The government’s admission that the accuracy of the India Meteorological Department (IMD) in its initial prediction of monsoon rainfall had been merely 50 per cent during 2007-11 has brought back into focus the models adopted by the official weather forecaster.
Ashwani Kumar, the minister of state for science, technology and earth sciences, had made the admission in Parliament yesterday. Experts said if so, there was something seriously wrong with the forecasts. If actual rainfall had been, for half the time, beyond the error margin given in the forecast, it meant the methods adopted were not working, they said.
IMD had adopted a statistical model for weather forecasts over almost 100 years but the severe droughts of 2002 and 2004 made it change that. At present, it adopts an improved version of the old mode, based on the ensemble multiple linear regression (EMR) and projection pursuit regression (PPR) techniques. IMD had said when the new forecast model was applied regressively on monsoons between 1981 and 2004, it showed far better accuracy.
Yet, since 2007, when IMD first began using the improved model, the actual rainfall beat its Long Range Forecast (LRF) thrice. The period includes the worst drought, of 2009. In that year, IMD had first said full season rainfall would be 96 per cent of the Long Period Average (LPA); actual rains were only 77 per cent.
“It is well known that IMD’s monthly monsoon forecast is bogus. Now, if the LRF, too, is not accurate in half of the years, then it means something is seriously wrong with the method,” Y K Alagh, the noted economist and academician who was minister in charge of the department between 1996 and 1998, told Business Standard.
Adding: “We seriously need to take note of this because if any model gives such poor results anywhere in the world, it would be immediately changed.”
Also Read
A group of Indian businessmen — rain impacts overall economic growth in a big way — recently called for changing the model IMD uses to make an LRF of the southwest monsoon.
However, the department has defended its model. "As far as I am concerned, this is by far the best possible model to forecast the southwest monsoon. Yes, it has failed sometimes, but that is possible with every system that you adopt," says LRF Director D S Pai.
Ironically, IMD’s LRF was relatively more accurate when it was using the old model between 1990 and 2000. Data shows actual rain was more or less than 10 per cent of the initial forecast in that time on only four occasions.
This year, too, the met office has faltered on the forecast. In April, it had said monsoon rainfall across the country would be around 99 per cent of the LPA. Till mid-August, rainfall had been 12 per cent deficient.