downscaling ability of the ncep regional spectral model v.97: the big brother experiment
DESCRIPTION
Large Limited Area High Resolution Model. Small Limited-area High Resolution model. Deborah Herceg * Adam Sobel *†‡ Liqiang Sun ‡ Stephen Zebiak ‡ *Department of Applied Physics and Applied Mathematics Columbia University New York, NY †Department of Earth and Environmental Sciences - PowerPoint PPT PresentationTRANSCRIPT
Tropical Domain Results
Downscaling Ability of the NCEP Regional Spectral Model v.97: The Big Brother Experiment
Conclusions:
Motivation: The Big Brother Experiment methodology allows us to assess the errors in regional climate simulations which are solely due to the nesting strategy.
Deborah Herceg*
Adam Sobel*†‡
Liqiang Sun‡
Stephen Zebiak‡
*Department of Applied Physics and Applied Mathematics
Columbia UniversityNew York, NY
†Department of Earth and Environmental Sciences
Columbia UniversityNew York, NY
‡International Research Institute For Climate
PredictionColumbia University
Palisades, NY
Setup of the Big Brother Experiment:
Our Experiment : Domains
STEPS:
1.We first perform a large-domain high resolution regional model simulation, the “Big Brother”.
2.This reference simulation is then degraded by filtering out short-scales which would be unresolved in the GCM
3.This filtered reference is then used to drive the same regional climate model, integrated at the same resolution as the Big Brother, but over a smaller domain embedded in the large domain, called the Little Brother
4.We then compare the climate statistics of the Little Brother with those of the Big Brother on the Little Brother Domain
5.Since the same model is used for both the Big Brother and Little Brother experiments, the climate of the two should in principle be identical. Any difference in the solutions can only be attributed to the nesting strategy, and not to model error.TROPICAL
DOMAIN:MIDLATITUDE DOMAIN:
Midlatitude Domain Results
Experimental Details:
NOAA 29th Annual Climate Diagnostics and Prediction Workshop; October 18-22, 2004, Madison, Wisconsin
Table Legend:
The Big Brother Experiment (BBE)
Large Limited Area High Resolution
Model
Regional Climate Validation
Filter Small Scales
Small Limited-area High Resolution
model
Small regional High Resolution
Simulation
Corresponding Data Set
Model
IC and LBC
IC LBC
IC = Initial Condition
LBC = Lateral Boundary Condition
LEGEND:
Global Low-Resolution Data
Set
Large regional High Resolution
Reference Simulation
BIG BROTHER
LITTLE BROTHER
LITTLE BROTHER
BIG BROTHER
The Model: NCEP Regional Spectral Model, v. 97 – Juang et al., 1997.
ECHAM4.5 data, archived every 6 hours was used to drive the Big Brother simulation. The Big Brother simulation data was archived every 6 hours for the nesting of the Little Brother simulation, as well.All experiments were executed for a month. All of the statistical computations were performed excluding the first day in order to reject any short time-scale spin-up phenomenon. A ten grid point lateral boundary zone was also excluded. To mimic the resolution of the operational GCM (ECHAM at T-42 resolution), a cutoff wavelength of ~900 km was used in the filtering of the Big Brother output before using it to drive the Little Brother.
This experiment was first performed by Denis et al. (2001), using a grid point model for a midlatitude winter simulation. Our study extends theirs by using both a tropical and a midlatitude domain case, and using a different (in this case, spectral) model.
One month simulation completed for April, 1994Both domains centered at 37.5°W and 5°SHorizontal resolution of 60 kmBig Brother domain at 181 x 182 grid pointsLittle Brother domain at 91 x 92 grid points18 vertical layers
One month simulation completed for February, 2001Both domains centered at 70°W and 47°NHorizontal resolution of 50 kmBig Brother domain at 151 x 142 grid pointsLittle Brother domain at 81 x 77 grid points18 vertical layers
BIG BROTHER LITTLE BROTHER
BIG BROTHER LITTLE BROTHER
PRECIPITATION
SEA+LAND SEA LAND
Stat. Corr
.
Trans.
Corr.
Stat. Var.
Ratio
Trans. Var.
RatioStat. Corr.
Trans. Corr.
Stat. Corr.
Trans. Corr.
Total
0.8147
0.4257 1.317 1.4374 0.7841 0.405
0.8847 0.4834
LW0.8534
0.5382 1.2967 1.3442 0.8227 0.5061
0.9272 0.577
SW0.2247
0.1293 1.7362 1.8379
0.2075
0.0921
0.2618
0.1258
SPECIFIC HUMIDITY (850 MBAR)
SEA+LAND SEA LAND
Stat. Corr
.
Trans.
Corr.
Stat. Var.
Ratio
Trans. Var.
RatioStat. Corr.
Trans. Corr.
Stat. Corr
.Trans. Corr.
Total
0.9935
0.9336 1.0061 1.0188 0.9938 0.9345
0.9775 0.835
LW0.9936
0.9391 1.0064 1.0182 0.994 0.9363
0.9792 0.8448
SW0.9841
0.7504 0.9834 1.072
0.9848 0.845
0.981
0.8305
PRECIPITATION
SEA+LAND SEA LAND
Stat. Corr
.
Trans.
Corr.
Stat. Var. Rati
o
Trans.
Var. Rati
oStat. Corr.
Trans. Corr.
Stat. Corr
.Trans. Corr.
Total
0.9452
0.7382
1.0639
1.0502
0.9663 0.7258
0.8844 0.6980
LW0.9589
0.7941
1.0492
1.0388
0.9750 0.7944
0.9153 0.7524
SW0.8211
0.4140
1.1863
1.1889
0.8232
0.4149
0.8196
0.4750
SPECIFIC HUMIDITY (850 MBAR)
SEA+LAND SEA LAND
Stat. Corr
.
Trans.
Corr.
Stat. Var. Rati
o
Trans.
Var. Rati
oStat. Corr.
Trans.
Corr.Stat. Corr.
Trans. Corr.
Total
0.9972
0.9641
1.0682
1.0482
0.9981
0.9621
0.9961 0.9526
LW0.9973
0.9645
1.0676
1.0471
0.9982
0.9627
0.9961 0.9558
SW0.9972
0.9588
1.0795
1.0899
0.9976
0.9645
0.9965
0.9494 Total Field (TOTAL) is separated into a small-scale
component (SW), and large-scale component (LW); defined as the wave numbers above and below the cutoff wave number.Stationary Correlation (Stat. Corr.) = spatial correlation of time-mean fields. Stationary Variance Ratio (Stat. Var. Ratio)=ratio of variances of time-mean fields.Transient Correlation (Trans. Corr.), Transient Variance Ratio (Trans. Var. Ratio) = spatial correlation, variance ratio of transient fields (total-stationary).Statistics provided for sea and land separately, as well as total.
In most fields, good agreement between Big and Little Brothers was obtained, for both small and large scales, after a brief initial spin-up period.
Precipitation was the exception. For our tropical domain experiment, the small-scale features of the precipitation field in the Big Brother were not reproduced by the Little Brother. The spatial and temporal variations of the large-scale component of the precipitation were well reproduced.
The reason for the poor reproduction of the small-scale component of the precipitation is that this component is inherently unpredictable, at least in this model, rather than because of any inherent flaw in the nesting strategy. This was shown by an ensemble of identically forced Little Brother simulations which were initialized at different times differing by 6-24 hours, in which the correlations of the small-scale precipitation between these different simulations was nearly as low as that between the Big and Little Brothers. Related to this, the agreement between Big and Little Brothers is fairly insensitive to the choice of the cutoff wave number used for the filtering.
Good reproduction of the small-scale precipitation was obtained for a wintertime midlatitude simulation. Thus, the relatively poor reproduction of this field for the tropical case appears to be due to specifically tropical (convective) precipitation processes.
“Ensemble” Simulations Results
ZONAL WIND (850 MBAR)
SEA+LAND SEA LAND
Stat. Corr
.
Trans.
Corr.
Stat. Var. Rati
o
Trans.
Var. Rati
oStat. Corr.
Trans.
Corr.Stat. Corr.
Trans. Corr.
Total
0.9936
0.9336
1.1407
1.1116
0.9951 0.9343 0.992 0.924
LW0.99
40.9344
1.1411
1.1121
0.9955 0.9354
0.9926 0.9263
SW0.9860
0.9202
1.1326
1.1137
0.9885
0.9278
0.9823
0.9113
ZONAL WIND (850 MBAR)
SEA+LAND SEA LAND
Stat. Corr
.
Trans.
Corr.
Stat. Var.
Ratio
Trans. Var.
RatioStat. Corr.
Trans. Corr.
Stat. Corr.
Trans. Corr.
Total
0.9762 0.918 0.8878 0.9766 0.977 0.9168
0.9836 0.8754
LW0.9768
0.9227 0.8879 0.9748 0.9773 0.9194
0.9847 0.8821
SW0.9600
0.786 0.8726 1.0451
0.9667
0.8440
0.9489
0.8028
Stat. Corr.
Stat. Var.
RatioTrans. Corr.
Trans. Var. Ratio
T=06 HR
Total 0.9025 0.9382 0.4744 1.0981
LW 0.9400 0.9396 0.5905 1.1085
SW 0.3561 0.9120 0.1484 1.0779
T=12 HR
Total 0.8742 0.8398 0.4346 1.0747
LW 0.9141 0.8410 0.5406 1.0809
SW 0.3158 0.8013 0.1231 1.0836
T=18 HR
Total 0.7747 0.9637 0.3590 1.2473
LW 0.8176 0.9540 0.4544 1.2457
SW 0.2266 0.9423 0.0783 1.3260
T=24 HR
Total 0.8845 1.0883 0.4398 1.0224
LW 0.9274 1.0969 0.5601 1.0428
SW 0.2343 1.0506 0.0765 1.0228
PRECIPITATION FIELD ENSEMBLES
The table on the left shows the “ensemble” results comparing the identically forced Little Brother simulations, but initialized at different times differing by 6-24 hours; the correlations of the small-scale precipitation between these different simulations was almost as low as that between the Big and Little Brother “ensemble” simulations (not shown here).