NCEP MJO Forecast Methodology
- We developed our own codes here but followed very closely WH2004. Our correlation with WH2004 over the 1979-2001 time period is 0.995 for both RMM1 and RMM2 so we are close.
- Remove the first 3 harmonics of the seasonal cycle to get the basic anomaly.
- Remove ENSO by using the SST1 index calculated by the BMRC (download from the web and extend the most recent month SST1 value to current month).
- Remove a 120 day running mean using the last 120 days of the analysis.
- Project onto EOF patterns (U850, U200 and observed OLR) obtained from historical data 1979-2001.
- Realtime U200 and U850 are obtained from NCEP GDAS and replaced by the NCEP reanalysis after 5 days so that there may be subtle changes to the diagram each day. Observed OLR is used.
Please note: We are working on better ways to do this and would like to talk to Matt, Harry, and Nick about this. We have started to calculate SST1 in realtime here using OI SST but this is just in the very early stages.
For the forecast:
- Calculated from the NCEP Global Ensemble Forecast System (GEFS).
- The ensemble mean is for 20 members.
- The method for the forecast is the same as for the observations except with some caveats below.
Unfortunately we do not have a recent consistent climatology for the GFS as the model is always changing. There is an activity within EMC at NCEP that is working on bias correction of the GFS based on the reanalysis and they have done this for U850 and U200 so we use bias corrected fields from the GFS.
Accordingly we then use the reanalysis climatology for the GFS forecast with the assumption that much of the bias has been removed from these two fields. SST1 is used similarly, and the 120 day mean is based on observations (the last 120 days from the reanalysis).