SAS
R-2
RAS
22
U. S. and Mexico300 hPa Temperature (K)July 1999
SAS
R-2
RAS
23
U. S. and Mexico850 hPa Temperature (K)August
1999
SAS
R-2
RAS
24
U. S. and Mexico500 hPa Temperature (K)August
1999
SAS
R-2
RAS
25
U. S. and Mexico300 hPa Temperature (K)August
1999
SAS
R-2
RAS
26
U. S. and Mexico850 hPa Temperature
(K)September 1999
SAS
R-2
RAS
27
U. S. and Mexico500 hPa Temperature
(K)September 1999
SAS
R-2
RAS
28
U. S. and Mexico300 hPa Temperature
(K)September 1999
SAS
R-2
RAS
29
South America850 hPa Temperature (K)January 1999
SAS
R-2
RAS
30
South America500 hPa Temperature (K)January 1999
SAS
R-2
RAS
31
South America300 hPa Temperature (K)January 1999
SAS
R-2
RAS
32
U. S. and MexicoJuly 1999
33
U. S. and MexicoAugust 1999
34
U. S. and MexicoSeptember 1999
35
South AmericaJanuary 1999
36
PHYSICAL INITIALIZATION FOR THE REGIONAL SPECTRAL
MODEL
Preliminary evaluations of the PI implementations
indicate that the RSM was able to successfully
assimilate the merged precipitation analysis and
SSM/I estimates as well. The correlation
coefficients exceeded 0.9 and the spin-up problem
was noticeably reduced during the continuous
assimilation period. The temperature fields are
not degraded by using a rainfall nudging and they
are well correlated with the R-2 fields. We are
now attempting to implement a physically
initialized analyses as part of our effort to
develop useful downscaled reanalysis fields