MJO forecast method
A statistical technique is used to forecast the MJO. A brief
technical description is given here. Full details can be found in Love BS, Matthews AJ, Janacek GJ,
Real-time extraction of the Madden-Julian Oscillation using empirical mode decomposition and statistical forecasting with a VARMA model.
J. Climate, 21, 5318-5335.
The data set used is the NOAA AVHRR gridded outgoing longwave
radiation (OLR). Pentad (5-day) mean maps of OLR anomalies are
calculated by subtracting the annual cycle at each grid point. Then,
the MJO signal at each grid point is isolated using empirical mode
decomposition (EMD). This acts as a time filter that can be applied
in real time with minimal end effects.
The EMD-filtered OLR maps up to the current pentad are then
projected onto the leading two empirical orthogonal functions (EOFs)
of tropical OLR. The resulting two principal component (PC) time
series, PC1 and PC2, describe the state of the MJO at each time.
Future values of PC1 and PC2 are then predicted using a vector
autoregressive moving average (VARMA) model. The parameters of the
VARMA model were determined by a maximum likelihood technique using a
training data set, from 1979 to 1996.
The forecast OLR anomaly maps are produced by multiplying
seasonally varying regression maps of OLR by the predicted values of
PC1 and PC2.