Is Monsoon Predictability through Statistical Methods decreasing?

The predictability of the Indian Monsoon Rainfall (IMR) through the statistical methods depends on the stability of the relationship between the predictors and the IMR. Several studies have shown secular variations in these relationships.

Statistical analysis (through Cramer's t-statistics) for the period 1871-1999 reveals that the IMR exhibits decadal variability. Whereas the periods 1880-1895 and 1930-1963 are characterised by above normal IMR, the periods 1895-1930 and 1963-1990 have witnessed below normal IMR. The IMR has entered into an above normal epoch around 1990. Analysis further reveals that the impact of El Nino is more severe during the below normal epochs than the above normal epochs. Thus the impact of El Nino is modulated by the decadal variability. This gives a convincing explanation for the failure of the El Nino-IMR relationship after the 1990s. Inspite of the severe 1997 El Nino of the century the IMR was 102% of the long term average.

Sliding correlations of the IMR with other regional and global predictors reveal that the relationships are stronger during the below normal epochs than the above normal epochs. This suggests that the predictability of IMR through statistical methods is more during the below normal epochs. This further implies that the predictability of the IMR after 1990s through statistical methods is decreasing. This is supported by the fact that most statistical models had forecasted monsoon rainfall for 1994 (1999) to be on the negative (positive) side, but in reality the rainfall was on the positive (negative) side.

The variability of the IMR not only depends on the surface boundary forcings, but also on its own internal dynamics i.e. intraseasonal low frequency oscillations. Hence it appears that the IMR has entered into a phase where dynamical prediction models may show better skills in prediction than the statistical models.