ENSO Prediction Skill in the NCEP CFS
Renguang Wu
Center for Ocean-Land-Atmosphere Studies
Coauthors: Ben P. Kirtman (RSMAS, University of Miami) Huug van den Dool (CPC, NCEP, NOAA)
CTB Joint Seminar Series
February 3, 2010, NCEP
1
Spring prediction barrier
What is the spring prediction barrier?
A large drop in the prediction skill of eastern equatorial Pacific SST during boreal spring
2
The seasonality of forecast skill (NINO3 SST) May
Kirtman et al. 2001
Initial Month
Jan Dec
Lead time (month) 3
Spring Prediction Barrier: Attribution
Why is there a spring prediction barrier?
. Low variance of Equatorial Pacific SSTAs in boreal spring e.g., Xue et al.’94; Torrence and Webster’98; Clarke and van Gorder’99
. The physical argument: weak Walker Cell and minimum zonal pressure and SST gradient (Equatorial Pac) in spring, initial errors project strongly onto ENSO modes, leading to large error growth
Issue: Perfect ICs in observations are not necessarily perfect for models used in predictions
. The statistical argument: Signal/Noise ratio lowest in spring
But, cannot explain the El Nino versus La Nina skill difference 4
Spring Persistence Barrier: auto-lag cor drop
Reasons:
(1) The phase transition of ENSO (Torrence and Webster’98;
Clarke and van Gorder’99; Burgers et al.’05)
(2) The seasonal change in the ENSO variance (Xue et al.’94;
Balmaseda et al.’95; Schneider et al.’03)
(3) The seasonal change in the S/N ratio (Webster’95; Torrence and Webster’98)
(4) Biennial oscillation/component (Clarke and van Gorden’99;
Yu’05)
5
ENSO: tropical Pacific air-sea interaction
Q: Any prediction/persistence barrier in thermocline and wind?
A boreal winter prediction/persistence barrier in the heat
content/warm water volume (Balmaseda et al.’95; McPhaden’03) SST
heating winds
ocean waves thermocline
6
Winter prediction barrier
QuickTime™ and a decompressor
are needed to see this picture.
Balmesada et al.’95 r(HC’pred, HC’sim): a large drop in winter > a winter barrier
7
Winter persistence barrier
McPhaden’03
May Jan
8
Winter persistence barrier
Yu and Kao’07
May Jan
9
Questions
. How is the ENSO prediction skill in the CFS? Is there a spring prediction barrier?
. Can the CFS capture the persistence barrier?
. What are plausible reasons for the drop of skill in spring?
. Are these related to the S/N ratio?
. Is the prediction skill related to the ENSO phase, initial or current state, different between El Nino and La Nina?
10
NCEP CFS 24-year ensemble forecasts
CFS (Climate Forecast System) model
Atmosphere: NCEP GFS (Global Forecast System) T62 64 sigma levels
Ocean: GFDL MOM3
long 1degree, latitude 1/3 degree 10S-10N and 1 degree 30S/30N, 40 levels (27 levels 400m)
15 forecasts (each 9-month length): three groups
1st: 9th,10th,11th,12th,13th (atm) & 11th (ocn, pentad);
2nd: 19th,20th,21th,22th,23th (atm) & 21th (ocn, pentad);
3rd: 29th,30th, 1st, 2nd, 3rd (atm) & 1st (ocn, pentad)
11
Measure of the prediction skill & noise
. Anomaly correlation coefficient (ACC)
. Root-mean-square error (Interannual component) (RMSE) Three quantities: NINO3.4 SST, NINO3.4 d20, WEP taux . ACC or RMSE calculated based on ensemble mean or individual members (mean value displayed), similar results . Spread (noise): standard deviation of members with respect to ensemble mean
12
Background
Phase relationship:
EP SST
EP d20 WP wind
1 4-5 Observations
CFS [lead3]
13 WEP:130-170E,5S-5N
Correlation Skill
Target Month
Saha et al.’06
14
Correlation Skill
long-lead forecast
Dec SST July
Dec taux Jan d20
15
Correlation Skill
short-lead forecast Dec SST
Jan d20
Dec taux
Dec.
16
Is the spring skill drop due to noise?
If noise is critical to the low skill, then
We would expect to see large noise when the skill drops.
Is that so?
17
Correlation Skill vs Spread (noise)
NINO3.4 SST
WEP taux NINO3.4 d20
Target Month
Initial Month
18
NINO3.4 SST
NINO3.4 d20
WEP taux
Initial Month
Target Month
Correlation Skill vs Signal-to-Noise Ratio
19
Correlation Skill
“perfect model approach”
- skill drop due to noise -
Target Month
Saha et al.’06
20
Perfect Model Skill vs Prediction Skill
Target Month
21
Plausible Reasons for spring prediction barrier
Noise cannot explain the spring prediction barrier
What are the plausible reasons?
Bias in atmospheric model wind response
22
Spring Persistence Barrier
Reasons:
(1) The phase transition of ENSO (Torrence and Webster’98;
Clarke and van Gorder’99; Burgers et al.’05)
(2) The seasonal change in the ENSO variance (Xue et al.’94;
Balmaseda et al.’95; Schneider et al.’03)
(3) The seasonal change in the S/N (Webster’95; Torrence and Webster’98)
(4) Biennial oscillation (Clarke and van Gorden’99; Yu’05)
23
Persistence Barrier
(Auto-lag correlation)
Initial Month
Target Month
Persistence barrier delayed in CFS long-lead forecasts
24 CFS vs OBS
NINO3.4 SST
NINO3.4 d20
WEP taux
Seasonal change:
composite anomaly
Target Month
Initial Month
Peak and decay time delayed in CFS long-lead forecasts
25 CFS vs OBS
NINO3.4 SST
NINO3.4 d20
WEP taux
Persistence Barrier vs Anomaly (obs)
Initial Month
Target Month
consistent
NINO3.4 SST
NINO3.4 d20
WEP taux
26
Persistence Barrier vs Anomaly (cfs)
Initial Month
Target Month
consistent
NINO3.4 SST
NINO3.4 d20
WEP taux
27
Seasonal change:
variance
Initial Month
Target Month
28 CFS vs OBS
Persistence Barrier vs Variance (obs)
Target Month
Initial Month
NINO3.4 SST
NINO3.4 d20
WEP taux
29
Persistence Barrier vs Variance (cfs)
Target Month
Relation is good for SST, but not for d20 and wind
NINO3.4 SST
NINO3.4 d20
WEP taux
Initial Month
30
Persistence Barrier vs S/N (cfs)
Target Month
Initial Month
Not explained by seasonal change in S/N ratio
NINO3.4 SST
NINO3.4 d20
WEP taux
31
Persistence Barrier vs Spread (cfs)
Target Month
Initial Month
32 Auto-lag cor starts
to drop before the largest spread
WEP taux NINO3.4 d20 NINO3.4 SST
Plausible Reasons for spring prediction barrier
Noise cannot explain the spring prediction barrier
What are the plausible reasons?
Bias in atmospheric model wind response
33
Regression wrt Dec NINO3.4 SST
SSTA Observations txA
CFS ensm July
CFS ensm December
34
Anomaly (2S-2N):
regression wrt DEC NINO3.4 SST
CFS ensm December
Observations
SSTA txA*40 d20A/20
35
Anomaly (2S-2N):
regression wrt DEC NINO3.4 SST
CFS ensm July
Observations
SSTA txA*40 d20A/20
36
Area-mean Anomaly NINO3.4 SST
NINO3.4 d20
NINO3.4 taux WEP taux
37
Prediction skill:
El Nino vs La Nina
38
Prediction skill (RMSE): El Nino vs La Nina
39
Target Month
Initial Month
Prediction skill:
Dependence on current state
Large RMSE composite
Small RMSE composite
40
Prediction skill:
Dependence on initial state
Large RMSE composite
Small RMSE composite
41
Prediction skill: Relation to noise
42 rmsIA
Summary
1. The spring prediction barrier in EEP SST is preceded by a boreal winter prediction barrier in the WEP zonal
wind stress
2. The seasonal change in noise cannot entirely explain the spring prediction barrier
3. The prediction barriers could be related to the erroneous atmospheric model wind response to SST anomalies
4. The prediction skill is better for El Nino than for La Nina
43
Thanks!
Wu, R., B. P. Kirtman, and H. van den Dool, 2009: An
analysis of ENSO prediction skill in the CFS retrospective forecasts. J. Climate, 22, 1801-1818.
Wu, R., and B. P. Kirtman, 2009: Variability of El Niño- Southern Oscillation-related noise in the equatorial Pacific Ocean. J. Geophys. Res., 114, D23106,
doi:10.1029/2009JD012456.
44