The Telegraph
Since 1st March, 1999
Email This Page

Only time will tell whether the new forecasting model for the Indian weather will lead to better predictions or more muddle

Predictive models are never perfect. And contrary to what people sometimes tend to think, prediction is stochastic, not deterministic — models predict probabilities of various outcomes. They do not predict any specific outcome with certainty. Having said this, models are as good (or as bad) as their specifications, data used and computational power. In the aggregate, Indian agriculture is much less sensitive to monsoon vagaries than was the case before the Green Revolution. Irrigation in Punjab, Haryana and western Uttar Pradesh has ensured this. However, there are parts of India that are not irrigated, and therefore still monsoon-dependent. Arguably, agricultural policy, including research and development, has not so far catered to these areas. Also, spatial and temporal distribution of rains is more important than aggregate rainfall. The severity of the 2002 drought is still debated. But the Indian Meteorological Department did not cover itself with glory last year, having predicted a normal monsoon.

Consequently, the 14-year-old weather forecasting model has now been junked for a model that incorporates variables like wind, temperature, pressure, El Nino, snow cover and historical data. Second, the forecast date has been preponed from May 25 to April 16. Thereafter, when additional data become available in June, there will be a new set of predictions in July. Third, predictions will have five categories — drought (less than 90 per cent of long-term average), below-normal (90-97 per cent), near-normal (98-102 per cent), above-normal (103-110 per cent) and excess (more than 110 per cent).

The IMD now predicts that 2003 will have a below-normal south-west monsoon of 96 per cent of the long-term average. This projection has a probability of 39 per cent. However, there is 21 per cent probability of drought, 14 per cent probability of near-normal, 23 per cent probability of above-normal and 3 per cent probability of excess. Assuming these projections are correct, rains lean towards the lower rather than the upper end. This does not augur well for a great agricultural recovery. The timing of the monsoon will only be forecast in end-May and the breakup into three regions in July. Time will tell whether last year’s flak has led to a better model or more muddle. For the record, the IMD may have been well off the mark in 2002. But deviations between forecasts and actuals have been significant in every other year also. While prediction is inherently stochastic, consistent error margins of more than 5 per cent do not enhance the IMD’s credibility. This is especially galling because weather forecasts emanating from California were far more accurate in predicting Indian rains in 2002. In 2001, the report of the national statistical commission removed whatever faith citizens had in the quality of Indian statistics. The commission’s “garbage-in, garbage-out” thrust should have also covered weather forecasts. The IMD is entitled to argue that it has learnt from its mistakes and thus evolved a refined and better model. Whether that is true or not only time will tell. Otherwise, there may well be a case for out-sourcing weather forecasts.

Email This Page