Nation’s top Met team admits: our monsoon forecasts are way off

‘‘The (monsoon) predictions have not improved since 1932. I was shocked by what we found,’’ Sulochana Gadgil, Professor, Centre for Atmospheric and Oceanic Sciences at the Indian Institute of Science (IISc), told The Indian Express from Bangalore.

‘‘It is amazing,’’ emphasised Gadgil, on a recent analysis of India’s monsoon predictions with M Rajeevan, Director, National Climate Centre, at Pune’s Indian Meteorological Department (IMD).

After a week of rain shortfall of 49 per cent, an urgency for corrective changes in how India forecasts the monsoon is evident at this office in Pune where 11 scientists use a statistical model to prepare the annual monsoon forecast released in the capital in April.

‘‘The magnitude of errors has not been small,’’ Rajeevan told The Indian Express. ‘‘The performance of our forecasts has not been impressive. An improvement needs a focussed national effort.’’

Gadgil, Ravi Nanjundiah of the IISc and Rajeevan’s critical appraisal of monsoon predictions — published in Current Science in May — makes the point bluntly:

‘‘Our analysis of the predictions generated by the operational models at the IMD from 1932 onwards suggests that the forecast skill has not improved over seven decades despite continued changes in the operational models. Clearly, new approaches need to be explored with empirical models.’’

That’s why in a little room on IMD’s first floor, Rajeevan’s team is experimenting with an alternate forecast model used globally and borrowed last year from the Experimental Climate Prediction Centre at San Diego, USA.

‘‘The present statistical model has a lot of limitations,’’ said Rajeevan. ‘‘Ultimately, we will have to switch to a dynamical model that can predict rainfall variation over every 200 km every month unlike the all-India index possible at present.’’

Gadgil pointed out that research could take five to 10 years. ‘‘We need to simulate year-to-year variations of the Indian monsoon to get a workable model and right now, none of the models in the world can do that. We just have to go ahead with the research.’’

Warning against a ‘‘quick-fix solution’’, the paper suggested the dynamical model as an alternate option provided it is improved to simulate monsoons realistically.

To do that seriously, the Pune-based team, which receives experimental data from centres nationwide, needs:

A supercomputer

The present high-performance computer for dynamical model experiments is worth Rs 1 crore but takes 10 days for a seasonal forecast. ‘‘Our computing capabilities are subdued for dynamical modelling. The programme is almost 20,000 lines log and needs a supercomputer,’’ said Rajeevan, pointing out that workstations are adequate only for statistical models.

A research group

‘‘This is very complex physics,’’ said Rajeevan. ‘‘We need a skilled research group to perfect the dynamical model. Right now our small operational group focuses on the annual monsoon forecast.’’

The Current Science paper stressed the gap in present forecasts: It is far more important to generate accurate predictions of droughts/excess rainfall seasons than of fluctuations within 10 per cent of the average. It is important to predict not only the total rainfall for the season but also the variation within the season…that has a direct impact on the livelihood of rural populations.’’