Seasonal Predictability of SMIP and SMIP/HFP

Seasonal Predictability of

SMIP and SMIP/HFP

In-Sik KangJin-Ho Yoo, Kyung Jin, June-Yi Lee

Climate Environment System Research CenterSeoul National University

SMIP (Seasonal prediction Model Intercomparison Project)

Organized by World Climate Research Programme Climate Variability and Predictability Programme (CLIVAR) Working Group on Seasonal to Interannual Prediction (WGSIP) Coordinators G. Boer(CCCma), M. Davey (UKMO), I.-S. Kang (SNU), and K. R. Sperber (PCMDI)

Purpose

Investigate 1 or 2 season potential predictability based on the initial condition and observed boundary condition

SMIP Experimental Design

- Model Integration : 7 month x 4 season x 22 year (1979-2000), 6 or more ensembles- 4 institute 5 models have been participated. : NCEP (USA), CCCma (Canada), SNU/KMA (Korea), MRI/JMA (Japan)

Model Institute Resolution Experiment Type

NCEP NCEP T62L28 SMIP (10 member)

GDAPS KMA T106L21 SMIP (10 member)

GCPS SNU/KMA T63L21 SMIP (10 member)

NSIPP NASA 2ox2.5o L43 AMIP (9 member)

JMA JAPAN T63L40 SMIP (10 member)

Participating Models

Total Variance of JJA Precipitation Anomalies

(a) CMAP (21yr)

(d) NASA (21yr×9member)

(b) SNU (21yr×10member)

(e) NCEP (21yr×10member)

(c) KMA (21yr×10member)

(f) JMA (21yr×6member)

Analysis of Variance of JJA Precipitation Anomalies (SNU case)

(a) Total variance

(b) Forced variance

(c) Free variance

Free variance

Intrinsic transients due to natural variability

Forced variance

Climate signals caused by external forcing

N

ii XX

N 1

2)(1

1

N

i

n

jiij XX

nN 1 1

2)()1(

1

Forced Variance Free Variance Signal-to-noise

Forced Variance Error Variance Forced/Error Variance

Prediction Skill of JJA Precipitation during 21 years

(a) MME1(Model Composite)

(d) NASA

(b) SNU

(e) NCEP

(c) KMA

(f) JMA

Temporal Correlation with Observed Rainfall

Prediction Skill of JJA Precipitation-Global Pattern Correlation (a) SNU

(b) KMA

(c) NASA

(d) NCEP

(e) JMA

Previous DJF NINO3.4

Recent NINO3.4

Pattern Cor. for Ensemble mean

Pattern Cor. for each member

5 Model Mean

MME1 – Model Composite

NINO3.4

Monsoon Region (40-160E, 20S-40N)

Pattern Correlation

Prediction Skill of JJA Monsoon Rainfall

Preferable Pattern for Asian Monsoon Rainfall Prediction in Model

(a) Good Prediction

(b) Bad Prediction

(c) (a) - (b)

OISSTMME1 CMAP

(d) Good Prediction

(e) Bad Prediction

(f) Good Prediction

(g) Bad Prediction

Selected Cases

Good Prediction: 81’ 95’ 96’ 98’

Bad Prediction: 80’ 82’ 85’ 88’

SMIP/HFP (Historical Forecast Project)

HFP Procedure ( ex: prediction for summer: JJA)

5/1

6/1

7/1

8/1

8/31

6 ensembles : started from 4/28/00,12Z, 4/29/00,12Z 4/30/00,12Z (12hr interval)

Initial condition : Atmosphere NCEP Reanalysis anomaly + model climatology

Land surface NCEP Reanalysis

AGCM integration (4 month)

Global SST prediction

4/1

Predicted SST

Dynamical prediction

To carry out 7-month ensemble integrations of atmospheric GCMs with observed initial conditions and observed (prescribed) boundary conditions

SMIP2

To carry out 4-month ensemble integrations of atmospheric GCMs with observed initial conditions and predicted boundary conditions or Coupled GCM

SMIP2/HFP

1st and 2nd Season

Potential predictability

1st Season

Actual predictability

Characteristics of Prescribed SST and Predictability

(a) Temporal Correlation

(b) Ratio of Standard Deviation (c) RMS error

Comparison with OISST

Forced Variance Free Variance

Signal-to-noise

(a) SNU (b) KMA

(c) SNU/HFP (d) KMA/HFP

(a) SNU (b) KMA

(c) SNU/HFP (d) KMA/HFP

(a) SNU (b) KMA

(c) SNU/HFP (d) KMA/HFP

Forced Variance Error Variance

Forced/Error Variance

(a) SNU (b) KMA

(c) SNU/HFP (d) KMA/HFP

(a) SNU (b) KMA

(c) SNU/HFP (d) KMA/HFP

(a) SNU (b) KMA

(c) SNU/HFP (d) KMA/HFP

37.5%

21.3%

11.1%

27.7%

15.8%

8.5%

Observation Prediction

Time coefficients

Observation

Prediction

Eigen Vectors

1st Mode

2nd Mode

3rd Mode

EOF Analysis of Summer Mean SST

Change of SST Influence: Decreased Forced Variance

SMIP signal – HFP signal

Absolute value of COV of Prcp & CEP. SST

Central Equatorial SST : 180E-220E, 5S-5N

(a) SNU (b) KMA

(c) SNU (d) KMA

Influence of Regional SST on the Asian Monsoon Rainfall Predictability

(b) SNU

(a) Observation

(c) KMA

TPAC NPAC WPAC IDO Local

MME1

Prediction skill of JJA Precipitation during 1979-2002

Global Pattern Correlation (0-360E, 60S-60N)

KMA

SNU

Cor=0.30 Cor=0.08Cor=0.22 Cor=0.08

Cor=0.23 Cor=0.02Cor=0.08 Cor=0.03

Monsoon Pattern Correlation (40-160E, 20S-40N)

KMA

SNU

Cor=0.04 Cor=0.09Cor=0.03 Cor=0.05

Cor=0.06 Cor=-0.22Cor=0.01 Cor=-0.20

Prediction skill of JJA Precipitation during 1979-2002

Perfect Model Correlation of JJA Precipitation during 1979-1999

Monsoon Region (40-160E, 20S-40N)

Global Domain (0-360E, 60S-60N)

EOF Analysis of Summer Mean Precipitation

(a) CMAP

(d) NASA

(b) SNU

(e) NCEP

(c) KMA

(f) JMA

(d) MME1 (e) PC time series

EOF Analysis

Truncation of small scale noise modes by retaining first 10 EOF

modes

SVD Analysis

Couple pattern of observation and model

Transfer Function

Replace the model SVD mode to the corresponding observation mode

ObservationX (x , t)

Forecast FieldY* (x*, t)

EOFei (x) , ti (t)

SVDi = cor [Ti , Yi]

Si , Ti (t)

EOFtj (t) , ej (x*)

Yi (t) , Pi

Ri (x)

projection of Ti(t) into X

Reproduction of Systematic ErrorX (x,t) = i Yi(t) Ri (x)

Statistical Correction Procedure

Systematic bias correction

GCM prediction

GCM prediction

GCM prediction

GCM prediction

GCM prediction

MME1(composite)

MME2 (SVD based super ensemble)

Correctedprediction

Corrected prediction

Corrected prediction

Corrected prediction

Corrected prediction

Statistical Correction (Post-processing)

MME3

Specio-Ensemble prediction

Model Institute Resolution Experiment Type

NCEP NCEP T63L17 SMIP (10 member)

GDAPS KMA T106L21 SMIP (10 member)

GCPS SNU/KMA T63L21 SMIP (10 member)

NSIPP NASA 2ox2.5o L43 AMIP (9 member)

JMA JAPAN T63L40 SMIP (10 member)

Participated Model

Ensemble procedure

APCN Multi Model Ensemble prediction

Prediction SST used (real forecast)

Prediction skill of APCN Multi Model predictions

Pattern correlation precipitation over monsoon region (40E-160E, 20S-40N)

MME3 MME2 MME1 SNU KMA NASA NCEP JMA

Avg. Skill

79-99

0.45 0.39 0.250.20 0.15 0.25 0.26 0.21

0.42 0.39 0.35 0.32 0.40

00-02

0.41 0.22 0.150.10 -0.21 0.31 0.31 N/A

0.26 0.15 0.31 -0.22 N/A

SMIP/HFP history after statistical correction

MME3 with 5 models (only SNU & KMA are different : SMIP vs SMIP/HFP)

MME3 with SMIP type history for statistical correction

MME3 with SMIP/HFP type history for statistical correction

Prediction SST used (real forecast)

Prediction dataset has inconsistency in SST boundary condition. During 1979-1999, observed SST was used for SMIP type simulation. However, the forecast after 2000 used predicted SST in real forecast mode. Thus, SMIP/HFP can be more skillful for later stage due to consistency in boundary condition for statistical correction based on previous forecast history