Development of Precipitation Outlooks for the Development of Precipitation Outlooks for the Global Tropics Keyed to the MJO CycleGlobal Tropics Keyed to the MJO Cycle

Jon Gottschalck1, Qin Zhang1, Michelle L’Heureux1, Peitao Peng1, Kyong-Hwan Seo2, Huug van den Dool1, Wanqui Wang1,Wayne Higgins1, Arun Kumar1

1 NOAA / NWS / NCEP Climate Prediction Center2 Pusan National University, Busan, Korea

Climate Diagnostics and Prediction WorkshopTallahassee, Florida October 22-26, 2007

Motivation, Background, and Goals

Methodology

1. Basis of the outlooks -- MJO MJO filtering MJO Forecast Method Descriptions Consolidation Specifics Initial Findings and Impressions

2. Procedure for Precipitation Outlooks

Potential Interactions and Upcoming Plans

OutlineOutline

MJO substantially modulates tropical rainfall when active

Objective forecast input for CPC weekly MJO and international benefits/hazards assessments

Companion to CPC empirical temperature/precipitation outlooks keyed to the ENSO cycle (Higgins et al. 2004)

Consolidation of MJO forecast methods is first step

Several tools are available for MJO prediction and include both statistical and dynamical approaches

Motivation and BackgroundMotivation and Background

MJO IdentificationMJO Identification Wheeler and Hendon (2004)

Multivariate EOF analysis using OLR, 850 hPa / 200 hPa zonal wind

Data pre-filtering:

1. Seasonal cycle removed

2. ENSO associated variability removed

3. Latest 120 day mean removed

Index is first two PCs (RMM1, RMM2) taken together

Farther from circle the greater the MJO strength

Counterclockwise movement indicates eastward propagation

Data record available extends from 1979-2004

Forecasts are based on pentad averaged data

Forecasts for leads 1 – 6 pentads

Forecasts are of RMM1 and RMM2 (WH2004 PCs 1 and 2)

Idea is to use methods of varying complexity, statistical and dynamical

MJO Forecast Method FrameworkMJO Forecast Method Framework

5 MJO forecast methods currently used:

1. Autoregressive model (ARM) – statistical, (Jones et al. 2004)--Training period 1979-1989, order = 4 pentads, uses information from one PC only

PC(t+1)=∑ CjPC(t–j+1) + εt+1

2. Lagged linear regression (PCL) – statistical, (Jones et al. 2004)--Training period 1979-1989, 5 pentad lags, uses information from both PCs

PC(t+h) = ∑∑ Cij(h)PCi(t–j+1)

3. Empirical Phase Propagation (EPP) – statistical (Seo et al. 2007)--Fixed amplitude, constant 30° per pentad propagation speed

4. Constructed Analogue (ANL) – statistical (Peng and van den Dool, 2005)--Training period 1980-2006 CV

5. Climate Forecast System (CFS) – dynamical (Saha et al. 2006)--Lead dependent climatology, observed EOFs

MJO Forecast Method DescriptionsMJO Forecast Method Descriptions

Forecasts Utilized: 1990-2004, standardized anomalies Consolidation Methods:

1. Equal Weights (CEQ): Weights sum to unity Each method is assigned a weight of 0.20

2. Ridge Regression (CRR):

Weights account for co-linearity between methods Weights are a function of method, time of year, and lead Pooled pentads (3,5,7 pentad tests) Weights based on combining RMM1 and RMM2

Consolidation SpecificsConsolidation Specifics

Results – Ridge Regression WeightsResults – Ridge Regression Weights

PCLGenerally small weights at longer leads during the entire year.

Largest weights at early leads during periods in the boreal spring and late fall.

Results – Ridge Regression WeightsResults – Ridge Regression Weights

ARM

Greatest weights at all leads during late summer and at time at longer leads

Little or no weight given at early leads during much of the year

Results – Ridge Regression WeightsResults – Ridge Regression Weights

EPPHigh weights during September and October at most leads.

Results – Ridge Regression WeightsResults – Ridge Regression Weights

Largest weights of all the methods mainly during the boreal winter and early summer.

ANL

Results – Ridge Regression WeightsResults – Ridge Regression Weights

CFS

Largest weights mainly during late summer and early fall.

Results – Sum of Individual Method WeightsResults – Sum of Individual Method Weights

Periods of little predictability

Periods during February, May, June, August, and October offer the greatest predictability

ALL

Results – Cross-Validated Correlation – RMM1Results – Cross-Validated Correlation – RMM1

Results – Cross-Validated Correlation – RMM2Results – Cross-Validated Correlation – RMM2

Precipitation Outlooks – BackgroundPrecipitation Outlooks – Background

Methodology is similar to Higgins et al. (2004) empirical prediction of seasonal temperature and precipitation keyed to the ENSO cycle

Precipitation Outlooks – MethodologyPrecipitation Outlooks – Methodology

Empirical prediction of MJO associated pentad precipitation Consolidated MJO index to determine MJO phase so precipitation keyed to the MJO cycle

Contour intervals are differences from 33%

11.8

Precipitation Probabilities Keyed to the MJO CyclePrecipitation Probabilities Keyed to the MJO Cycle

Pentad CPC Merged Analysis of Precipitation (CMAP) --1979-2006, 2.5x2.5

Determined threshold limits for upper, middle, and lower terciles--Gamma distribution--Each grid point--Extended winter/summer seasons, 3-month running window

Identified MJO events (WH2004) in the historical record

Combining CMAP data and historical MJO information we can calculate probabilities of precipitation by MJO phase for upper, lower, and middle categories

Results – Consolidation Example in Phase SpaceResults – Consolidation Example in Phase Space

Contour intervals are differences from 33%

11.8

10.7

9.6

11.9

Closing CommentsClosing Comments

Further investigate and improve the stability of weights--Stratifying by season, additional “pooling” tests, etc.

Procedure can leverage work being conducted as part of the US CLIVAR MJO working group

Applying WH2004 methodology to operational models Current participating centers: NCEP, ECMWF, UKMET, CMC, BMRC Other dynamical model input may aid the consolidated MJO index forecast

Proceed with the development of precipitation outlooks if warranted

Objective input into international hazard assessments

Thank You. Comments/Suggestions/Questions?Thank You. Comments/Suggestions/Questions?