The Hindu : Monsoon forecasting

Online edition of India's National Newspaper
Friday, Jun 21, 2002

Opinion

Opinion - Leader Page Articles

Monsoon forecasting

By N.Gopal Raj

Despite all the advances in science... the mechanisms which cause the year-to-year variations in the monsoon are still poorly understood.

ON JUNE 4, 1886, the India Meteorological Department (IMD) issued its first operational seasonal forecast for the monsoon. Since then, each year the IMD has been forecasting how good or bad the summer monsoon was likely to be over the country as a whole. It undertook this task after the failure of the 1877 monsoon, the worst on record, led to a severe famine. The primary purpose of the IMD forecasts were, therefore, to forewarn the Government so that it could be prepared in case of a bad monsoon.

Although the spectre of famines has been abolished, the large year-to-year variability in the monsoon rainfall still causes concern. After all, irrigated area forms only about 40 per cent of the area under foodgrain output of which in a monsoon deficient year (1987-88) can be almost 20 per cent lower than in a normal year (1989-90).

Despite all the advances in science, the advent of powerful computers and availability of satellite-based sensors, the mechanisms which cause the year-to-year variations in the monsoon are still poorly understood. "Understanding the physics of the vagaries of the Indian monsoon is one of the most challenging problems in atmospheric science today," says J. Srinivasan of the Indian Institute of Science.

In the absence of such knowledge, the IMD has depended on statistical methods to predict the all-India rainfall during the summer monsoon (June to September). Its early forecasts depended on just one predictor, the Himalayan snow cover. More snow in the Himalayas was bad for the monsoon while less snow presaged a good monsoon. But this soon proved unsatisfactory. A major breakthrough came in 1907 were Sir Gilbert Walker, then IMD Director General, scientifically established the correlation of various meteorological parameters with the rainfall. The subjectivity of the earlier method was replaced with mathematical multiple regression models to compute the rainfall from these predictors. Since 1988, the IMD has relied on its power regression model for forecasting the monsoon for the country as a whole. This statistical model uses 16 parameters to compute the rainfall. But on nine occasions the actual all-India rainfall has fallen outside the specified error margin of plus or minus four per cent, giving an error rate of close to 65 per cent. Fortunately for the IMD, India is currently going through a period of normal or above normal monsoons. A normal monsoon is defined as one where the rainfall is within 10 per cent of the long period average of 88 cm. The last deficient monsoon was in 1987. Even if the actual rainfall was outside the error margin, IMD was able to correctly predict that the rainfall would be normal or excessive (as happened in 1988).

Any statistical model depends on the past correlation between predictor and rainfall continuing. Unfortunately, the strength of this correlation can vary and even reverse itself. Consequently, after 1930, the IMD had to periodically update its multiple regression models. Analysis of more than 20 known predictors has revealed that many of them had lost their significant relationship with the monsoon during recent years, says M. Rajeevan, IMD's Director of Long Range Forecasting, in a paper published recently. As a result, the IMD's power regressional model has performed particularly badly in the last few years. Since 1994, the actual rainfall has been within the error margin only in one year. In 2000, the IMD replaced four of the original 16 parameters, saying this was necessary to contain the model error to within four per cent. Despite this, the actual rainfall during both the 2000 and 2001 monsoons remained outside the forecasted margin of error. Even more interesting is Dr. Rajeevan's finding that periods of normal rainfall coincide with those of weaker correlation between predictors and the monsoon. As a result, these were also periods when the statistical models showed poor predictive skills. If that is so, then perhaps no model need be used for forecasting during such periods of normal monsoon!

Moreover, as D.R. Sikka, former director of the Indian Institute of Tropical Meteorology, observes, "spatial averages of a highly variable parameter like rainfall are meaningful only if there is homogeneity within the area being averaged". Instead, the all-India rainfall figure hides considerable regional disparity. Although the 12 years from 1989 to 2000 are all categorised as having had "normal" monsoons, between 12 and 35 per cent of the districts in the country suffered deficient rainfall. Last year, when too the monsoon was normal for the country as a whole, there were floods in Orissa and Bihar while many parts of Rajasthan and Madhya Pradesh had drought.

The obvious solution is to predict rainfall for smaller and more homogenous regions. Since 1999, the IMD has reintroduced separate monsoon forecasts for north-west India, peninsular India and north-east India separately. The year-to-year variations in rainfall are greater for these regions than for India as a whole.

Consequently, the IMD gives its forecasts a greater error margin of plus or minus eight per cent. Moreover, north-west and peninsular India are still quite large and disparate regions, the first forms about 30 per cent of the Indian land area and the latter close to half. So even if the monsoon forecasts are tolerably accurate, the issue could still arise as to whether the aggregated rainfall is truly representative of the region.

Ideally, rainfall recorded in all stations in a region should be strongly correlated with one another. One scheme which divided the country into coherent rainfall zones resulted in over 30 divisons. This poses another set of problems. Some regions have good predictability, typically those with high rainfall variability. But the rainfall in other areas, such as the north-east and coastal regions, is not easily predicted. On a sensitive issue such as the outcome of the monsoon, issuing predictions only for some regions and not others, however sound the scientific justification, is not an option open to a national body such as the IMD.

Even if such factors limit how much in advance the monsoon can be predicted, shorter range forecasts too can be valuable. Already, computer weather simulation models show promise in predicting rainfall a few days in advance. Farmers, however, would like to have forecasts 10 days or more in advance to plan their activities accordingly.

Ultimately, improving prediction is going to depend on understanding the complex processes which determine the progress of the monsoon. There are still major gaps in the availability of data needed for this purpose, says Prof. Srinivasan.

More meteorological information is needed from Afghanistan and Central Asia. Vertical profiles of temperature and water vapour over the oceans surrounding India, which cannot be got from satellites, are also required.

Predicting the monsoon remains the challenging and important problem it was when the first efforts began in India more than a century back.

Send this article to Friends by E-Mail

Opinion

Stories in this Section

Disingenuous reckoning

Tactics and terror in West Asia

From the jaws of self-defeat