Breeding with the NSIPP global coupled model: applications to ENSO prediction and data assimilation

1
Breeding with the NSIPP global coupled model
applications to ENSO prediction and data
assimilation

• Shu-Chih Yang
• Advisors Profs. Eugenia Kalnay and Ming Cai

2
Outline

• Introduction
• Objectives
• NASA/NSIPP CGCM
• Breeding method
• Results from a 10-year perfect model experiment
• Comparison with breeding in NCEP CGCM
• Summary
• NSIPP operational system preliminary results

3
Introduction

• ENSO simulation
• Because the coupled nature of ENSO phenomenon,
  the key factor to simulate and predict ENSO lies
  in the correct depiction of SST.
• ENSO prediction skill
• The prediction skill of a coupled model can be
  significantly improved through more refined
  initialization procedures (ex Chen et al.,1995
  and Rosati et al, 1997)
• Initialization of operational ensemble forecast
  for CGCMs
• Two-tier (Bengtsson et al., 1993)
• An ensemble of atmospheric forecast generated by
  a forecasted SST
• One-tier (Stockdale et al., 1998, adopted in
  ECMWF)
• Generate all the ensemble members via CGCM
• Initial perturbations are introduced in
  atmosphere components only

4

• How to construct effective ensemble members?
• 2 methods have been considered to construct
  initial perturbations
• Singular vectors have been used for ENSO
  prediction with the Cane and Zebiak model
• Limitations
• Strong dependence on the choice of norm and
  optimization time
• High computational cost makes it impractical for
  CGCMs

5

• Breeding method
• Toth and Kalnay (1996)
• Cai et al. (2002) with CZ model
• Bred vectors are sensitive to the background
  ENSO, showing that the growth rate is weakest at
  the peak time of the ENSO states and strongest
  between the events.
• Bred vectors can be applied to improve the
  forecast skill and reduce the impact of the
  spring-barrier.
• The results show the potential impact for
  ensemble forecast and data assimilation

6
Spring Barrier The dip in the error growth
chart indicates a large error growth for
forecasts that begin in the spring and pass
through the summer. Removing the BV from the
initial errors reduces the spring barrier
7
Improvement on ensemble forecasts
FCT error with ?BV
FCT error with ?RDM
8
Objectives of this research

• Implement the breeding method with the NASA/NSIPP
  CGCM
• Construct effective perturbations for initial
  conditions of ENSO ensemble forecasts
• Test methods first with a perfect model
  simulation to develop understanding
• Apply methodology to NSIPP operational system,
  which is more complex (e.g. model errors)
• The ultimate goals is to improve seasonal and
  interannual forecasts through ensemble
  forecasting and data assimilation using coupled
  breeding

9
NASA Seasonal-to Interannual Prediction (NSIPP)
coupled GCM

• AGCM (Suarez, 1996)

Model features Primitive equations Empirical cloud diagnostic model 4th-order version of the enstrophy conserving scheme 4th-order horizontal advection schemes for potential temperature, moisture Penetrative convection parameterized with Relaxed Arakawa-Schubert scheme
Coordinates Finite-difference C grid in horizontal Generalized sigma coordinate
Resolution 2?? 2.5??34 levels
10

• OGCM
• Poseidon V4, (Schopf and Loughe,1995)

Model features Quasi-isopycnal model Reduced-gravity formulation Turbulent well-mixed layer with entrainment parameterized according to a Kraus-Turner bulk mixed layer model Vertical mixing and diffusion are parameterized using a Richardson number dependent scheme Horizontal mixing is implemented with high order Shapiro filtering
Coordinates generalized horizontal and vertical coordinates
Resolution 1?3?? 5?8?? 27 layers
11
Current prediction skill from NSIPP CGCM
hindcasts
Niño-3 Forecast SST anomalies up to 9-month lead
April 1 starts
September 1 starts
12
Breeding method
?

• Bred vectors
• The differences between the control
  forecast and perturbed runs
• Tuning parameters
• Size of perturbation
• Rescaling period (important for coupled system)
• Advantages
• Low computational cost
• Easy to apply to CGCM

13
10 years breeding perfect model experiment 10 years breeding perfect model experiment 10 years breeding perfect model experiment
Breeding Breeding Size of perturbation 10 of the RMS of the SSTA (0.085?C) Rescaling period one month
CGCM AGCM NSIPP-1 3 X 3.75 X 34 (global)
CGCM OGCM Poseidon V4 1/2 X 1.25 X 27 (90S - 72N)
14
Snapshot of background SST (color) and bred
vector SST (contour)
model year JUN2024
Instabilities associated with the equatorial
waves in the NSIPP coupled model are naturally
captured by the breeding method
15
Lead/lag correlation between BV growth rate and
absolute value of background NINO3 index
?
16
Lead/lag correlation between BV growth rate and
absolute value of BV NINO3 index
17
EOF analysis of SST
Background SST anomaly EOF1 (46)
BV SST EOF1 (11)
18
EOF analysis of thermocline (Z20)
Background Z20 EOF1 (22)
BV Z20 EOF1 (10)
Background Z20 EOF2 (16)
BV Z20 EOF2 (7)
Z20 EOF2, SST EOF1 represent the mature phase of
ENSO
19
Oceanic maps regressed with PCs
BV regressed with Z20 PC1
Background regressed with SST PC1
SST
Thermocline (Z20)
Surface zonal current
20
Atmospheric maps regressed with PCs
Tropical Pacific domain
BV
Background
Wind at 850mb
Surface pressure
Geopotential at 500mb
OLR
21
Atmospheric maps regressed with PCs
Northern Hemisphere
BV
Background
Sea-level pressure
Geopotential at 200mb
Even though the breeding rescaling is in the
Nino3 region, the atmospheric response is global
22
Atmospheric maps regressed with PCs
Southern Hemisphere
BV
Background
Sea-level pressure
Geopotential at 200mb
23
Lead/lag regression maps
BV zonal wind stress vs. CNT NINO3
BV SST vs. CNT NINO3
BV surface height vs. CNT NINO3
Bred vector lags ENSO episode in the Central
Pacific
?
Bred vector leads ENSO episode in the Eastern
Pacific

24
NASA/NSIPP BV vs. NCEP/CFS BV
Tropical Pacific domain
NCEP
NSIPP
SST
Z20
?x
25
NASA/NSIPP BV vs. NCEP/CFS BV
NCEP
NSIPP
SST EOF1
Z20 EOF1
Z20 EOF2

• Results obtained with a 4-year NCEP run are
  extremely similar to ours

26
NASA/NSIPP BV vs. NCEP/CFS BV
Northern Hemisphere
NSIPP geopotential height at 500mb
NCEP geopotential height at 500mb
27
Summary of perfect model results

• Larger BV growth rate leads the warm/cold events
  by about 3 months.
• The amplitude of BV in the eastern tropical
  Pacific increases before the development of the
  warm/cold events.
• The ENSO related coupled instability exhibits
  large amplitude in the eastern tropical Pacific.
• In N.H, BV teleconnection pattern reflect their
  sensitivity associated with background ENSO.
  Rossby wave-train atmospheric anomalies over both
  Hemispheres.
• Breeding method is able to isolate the slowly
  growing coupled ENSO instability from weather
  noise
• Bred vectors can capture the tropical instability
  waves
• Results of a perfect model experiment with the
  NCEP CGCM are very similar

28
Current work

• Develop breeding strategy for the NASA/NSIPP
  coupled operational forecasting system
• Perform breeding runs with different rescaling
  norms
• Perform experiments with a modified breeding
  cycle to reduce spin-up
• ?
• Replace the restart file from an AMIP run to NCEP
  atmospheric re-analysis data

B2month
F1month
?
B
B
?
A
?
?
?
?
t1
t2
t3
t4
t5
29
Relationship between bred vectors and background
errors
BV Temp (contour) vs. analysis increment (color)
at OCT1996
This case was chosen because the BV growth rate
was large. The excellent agreement suggests that
the operational OI could be improved by
augmenting the background error covariance with
the BV as in Corazza et al, 2002
30
For this case, we performed the first ensemble
forecast (BV fcst)(-BV fcst)/2
SST Analysis - Control forecast
OCT1996
Analysis BV ensemble ave fcst
OCT1996
31

• Summary of plans for application to the
  operational NSIPP system
• Develop a strategy to include the coupled growing
  modes extracted from coupled bred vectors in the
  initial condition of the ensemble system For
  example, use perturbations BV and BV with an
  appropriate amplitude in the ensemble forecast
  system
• Develop a methodology for using advantage of the
  ENSO BVs within the operational NSIPP ocean
  ensemble data assimilation For example, augment
  the OI background error covariance with BVs.

32
(No Transcript)
33
BV Geopotential at 500mb
NSIPP
NCEP
34
From 10 year perfect model simulation
35
Joint EOF map of BV SST
36
BV1 Z20PC1 vs. BV1 growth rate
Growth rate
Z20 PC1
BV2 Z20PC1 vs. BV2 growth rate
Growth rate
Z20 PC1
37
CNT
Background Z20 EOF1
Background Z20 PC1
Background Z20 EOF2
Background Z20 PC2
38
Background ENSO vs. ENSO embryo
CNT EOF1
BV1 EOF1
BV2 EOF1
CNT EOF2
BV1 EOF2
BV2 EOF2
39
BV growth rate
BV SST vs. (SSTfcst-SSTa)
MAR1996
40
BV regression maps constructed with Z20 PC1
41
Vertical cross-section along the Equator
Color Tfcst-Ta Contour BV (SST norm)
Color Tfcst-Ta Contour BV (Z20 norm)
JAN2000
42
Vertical cross-section along the Equator
Color Tfcst-Ta Contour BV (SST norm)
Color Tfcst-Ta Contour BV (Z20 norm)
MAR1996
43
(No Transcript)
44
(No Transcript)