Monsoon forecasting!!!

The India Meteorological Department (IMD) is once again in the uncomfortable position of finding its prediction for the south-west monsoon going awry. Using a new statistical forecasting methodology, developed in consultation with Indian statisticians and published in an internationally reputed peer-reviewed journal, the department issued its long-range forecast for the monsoon in April, indicating that there could be a five per cent shortfall in countrywide rainfall, with an error bar of five percentage points. The updated forecast it issued at the end of June hiked the predicted deficit in rainfall to seven per cent with an error bar of four percentage points. Happily, the monsoon seems set to end with a surplus, if only a small one of about five per cent more than the long-term average. But that is no reason to abandon the new model and cast about frantically for a fresh way to foretell the monsoon’s outcome. A statistical model must be evaluated over a reasonable period of time and not on the basis of its performance in just one year. Since 2000, the IMD has thrice tweaked and changed the operational model it uses for long-range forecasting of the monsoon, and the new model deserves a fair trial.


In all of this, the primary purpose behind long-range forecasting of the south-west monsoon must be borne in mind. India’s Met Department began such forecasts more than a century ago. The government wanted advance warning of a drought as such a catastrophe could potentially set off a famine in those days. Although the spectre of famine has long been banished, a major drought, as happened in 2002, can still sharply reduce foodgrain production and curb the country’s economic growth. However, the monsoon has ended in a drought only 16 per cent of the time over the last 130-odd years while it has been ‘normal,’ with countrywide rainfall within 10 per cent of the long-term average, in seven years out of 10. The ability to predict a drought well in advance remains a challenge. The IMD’s new model, which might not have fared well this year, is expected to have a good chance of successfully doing so. But if the objective is to distinguish drought years from times when the monsoon is normal or on the excess side, then the long-range forecasts issued to the public need not try to quantify the amount of rain that will occur. Instead, the IMD ought to look at combining the output from its own models with information from other groups so as to provide the best assessment it can of the chances that the monsoon might be normal, bring excess rain, or end in a drought. This year, for instance, if the IMD had indicated that the odds favoured a normal monsoon, its prediction would have been correct.
Source: The Hindu