New Delhi: India’s top weather man R.C. Bhatia is a few weeks away from making the most important weather call that the Indian Meteorological Department (IMD) makes every year. If the chief of IMD gets it right, no one will particularly notice, but if he gets it wrong, everyone, from farmers and politicians to the makers of consumer goods and bankers, will. The prediction IMD and Bhatia will make has to do with monsoon, the annual rains that still decide the fate of India’s agricultural and rural economy, and, by extension, the country’s economy as a whole.
A good monsoon could ensure a plentiful harvest and ease concerns regarding the short supply of agricultural commodities—one reason for high inflation—extending into next year. Which explains why everyone is looking to Bhatia for IMD’s take on the monsoon. In the past four years, ever since it moved to two new weather-predicting models, the agency has got it wrong every other year.
The meteorological department (or Met department as it is popularly known) got it right in 2003 and 2005 and wrong in 2004 and 2006. Bhatia, a veteran weather man, isn’t worried by that 50% hit-rate. “We are still experimenting,” said Bhatia. “Predicting the monsoon, especially the Indian monsoon, is not an easy job,” he added.
The reason why the department is “still experimenting” dates back to 2002, when the IMD, now 132 years old, got it horribly wrong. That year, the department predicted a normal monsoon and India ended up experiencing a major drought.
IMD realized that the problem was with the 16-parameter model it was using. “Our analysis said 10 of the 16 parameters showed weakening correlation, so we did away with them,” said National Climate Centre (Pune) director Madhavan Rajeevan, who authored a research paper on the subject around the same time.
A weakening correlation means the influence of a particular parameter on the outcome (in this case, the monsoon) isn’t as significant as it once was. For instance, the temperature in North India was an important parameter in the 16-parameter model.
It was only after the 2002 debacle that the Met department realized that its influence on the monsoon had been waning for some time. “There is no greater foe (to prediction) than a weakening parameter,” said a scientist who studies weather prediction models and did not wish to be identified.
The department replaced the 16-parameter model and its single prediction based on it, with two models and two predictions. The first prediction, based on an eight-parameter model is made in the second half of April. And the second prediction is made in the second half of June, based on a 10-parameter model. India receives 80% of its rainfall between June and September.
Monsoon plays an important role in determining the sowing patterns for what is called the kharif season where the primary crop is rice.
The two new models junked the 10 parameters that showed a weakening correlation and replaced them with four new ones; data for two of these becomes available by the end of March and is used in the April prediction; and data for the other two becomes available by June.
Around the same time, the Met department redefined normal. Previously, explained Rajeevan, “All rainfall that was 10% less or more than the long period average was considered normal.” The long period average (LPA) is the mean of measured rainfall between 1958 and 2005. LPA is 89cm, and the 10% tolerance meant that anything between 80cm and 98cm was considered normal. “That’s a wide range and it’s known that such a variation in rainfall can have a marked consequence for agriculture,” said Rajeevan.
The new models used by IMD use a mathematical technique called the linear discriminant analysis to further divide the 20% range (10% less or 10% more) into five categories: drought, below normal, near normal, above normal, and excess. Now, normal just means 98-102% of LPA.
The new definition of normal is one reason why the Met department’s predictions over the past four years fared badly, according to a senior meteorologist with IMD who did not wish to be identified. “Had the modified definition not been effected, IMD could have gotten away with the claim that the rainfall was normal in the past four years,” he added.
The Met department might have changed the parameters and the way it looks at them, but its weather predictions could still go faulty because of the very nature of the models, said an expert. “Unfortunately, most of our models are statistical,” said S.K. Dash, professor, Centre for Atmospheric Sciences, Indian Institute of Technology, Delhi. “Internationally, it is dynamic models that are used,” he added.
A statistical model uses historical data, often collected over a long period of time, sometimes as much as 40 years. A dynamic model measures a clutch of variable related to existing weather conditions—some could relate to oceans, others to land, and still others to atmosphere—and uses these to predict the weather. Dynamic models are usually more accurate in predicting weather than statistical ones. “The catch is, they require precise and detailed measurement,” said Dash.
Bhatia is aware of the importance of dynamic models. “We are setting up automatic weather stations, and have installed Doppler radar at certain locations,” he said. These machines installed by IMD have the ability to measure wind speeds and atmospheric pressure at significant heights. Both measures, according to Dash, are crucial for a dynamic model of predicting monsoon. It could take time though, said Rajeevan, who added that the shift to dynamic models would require more experiments.
For those not interested in whether the Met department uses a dynamic or statistical model, but keen to know whether India will have a bountiful monsoon this year, Dash has just the measure.
“Historically, it has been seen that whenever there’s a La Nina, India has never had a drought,” he said, referring to the phenomenon which describes unusually low temperatures in the central Pacific ocean (or cold currents).
Data from the World Meteorological Organization points to an emerging La Nina this year.