Long-range prediction of the low-frequency mode in the low-level Indian monsoon circulation with a simple statistical method

The time evolution of the Indian monsoon is closely related to locations of the northward migrating monsoon troughs and ridges which can be well depicted with the 30–60 day filtered 850-mb streamfunction. Thus, long-range forecasts of the large-scale low-level monsoon can be obtained from those of the filtered 850-mb streamfunction. These long-range forecasts were made in this study in terms of the Auto Regressive (AR) Moving-Average process. The historical series of the AR model were constructed with the 30–60 day filtered 850-mb streamfunction [ψ˜ (850 mb)] time series of 4 months. However, the phase of the last low-frequency cycle in the ψ˜ (850 mb) time series can be skewed by the bandpass filtering. To reduce this phase skewness, a simple scheme is introduced. With this phase modification of the filtered 850-mb streamfunction, we performed the pilot forecast experiments of three summers with the AR forecast process. The forecast errors in the positions of the northward propagating monsoon troughs and ridges at Day 20 are generally within the range of 1˜2 days behind the observed, except in some extreme cases.