A Framework for Assessing Operational Model MJO Forecasts US CLIVAR MJO Working Group Forecast Team** Climate Diagnostics and Prediction Workshop October 26-30, 2009 Monterey, CA ** Jon Gottschalck: NOAA / Climate Prediction Center Matt Wheeler / Harry Hendon: Australia Bureau of Meteorology Klaus Weickmann: NOAA / Earth System Research Laboratory Frederic Vitart: European Centre for Medium Range Weather Forecasts Hai Lin: Environment Canada Nick Savage: UK Met Office Duane Waliser: NASA / Jet Propulsion Laboratory Ken Sperber: DOE / Lawrence Livermore National Laboratory
Outline Project motivation and background Current participation MJO identification diagnostic and uniform display Operational applications Project next steps
Project Background The MJO is the dominant mode of tropical intraseasonal variability Eastward propagation of large-scale anomalous rainfall and wind Period of approximately 30-60 days Can lead to numerous impacts Realtime forecasts of the MJO are increasingly being recognized for their potential to improve extended range weather forecasting MJO prediction studies using operational realtime dynamical models have increased in recent years but have used varying methodologies, datasets, and validation metrics
Project Background The US CLIVAR MJO Working Group (MJOWG) designated a team to adopt a uniform diagnostic for MJO identification and skill metrics Standard measures allow for consistent evaluation and display of MJO forecasts from multiple sources over time Invitation letter from the MJOWG and Working Group on Numerical Experimentation (WGNE) was distributed to operational centers around the world to introduce the project and request participation
Center Participation US NCEP ECMWF United Kingdom Brazil US NRL India Taiwan Australia Japan Canada CMC
Center Data Specifics Multiple contributions for several centers High-resolution operational run data as well as data from ensemble prediction systems Varying forecast duration
MJO Diagnostic Details Several MJO extraction methodologies were considered: (1) Combined EOF analysis (OLR, u850, u200) Wheeler and Hendon (2004) (2) Fourier filtering of OLR anomalies for zonal wavenumbers and frequencies consistent with the MJO Wheeler and Kiladis (1999) Wheeler and Weickmann (2001) (3) EOF analysis of 20-90 day bandpassed OLR anomalies Jones et al. (2004)
MJO Diagnostic Details Slight variant of the WH2004 MJO index was chosen: (1) Widespread acceptance as a relatively well-characterized measure of the MJO and its evolution (2) Well suited for real-time application as it requires only spatial, no temporal, filtering an important consideration when applied to real-time operational model data (3) The method is relatively straightforward to adopt
MJO Diagnostic Details Weak MJO Strong MJO
MJO Diagnostic Details Application to operational model output: Centers send total OLR, u850, u200 data to CPC ftp site in realtime Model forecast anomalies based on observational climatological data from NCEP Reanalyses and NOAA satellite OLR The most recent 120 day mean of model analysis/forecast anomaly data is subtracted to remove low-frequency variability Forecast data are projected onto observed EOFs currently Resulting RMM1 and RMM2 values displayed in phase space
MJO Forecast Examples NCPE CANM UKME Differences in: (1) Ensemble spread (2) Propagation speed (3) Amplitude ECMF BOME
CPC Project Webpage http://www.cpc.ncep.noaa.gov/products/precip/cwlink/mjo/clivar/clivar_wh.shtml
Some Applications Other users include: Interests in the private sector and other government agencies
MJO Forecast Verification Forecast skill metrics: (1) Bivariate correlation (COR) (2) Root-mean-square error (RMSE) between the observed ( multi-model analysis ) and forecasted RMM indices
Current Issues and Next Steps (1) Considerable bias at times between model data and obs Function of MJO strength and phase, model and seasonal cycle Evaluation is planned to minimize and quantify the biases Hindcast datasets from operational models needed for calibration (2) Development and display of realtime verification database Statistical model as a benchmark Stratify by seasonal cycle, MJO strength and phase (3) Eventual development of a multi-model ensemble Equally-weighted average of the model ensemble mean forecasts Historical skill based weighted average
Thank You Comments and Questions Jon.Gottschalck@noaa.gov
MJO Diagnostic Details Daily time series projection coefficients of a pair of EOFs derived from a combined EOF analysis of OLR, 850-hPa, and 200-hPa zonal wind latitudinally averaged from 15S to 15N Seasonal cycle and low frequency variability are removed (the mean of the most recent 120 days) Resulting time series projection coefficients are termed Realtime Multivariate MJO index 1 and 2 (RMM1 and RMM2) EOF 1 Structure EOF 2 Structure 1.5E-01 1.0E-01 OLR U850 U200 1.5E-01 1.0E-01 OLR U850 U200 Eigenvector 5.0E-02 0.0E+00 0 60 120 180 240 300 360 Eigenvector 5.0E-02 0.0E+00 0 60 120 180 240 300 360-5.0E-02-5.0E-02-1.0E-01-1.0E-01 Longitude (degrees) Longitude (degrees)
(a) Operational MJO Prediction -- Example (b) (c)