Sensitivity of MJO Predictability to SST

1
Sensitivity of MJO Predictability to SST
Kathy Pegion Center for Ocean-Land-Atmosphere
Studies Ben Kirtman University of Miami Center
for Ocean-Land-Atmosphere Studies
NOAA 32nd Annual Climate Diagnostics and
Prediction Workshop Tallahassee, FL
2
Motivation
Prediction Skill Studies from DERF experiments
(Chen Alpert 1990, Lau Chang 1992, Hendon et
al. 1999, Jones et al. 2000, Seo et al.
2005) Use atmosphere-only with initial SST
values damped to climatology with a 90-day
e-folding time
Predictability Studies Climatological SST
(Waliser et al. 2003, 2004, Liess et al.
2004) Coupled and uncoupled with daily SST
w/intraseasonal variability removed (Fu et al.
2006) Coupled and uncoupled w/perfect SST
(Pegion and Kirtman 2007)
How sensitive is the predictability of the MJO to
SST?
3
Predictability Experiments

• Ten model intraseasonal events (gt2?) selected
  from a 52-year CFS03 (T62L64) control simulation
• Initialized when MJO-related precip is in Indian
  Ocean
• Perturb atm ICs to generate 9 member ensembles
• 60-day forecast
• Perfect Model - Forecast skill calculated with
  control as truth

4
Initial Conditions
9 Atm Perturbations Generated by running the
model in 1 hour increments resetting the
calendar
TIME (Hours)
- 24
0
1
2
3
4
5
- 1
- 2
- 3
- 4
Coupled Ocn ICs from Control
Uncoupled Prescribed SSTs
5
Predictability Experiments
6
Example Event Control Simulation Unfiltered
Anomalies Averaged 10S-10N
Precipitation (mm/day)
U200 (m/s)
SST (degrees C)
7
Example Event Unfiltered, Ensemble Mean
Precipitation Anomalies Averaged 10S-10N
Clim
Persisted Anomaly
FCST SST
Perfect SST
Forecast Day
mm/day
8
Example Event Predictability Estimates
Correlation Ensemble Mean with Control
Filtered (30-day) Precipitation Indo-Pacific
Region
Coupled
Perfect SST
FCST SST
Persisted Anoms
CLIM
Persistence
A good SST forecast is important to the
predictability of the TISO.
9
SST Sensitivity Experiments (All 10
Events) Unfiltered, Ensemble Mean Precipitation
Anomalies Averaged 10S-10N
Coupled
Perfect
Control
Forecast Day
mm/day
10
SST Sensitivity Experiments (All 10
Events) Unfiltered, Ensemble Mean Precipitation
Anomalies Averaged 10S-10N
Persist Anom
Forecast
Clim
Monthly
Forecast Day
mm/day
11
Predictability Estimates (Ten Events) Correlation
Ensemble Members with Control
Filtered (30-day) Precipitation Indo-Pacific
Region
Coupled
Perfect SST
FCST SST
Persisted Anoms
Monthly
CLIM
Predictability (Days)
Correlation Coefficient
Coupled 18
Perfect 17
Fcst 16
Persist 16
Monthly 14
Clim 9
Forecast Day
12
Predictability Estimates (Ten Events)
Correlation Ensemble Mean with Control
Filtered (30-day) Precipitation Indo-Pacific
Region
Coupled
Perfect SST
FCST SST
Persisted Anoms
Monthly
CLIM
Predictability (Days)
Correlation Coefficient
Coupled 36
Perfect 25
Fcst 23
Persist 20
Monthly 17
Clim 10
Forecast Day
13
Point Correlation of Unfiltered Precipitation
Anomalies Ensemble Members with Control
Week 2
Coupled
Perfect
Forecast
Persist Anom
Monthly
Precipitation Anoms
14
Point Correlation of Unfiltered Precipitation
Anomalies Ensemble Members with Control
Week 3
Coupled
Perfect
Forecast
Persist Anom
Monthly
Precipitation Anoms
15
Point Correlation of Unfiltered Precipitation
Anomalies Ensemble Members with Control
Week 4
Coupled
Perfect
Forecast
Persist Anom
Monthly
Precipitation Anoms
16
Conclusions

• 1. Degrading the quality of the SST degrades
  the skill of the precipitation forecast beyond
  week-1.
• If we hope to make better forecasts of the MJO,
  forecasts for week-2 and beyond should be made
  using ensembles and a coupled model.
• Most of the model skill on intraseasonal
  timescales at lead times beyond week-2 comes from
  regions outside the active/supressed
  precipitation of the MJO and in regions where
  precipitation is small.
• Forecasting MJO-related precipitation beyond
  week-2 is a challenge even under a perfect
  model assumption.

17
Caveats Future Work

• 1. Time filtering - not realistic for operational
  forecasting and not particularly satisfying
• Plan to apply the Wheeler and Hendon real-time
  multivariate MJO index as is being used by the
  Clivar MJO working group
• 2. Model Error - these are perfect model
  predictability experiments. What happens for
  observed MJO events using observed SST?
• Plan to perform hindcast SST sensitivity
  experiments using observed SST

18
(No Transcript)
19
Point Correlation of Unfiltered Precipitation
Anomalies Ensemble Members with Control
Week 1
Coupled
Perfect
Forecast
Persist Anom
Monthly