An Improved Multivariate ENSO Index (MEI) - A Progress Report Klaus Wolter and Michael Timlin NOAA-ESRL, Physical Science Division, and CIRES Climate Diagnostics.

An Improved Multivariate ENSO Index (MEI)- A Progress Report

Klaus Wolter and Michael TimlinNOAA-ESRL, Physical Science Division, and CIRES Climate

Diagnostics Center; University of Colorado at [email protected]

http://www.mei.noaa.gov/

• We already have Niño 3.4 SST to monitor ENSO - Why bother with something more complicated?

• Variations on the MEI theme - How robust is it?

• How to pick the “best” ENSO index & Where to go from here?

CPASW, Chapel Hill, 6mar08

MEI Background

• The Multivariate ENSO Index (MEI) is an outgrowth of my dissertation work two decades ago. It was inspired by the original definition of the ‘Southern Oscillation’ by Walker and Bliss (1932) that includes much more than the familiar sea level pressure seesaw between the western and eastern Pacific, but also temperature and precipitation fields, stratified by season, an early multivariate approach to this phenomenon.

• The MEI was engineered in its present form with support from Michael Timlin who is also helping with the MEI revisions. It is based on surface marine data (COADS) - filtered through spatial cluster analysis and based on six different observational fields: sea level pressure (P), zonal- and meridional wind component (U, V), sea surface temperatures (S), near-surface air temperatures (A), and total cloudiness (C). Separately analyzed for twelve sliding bi-monthly seasons, the MEI is the first Principal Component (EOF) of all six observational variables analyzed jointly for the tropical Pacific basin (normalized variance for each field).

• The MEI was put on the internet in mid-1997. I update it monthly and discuss its recent evolution, giving input to CPC ENSO discussions as well.

MEI Loading Maps

‘Loadings’ refer to the correlation of local time series with the MEI time series. Jan-Feb is slightly past the peak explained variance season of the original cluster-filtered version of the MEI, but reflects the peak of newer MEI versions.

MEI Time Series

• Long-term trends and shifts (~1976) are NOT filtered out in its present form, while the sliding bimonthly averaging removes most of the intra-seasonal ‘noise’.

• Month-to-month fluctuations in the MEI can be partially explained by seasonal changes in the clustering filter, especially in the ‘weaker’ fields (V and C). Using near-real time radio-transmitted ship (and buoy-) data adds noise to the monthly updates. In my estimate this translates into error bars around 0.1 sigma for the last decade of web updates.

MEI - Why Bother?

• ENSO is a multivariate, coupled ocean-atmosphere phenomenon, so an index that reflects these characteristics is the ‘right thing to do’;

• There is less vulnerability to errors in single variable fields (or point data);

• The MEI allows for spatial variations in the seasonal cycle as well as intensity with standardized anomalies (more impact on wind fields and cloudiness than SST);

• MEI is the ONLY ENSO index that includes at least the fundamental tropical ‘atmospheric bridges’ - if these fundamental teleconnections don’t work, how can you expect to see reliable associations with the extratropics?

• Despite all this, the MEI correlates quite highly with all other ENSO indices, so there is no question that they all monitor the same phenomenon - in boreal winter.

ENSO Index Ranks 50-06: MEI, Niño 3.4, SOI, Niño 3

Proposed/Currently Tested Changes to MEI

√ Switch from clustered to gridded data input;

√ Replace cloudiness with Outgoing Longwave Radiation (OLR);

√ Explore switch from COADS to NOAA-NCAR Reanalysis data;

√ Explore impact of change in base period;

√ Introduce subsurface ocean data (explore different data sets);

* Explore domain changes (currently mostly tropical Pacific)

* Explore different ‘flavors’ of ENSO (ENSO-subregional indices?)

* Support ENSO ‘impacts’ research (sensitivity to ENSO Index)

* Filter out climate change (‘hinge-fit’?)

MEI Loading Maps (1950-2005; no clusters)

Match of old (clustered) MEI with new MEI: 1950-2005

Match between clustered (old) and unclustered (new) MEI is better than between MEI and other ENSO Indices (correlation is highest with Niño 3.4). Clustered MEI data indicates stronger La Niña conditions in the 1950s than newer version, possibly due to lack of pre-IGY data.

1996-2005 Clustered vs. New PUVSAC 5005 MEI

Noisy real-time data?

MEI Loading Maps (1979-2005; no clusters)

Cloudiness has lowest explained variance/least influence on MEI

MEI Loading Maps, OLR (1979-2005; no clusters)

OLR has highest explained variance!

Match of PUVSAC MEI vs. OLR MEI - 7905

Match between original mix of unclustered variables (PUVSAC) & new mix of variables (PUVSA&OLR) MEI is very good, despite the big change in explained variance going from cloudiness to OLR.

Match between 1950-05 and 1979-05 versions is also very good (not shown here).

1996-2005 OLR- vs. Cloudiness-based MEI

MEI Loading Maps (1979-2007; no clusters; OLR; 300m ocean heat content)

Swapping subsurface heat content for air temperature works!

1996-2007 Four different MEI’s (r≥0.96)

How can we pick the best index for the job?• Philosophically, the MEI has intuitive appeal, since it at least tries to

address the multivariate ‘nature of the beast’; Niño 3.4 SST was only introduced a decade ago (Barnston et al., 1997), but has become the standard ENSO index in the U.S.;

How can we pick the best index for the job?• Philosophically, the MEI has intuitive appeal, since it at least tries to

address the multivariate ‘nature of the beast’; Niño 3.4 SST was only introduced a decade ago (Barnston et al., 1997), but has become the standard ENSO index in the U.S.;

• From an impact perspective, one can compare the explained variance of, say, global temperature and precipitation fields based on various ENSO indices to decide which index is best (MEI beats Niño 3.4 and SOI in Western U.S.);

How can we pick the best index for the job?• Philosophically, the MEI has intuitive appeal, since it at least tries to

address the multivariate ‘nature of the beast’; Niño 3.4 SST was only introduced a decade ago (Barnston et al., 1997), but has become the standard ENSO index in the U.S.;

• From an impact perspective, one can compare the explained variance of, say, global temperature and precipitation fields based on various ENSO indices to decide which index is best (MEI beats Niño 3.4 and SOI in Western U.S.);

• From an operational perspective, there is a need for an ENSO index that is most reliable (lowest False Alarm Rate, etc.) - initial results from Australia favor MEI over Niño 3.4 and SOI;

How can we pick the best index for the job?• Philosophically, the MEI has intuitive appeal, since it at least tries to

address the multivariate ‘nature of the beast’; Niño 3.4 SST was only introduced a decade ago (Barnston et al., 1997), but has become the standard ENSO index in the U.S.;

• From an impact perspective, one can compare the explained variance of, say, global temperature and precipitation fields based on various ENSO indices to decide which index is best (MEI beats Niño 3.4 and SOI in Western U.S.);

• From an operational perspective, there is a need for an ENSO index that is most reliable (lowest False Alarm Rate, etc.) - initial results from Australia favor MEI over Niño 3.4 and SOI;

• From a modeling perspective, a whole generation of coupled ocean-atmosphere models has been ‘trained’ to generate Niño 3.4 SST forecasts, but there has been a plateau in skill lately. Could it be that modeling and predicting the key features of the MEI is a more rewarding target?!

Proposed Changes to MEI Web Site (new: mei.noaa.gov)Current web site includes historic MEI time series values, their seasonal ranks, as well as comparisons with well-established ENSO events.

• How can this be enhanced to make the MEI a better decision support tool for global ‘stakeholders’?

• What kind of tutorials should be developed?

• Should there be a discussion and assessment of recent climate anomalies in the context of recent ENSO conditions?

An historical assessment of global and regional impacts (at least seasonal temperature and precipitation) will be performed in reference to MEI-based assessments of historic ENSO event strength. This will be coordinated with the WMO Expert Team on El Niño.

• Should there be a parallel effort to predict the MEI?

• Would there be interest in regional MEI’s that convey ‘flavor’ of ENSO, say, over the Indian Ocean vs. Western-Central Pacific vs, Eastern Pacific

Stay tuned & Feedback welcome!

ENSO Index Ranks 50-06: MEI, Niño 3.4, SOI, Niño 3

Using four-month seasons; color terciles (1-19=La Niña, 39-57=El Niño), underlined quintiles (1-11/47-57), and deciles (1-6/52-57) could serve as a rough classification for ‘weak’, ‘moderate’, and ‘strong’ events.