Following the India Meteorological Department’s forecast of an ‘above normal’ monsoon of 106 per cent of the Long Period Average (LPA) on Tuesday, The Hindu decided to find out whether these forecasts have worked in the past. Answer (based on 10 years of forecast data): not very accurately. Still, as the IMD puts it, they convey useful information and forecasting itself will only get better with better tools. Here’s a Q&A on the things to know about monsoon forecasting:
To start with, what is LPA? Long Period Average (or LPA) is defined as the average of the rainfall received during a fifty year period between 1951 and 2000, which comes to about 89 cm.
How does IMD give out its forecasts? IMD classifies its rainfall forecast into five ‘ranges’ based on the percentage value of its LPA: deficient (less than 90), below normal (90-96), normal (96-104), above normal (104-110) and excess (more than 110).
How good have these forecasts been in the past? An analysis of ten years’ forecast data shows that the IMD’s April forecast got the ‘rainfall range’ wrong 70 per cent of the times. The June-July forecast, considered a revised and a more accurate monsoon forecast got the ‘rainfall range’ wrong 60 per cent of the times. (See charts). In other words, if the IMD said that rainfall would be ‘below normal’, it could turn out to be ‘deficient’. There have also been a few years when despite a prediction of below-normal rainfall, rainfall was above normal, with many regions experiencing devastating floods.
June forecast against actual rainfall (2006-2015)
April forecast against actual rainfall (2006-2015)
In deciding whether the IMD’s rainfall range was right or not, The Hindu factored in the margin of error that the IMD attaches to its prediction statistics. Usually the April forecast comes with a margin of error of ± 5% and the June forecast with a margin of error of ± 4%. Even accounting for this error margin, The Hindu found the actual rainfall range deviated from predicted levels.
M.S. Swaminathan, head of the M.S Swaminathan Research Foundation and a keen watcher of monsoon forecasts as they matter to the farmers, said that in most cases when the IMD predicted a normal monsoon, it would so happen that rain that is supposed to fall in a month, falls in a matter of two days, with many places even flooding. “IMD forecasts do not provide much insight into the spatial distribution patterns of the rain and that is where they need to improve their services,” he said.
What purpose do they serve, then? We asked the IMD to explain how important these rainfall ranges were, and how it affected the extent of monsoon rainfall to be received in a given year. Laxman Singh Rathore, Director General of Meteorology at the IMD, told The Hindu that the intention of the agency in giving out these predictions was only to convey the direction in which the weather pattern is likely to develop and not to exactly predict what would happen. “If we say that the monsoon is below normal and it turns out to be deficient, then what that means is that we got the tendency correct though the magnitude may not be right,” he said.
And are there reasons to believe that these forecasts will get better? D.S. Pai, Director, Long Range Forecast at IMD in Pune, told The Hindu that five predictors were used for creating models for monsoon forecast, the methodology known as ensemble statistical forecasting system. By way of multiple regressions, the five predictors - Sea Surface Temperature (SST) Gradient between North Atlantic and North Pacific, Equatorial South Indian Ocean SST, East Asia Mean Sea Level Pressure, Northwest Europe Land Surface Air Temperature and Equatorial Pacific Warm Water Volume - were projected to create 62 models using various permutations and combinations of these, and the average of the forecast emerging from the best of these models were taken to arrive at the final figure. Mr. Pai said that this method was introduced in 2007, and since then there have been considerable improvements in predicting the drift of the weather pattern. Even if that was the case in 2009, the Long Range Forecast in June predicted 93% rainfall (below normal) but only 77% rainfall was actually received across India by the end of the year- a huge margin of difference.
Meteorologists agree that with a statistical model, accuracy of predictions is hard to come by. Mr. Rathore admitted that the agency’s skill in the statistical domain was rather poor and more refinement is desirable. He pointed to steps being taken in this direction with efforts to switch to a dynamic model of prediction, in which the ocean and atmospheric temperature levels were coupled to create a single model. The supercomputer Aditya was being tested for this dynamic model of forecasting in the IMD’s Pune centre and “we would be able to switch to this method of forecasting in a few years from now,” Mr. Rathore said.
