Madden-Julian Oscillation Real Time Forecasts
The Madden-Julian Oscillation (MJO) is the most important mode of tropical intraseasonal variability with potentially important influences on extra-tropical medium-to-extended range weather forecasting, monsoon activity in the Asian-Australian and Americas regions and variability in the equatorial Pacific thermocline structure via its close association with westerly wind bursts.
This project shows our research on Madden-Julian Oscillation forecasts. The current statistical forecasts are based on the model described in Jones, C., L. M. V. Carvalho, W. Higgins, D. Waliser, J-K. Schemm, 2004: A Statistical Forecast Model of Tropical Intraseasonal Convective Anomalies. J. Climate, 17, 2078-2095.
The model uses principal components (PC’s) of empirical orthogonal function analysis of 20-90 days anomalies of outgoing longwave radiation (OLR). Forecast models are developed for each lead time from 1 to 10 pentads and for winter and summer seasons separately. The forecast models use a combination of the 5 most recent pentad values of the first 5 PC’s of OLR to predict future values of a given PC. The spatial structures are obtained by reconstructing the fields of OLR using the forecasts of PC's and the associated EOF’s.
Check the log for the latest run "Most Recent Available Forecast"
Current Observations and Forecasts [it shows observed filtered OLR (left) and forecasts of OLR Anomalies (right); time oriented downward].
The links below show the last three previous forecasts. It can be used to evaluate the tendency of model forecasts and get a sense of current skill
Observations and Previous Forecasts (Initialized 1 pentad before current pentad)
Observations and Previous Forecasts (Initialized 2 pentads before current pentad)
Observations and Previous Forecasts (Initialized 3 pentads before current pentad)
The link below shows forecast verification for the model run initialized 4 pentads before the latest available pentad of OLR data (inclusive). The verification is presented in terms of pattern (spatial) correlation between forecast and observation. One can use this product to monitor the forecast skill near the current date. Our experience indicates that statistical forecasts of the MJO will have high skill when an MJO event has already started, i.e. enhanced convective activity is fully developed over the tropical Indian Ocean and is propagating eastward towards Indonesia. This limitation is perhaps true for any statistical forecast model of the MJO. More information can be seen in this presentation (click). The statistical forecasts can be useful to monitor and forecast the impacts of the MJO downstream from the region of convective anomalies. For instance, several studies have shown that the MJO modulation on weather variability in the Americas is observed when enhanced convective anomalies associated with the MJO have propagated over the tropical western Pacific Ocean.
NOTE: since filtering is done in real time, filtered OLR anomalies may not match exactly from one run to another including in the verification process.
Monitoring Precipitation and Extremes
GPCP Data 5-Day averages (pentads) . Extremes are relative to seasonal percentiles. Additional information can be viewed at the GEM website maintained by Dr. Leila Carvalho
Other Sites of Interest
Experimental MJO Forecasts (CDC Host) (opens in a new window)
For more information about our research, please email Dr. Charles Jones. The research described in this web site was partially funded by the National Science Foundation and the NOAA Office of Global Programs. Their support is greatly appreciated.
Send mail to
with questions or comments about this web site.