11
The Impact of Coupled versus Observed SST on Summer Season Predictions over America
with the NCEP CFS Land Upgrades
Rongqian Yang, Michael Ek, Jesse Meng and Ken Mitchell
NCEP Environmental Modeling Center NCEP-COLA CTB Joint Seminar Series
24 March 2010
2
Objectives Objectives
A.A. Examine the impact of land upgrades in the next-Examine the impact of land upgrades in the next-
generation Climate Forecast System (CFS) on summer generation Climate Forecast System (CFS) on summer season predictions. Any meaningful improvements?
season predictions. Any meaningful improvements?
B. Examine the relative contribution from land anomaly B. Examine the relative contribution from land anomaly forcing to summer seasonal predictability ( vs.SST forcing to summer seasonal predictability ( vs.SST anomaly forcing).
anomaly forcing).
Motivation Motivation
SST anomalies
SST anomalies : : the foremost source of seasonal predictabilitythe foremost source of seasonal predictability in coupled global models.
in coupled global models.
Land surface anomalies
Land surface anomalies:: the secondthe second most important source most important source of seasonal predictability (e.g. anomalies of
of seasonal predictability (e.g. anomalies of soil moisturesoil moisture, , snowpack, vegetation cover).
snowpack, vegetation cover).
33
Outline Outline
Overview of Overview of Currently Operational CFS Currently Operational CFS Skill Skill
Next-generation CFS Next-generation CFS Upgrades (with land) Upgrades (with land)
CFS Experiments CFS Experiments
Part A: Part A: Coupled SST (CMIP) runs and resultsCoupled SST (CMIP) runs and results
Part B: Part B: Observed SST (AMIP) runs and comparison with part AObserved SST (AMIP) runs and comparison with part A
Part C: Part C: Ocean-land-atmosphere interactionOcean-land-atmosphere interaction comparison between A & B comparison between A & B
Conclusions Conclusions
4
Overview of Overview of
Currently Operational CFS Currently Operational CFS Seasonal Prediction Skill Seasonal Prediction Skill
(next 3 frames)
(next 3 frames)
55
Ops CFS Seasonal
Ops CFS Seasonal SST SST Forecast Skill Forecast Skill : :
Correlation of CFS SST forecast with observed SST: 1982-2003 Correlation of CFS SST forecast with observed SST: 1982-2003
For For April initial conditionsApril initial conditions: 15-member ensemble mean: 15-member ensemble mean 1-Month Lead (valid summer) 6-Month Lead (valid winter)
high correlation skill in tropical Pacific high correlation skill in tropical Pacific
66
Ops CFS Summer Season Ops CFS Summer Season Precipitation
Precipitation Forecast Skill Forecast Skill : :
Correlation of CFS precip forecast with observed precip: 1982-2003 Correlation of CFS precip forecast with observed precip: 1982-2003
For For April initial conditionsApril initial conditions: 15-member ensemble mean: 15-member ensemble mean
Short-lead summer forecast correlation skill is low across bulk of CONUS (lower than longer-lead winter forecast)
77
Ops CFS Summer Season Ops CFS Summer Season Temperature
Temperature Forecast Skill Forecast Skill : :
Correlation of CFS 2m-T forecast with observed 2m-T: 1982-2003 Correlation of CFS 2m-T forecast with observed 2m-T: 1982-2003
For For April initial conditionsApril initial conditions: 15-member ensemble mean: 15-member ensemble mean 1-Month Lead (valid summer) 6-Month Lead (valid winter)
Short-lead summer forecast correlation skill is low across bulk of CONUS
8
Next-generation CFS Upgrades Next-generation CFS Upgrades
Upgrades to CFS physics Upgrades to CFS physics
– Atmosphere (with co2 trend added), Atmosphere (with co2 trend added) , ocean, ocean , land land model
model , , sea-ice sea-ice
New 4DDA analysis systems (CFSR) New 4DDA analysis systems (CFSR) – Atmosphere, ocean, Atmosphere, ocean, land land
Double the CFS resolution Double the CFS resolution
– T126 / L64 T126 / L64 (versus T62 / L28) (versus T62 / L28)
More details about the CFS seeMore details about the CFS see
Saha et al., The NCEP Climate Forecast System, 2006, J. Clim, 19(15), 3483-3517 Saha et al., The NCEP Climate Forecast System, 2006, J. Clim, 19(15), 3483-3517
The CFS experiments presented below incorporate the CFS upgrades (land related upgrades highlighted above in red)
99
Noah LSMNoah LSM
– 4 soil layers (10, 30, 60, 100 cm)4 soil layers (10, 30, 60, 100 cm) – Frozen soil physics includedFrozen soil physics included
– Surface fluxes weighted by snow cover Surface fluxes weighted by snow cover fraction
fraction
– Improved seasonal cycle of vegetation Improved seasonal cycle of vegetation cover
cover
– Spatially varying root depthSpatially varying root depth
– Runoff and infiltration account for sub-grid Runoff and infiltration account for sub-grid variability in precipitation & soil moisture variability in precipitation & soil moisture – Improved soil & snow thermal conductivityImproved soil & snow thermal conductivity – Higher canopy resistanceHigher canopy resistance
– OtherOther
Land Model
Land Model Upgrade Upgrade
Noah LSM (new) versus OSU LSM (old):
Noah LSM (new) versus OSU LSM (old):
OSU LSMOSU LSM
– 2 soil layers (10, 190 cm)2 soil layers (10, 190 cm) – No frozen soil physicsNo frozen soil physics
– Surface fluxes not weighted by Surface fluxes not weighted by snow fraction
snow fraction
– Vegetation fraction never less than Vegetation fraction never less than 50 percent
50 percent
– Spatially constant root depthSpatially constant root depth
– Runoff & infiltration do not account Runoff & infiltration do not account for subgrid variability of
for subgrid variability of precipitation & soil moisture precipitation & soil moisture – Poor soil and snow thermal Poor soil and snow thermal
conductivity, especially for thin conductivity, especially for thin snowpack
snowpack
Noah LSM replaced OSU LSM in operational NCEP medium-range
Global Forecast System (GFS) in late May 2005
More details see:
Ek et al., 2003: Implementation of Noah land surface model advances in the National Centers for Environmental Prediction operational mesoscale Eta model. J. Geophys.
Res., 108(D22).
10
Land Data Assimilation System
Land Data Assimilation System Upgrade Upgrade
GLDAS/Noah
GLDAS/Noah & Global Reanalysis 2 ( & Global Reanalysis 2 ( GR2/OSU GR2/OSU ): ):
GLDAS GLDAS : : an an uncoupled uncoupled G G lobal lobal L L and and D D ata ata A A ssimilation ssimilation S S ystem ystem driven by observed driven by observed
precipitation
precipitation analyses (CPC CMAP analyses) analyses (CPC CMAP analyses)
– Executed using Executed using same grid, land mask, terrain field and same grid, land mask, terrain field and four-layer
four-layer Noah LSMNoah LSM as in experimental CFS forecastsas in experimental CFS forecasts – Non-precipitation land forcing is from GR2Non-precipitation land forcing is from GR2
– Executed retrospectively from 1979-2006 (after spin-up)Executed retrospectively from 1979-2006 (after spin-up)
GR2 GR2 : : a a coupled coupled atmosphere/land assimilation atmosphere/land assimilation system wherein land component is
system wherein land component is driven by model driven by model predicted precipitation
predicted precipitation
– applies the applies the OSU LSMOSU LSM with two soil layerswith two soil layers
– nudges soil moisture based on differences between nudges soil moisture based on differences between model and CPC CMAP precipitation
model and CPC CMAP precipitation
1111
01 May
Similar climatology (% volume) 01 May 1999
Different anomaly (% volume)
Time series of Total SM (mm) over Illinois (81-04)
Annual Cycle (mm)
Noah/GLDAS has higher SM
Soil Moisture
Soil Moisture Difference between GLDAS Difference between GLDAS and GR2 over CONUS
and GR2 over CONUS
(land anomaly forcing comparison) (land anomaly forcing comparison)
90-day ending 01 May 99 Precip Anomaly &
climatology
01 May 1999
Soil Moisture Anomaly disagree.
Noah/GLDAS Closer to Observed Precip
Anomaly over western CONUS
GR2 GLDAS
12
How the land upgrades in the new CFS system
perform? Any improvements?
Question?
1313
CFS Land Experiments: Part A
CMIP Runs (Coupled SST) Noah/GLDAS
vs
OSU/GR2
Highly Controlled
• 25-year summer reforecasts (80-04)
• 10 member (ICs from 00Z of Apr 19-23, Apr 29-May 03)
• Same atmospheric ICs, physics, and Oceanic initial states
• Same resolution (T126/L64)
14
Verification Methods and Data Verification Methods and Data
Precip: 1/8Precip: 1/8thth NLDAS (Mitchell et al.,2004, JGR 109 ) and CMAP NLDAS (Mitchell et al.,2004, JGR 109 ) and CMAP
SST:1X1 degree daily OI SST (Reynolds et al,2002, J. Clim,15) with CMIPSST:1X1 degree daily OI SST (Reynolds et al,2002, J. Clim,15) with CMIP
T2m: monthly T126 (Fan and H. van den Dool, 2008, JGR 113)T2m: monthly T126 (Fan and H. van den Dool, 2008, JGR 113)
Atmospheric data: 2.5x2.5 degree NCEP/DOE GR2Atmospheric data: 2.5x2.5 degree NCEP/DOE GR2
Main Skill Measures: Anomaly Correlation (AC)
Area-averaged AC scores (AAC)
Percentage Count of Positive AC (PAC)
Anomaly Cross Correlation
Focusing on
CONUS summer season
(June-July-August)1515
CFS Noah/GLDAS CFS Noah/GR2
CFS OSU/GR2
JJA CFS AC skill: SST (25 yrs ensemble mean)
The two configurations yield similar SST correlation patterns
0.1 0.6 0.9 0.1 0.6 0.9
Slightly better with Noah/GLDAS
over higher latitudes
Similar SST Performance over Nino 3.4
16
Time series of predicted JJA Nino 3.4 SST Anomaly over the 25 yrs
Predicted JJA Niño 3.4 anomaly and Obs
Good Agreement with Observations over most years
Similar performance
over most years &
1999 (defined as 5˚S-5˚N, 120˚W-170˚W)
(defined as 5˚S-5˚N, 120˚W-170˚W)
So the difference in CFS performance (next) is likely due to land upgrades
2.0
0
-2.0 0.5
-0.5
1717
0.6
- 0.6
0.6
- 0.6
JJA CFS AC Skill: precipitation (25 yrs ensemble mean)
Better Performance
over Pacific Northwest and northern Great
Plains
Better performance
over Central great plains and the
Gulf States
Positive AC points (%) are 64.9 (Noah/GLDAS & 58.4 (OSU/GR2)
18
CONUS-average Anomaly Correlation:
CFS JJA ensemble mean precipitation forecasts
from the 25-year 10-member reforecasts of the two CFS configurations
Improvement with Noah/GLDAS, but the score is still low
1919
Anomaly Correlation: CFS JJA ensemble mean
Anomaly Correlation: CFS JJA ensemble mean temperaturetemperature forecasts forecasts
from the 25-year reforecasts of 2 T126 CFS tests from the 25-year reforecasts of 2 T126 CFS tests
(10 members each from same late April and early May initial times) (10 members each from same late April and early May initial times)
better performance with OSU/GR2
from Mid-west
to
The Great Plains
20
CONUS-average Anomaly Correlation:
CFS JJA ensemble mean temperature forecasts
from the 25-year 10-member reforecasts of the two CFS configurations
2121
Partition into ENSO
Partition into ENSO Neutral Neutral & & Non-neutral Non-neutral samples using
samples using MJJ MJJ Nino3.4 SST anomaly of Nino3.4 SST anomaly of 0.7C 0.7C as a threshold magnitude. as a threshold magnitude.
10 non-neutral summers:
82,83,87,88,91,92,93,97,99,02 (red: warm, blue: cold)
15 neutral summers:
80,81,84,85,86,89,90,94,95,96,98,00,01,03,04
Further Analysis Further Analysis
To examine the CFS performance with
different ENSO signals
22
10 10 non-neutral non-neutral ENSO years: ENSO years:
JJA
JJA precipitation precipitation AC Score AC Score
Much better performance
than 25-yr avgs
Due to Strong SST signals
Extremely similar geographical patterns from
west of the Mississippi River to the Pacific states
2323
15 15 neutral neutral ENSO years: ENSO years:
JJA
JJA precipitation precipitation AC Score AC Score
Worse performance
than 25-yr avgs As expected
Weak SST impact
Similar to 25-yr avgs
The large differences are mostly over the western CONUS
2424
CONUS-average JJA precipitation AC score
Significance test (T-statistic) shows differences are
significant at 90%
confidence level.
Neutral years
0.04
-0.15 0.00
Significance test (T-statistic) shows differences are not significant at 90%
confidence level.
Non-Neutral years
0.18
0 0.10
2525
10 10 non-neutral non-neutral ENSO years: ENSO years:
JJA
JJA temperature temperature AC score AC score
Due to Strong SST signals
Similar geographical
patterns With both CFS
Slightly better with OSU/GR2
26
15 15 neutral neutral ENSO years: ENSO years:
JJA
JJA temperature temperature AC score AC score
Degraded Performance with
both cases
Slightly better over with Noah/GLDAS
Over the Rocky Mountain states and the N. Pacific
states
Disagree over the southern Great
Plains
2727
CONUS-average JJA temperature AC score
Neutral years
0.2
0
Significance test (T-statistic) shows differences are not significant at 90%
confidence level.
Non-Neutral years
0.5
0 0.2 0.4
0.14
28
Question?
Question?
Is the skill gain really from the land upgrades Is the skill gain really from the land upgrades
Noah/GLDAS Noah/GLDAS
or or
better SST?
better SST?
CFS Land Experiments: Part B
Replace the
coupled SST
in CFS Experiments Part Awith
observed SST
(Extremely Controlled AMIP runs)
To answer:
How are the AMIP runs compared to the CMIP runs?
How are the AMIP runs compared to the CMIP runs?
2929
CMIP and AMIP Precip AC Skill Avgd over 25 yrs
AAC
PAC
C-Noah A-Noah C-OSU A-OSU
AMIP OSU CMIP Noah
CMIP OSU
AMIP Noah
Noah performs better in CMIP AMIP loses skill in these regions with both CFS
3030
T2m AC skill averagd over the 25 yrs
AAC
PAC
Noah performs worse with AMIP, no big difference with OSU
AMIP OSU CMIP Noah AMIP Noah
CMIP OSU
3131
Precip AC skill averaged over the 10 non-neutral yrs
AAC
PAC
AMIP runs are better than CMIP with both LSMs during non-neutral yrs, but not statistically
significant (90%)
CMIP Noah
AMIP OSU CMIP OSU
AMIP Noah
3232
Precip AC skill averaged over the 15 neutral yrs
AAC
PAC
Noah is better than OSU, the difference is
statistically significant (90%) in AMIP
CMIP Noah AMIP Noah
CMIP OSU AMIP OSU
Noah is better than OSU,
statistically significant (90%) in CMIP
3333
T2m AC skill averaged over the 10 non-neutral yrs
AAC
PAC
No big difference with either LSM in both CMIP/AMIP modes – no clear advantage -- Mixed in both AAC and PAC
AMIP OSU CMIP Noah AMIP Noah
CMIP OSU
3434
T2m AC skill avgd over the 15 neutral yrs
AAC
PAC
Both LSMs perform slightly better with AMIP; All differences in T2m are not statistically significant at 90% confidence level
AMIP OSU CMIP Noah AMIP Noah
CMIP OSU
3535
Wet Southwest U.S.
Wet Southwest U.S. Monsoon CaseMonsoon Case Predicted Precipitation Anomaly (in mm)
1999: Predicted
1999: Predicted JJA JJA
-50 50
No big difference over Southwest With Noah/GLDAS
Bad performance over Southwest
With OSU/GR2
CMIP Noah/GLDAS performs better over these areas than AMIP Noah/GLDAS
CMIP Noah/GLDAS AMIP Noah/GLDAS
CMIP OSU/GR2 AMIP OSU/GR2
Obs
36
Why the AMIP runs are not as good as CMIP runs?
Perfect SST means the upper limit of predictability or only true with perfect atmosphere and land surface?
Ocean-Land-Atmosphere Ocean-Land-Atmosphere
interaction comparison interaction comparison
Question?
Represented by Represented by
SST SST , , 500 GPH, 500 GPH, and and Soil Moisture Soil Moisture
3737
GR2
CMIP Noah AMIP Noah
CMIP OSU AMIP OSU
SST- 500GPH Anomaly Cross Correlation
All have good agreements with
observations, but
weaker in CMIP and stronger in AMIP
3838
500 GPH AC skill Averaged over 25 yrs
CMIP : Noah better over CONUS; AMIP worse than CMIP over bulk of CONUS (Noah better over ocean)
AMIP OSU AMIP Noah
CMIP OSU CMIP Noah
3939
Comparison of Predicted 500GPH JJA Climatology with GR2
The JJA 500GPH Climatology is
too low compared to GR2 in AMIP
where CMIP shows better agreement with Obs
AMIP OSU CMIP Noah AMIP Noah
CMIP OSU GR2
588
588
588 588
588
40
JJA Soil Moisture and 500 GPH Anomaly Cross Correlation
Negatively Correlated in both GLDAS and GR2
4141
Predicted JJA Soil Moisture and 500 GPH Cross Correlation
CMIP Noah AMIP Noah
CMIP OSU AMIP OSU
Predict wrong signs with AMIP runs
42
Conclusions Conclusions
The land upgrades in CFS do improve summer season The land upgrades in CFS do improve summer season precipitation
precipitation prediction over CONUS, especially prediction over CONUS, especially during during ENSO-neutral years in both modes
ENSO-neutral years in both modes when the when the land anomaly land anomaly forcings
forcings and and land-atmospheric interactionsland-atmospheric interactions contribute more contribute more to seasonal predictability.
to seasonal predictability.
AMIP runs are usually assumed to be the upper limit of
potential predictive skill, but they ignore feedbacks from the atmosphere, and lead to degraded large-scale atmospheric circulation performance which contributes to the skill loss over CONUS (on average).