Chennai: In a country where it is difficult to predict even the arrival and departure of trains, forecasting the onset of the monsoons has been speculative. The situation seems to be changing. The Indian Meteorological Department has developed a new, two-stage, eight-parameter power regression and probabilistic model, which it used for the first time in 2003, to predict country wide rainfall with remarkable accuracy.

The power regression model is a statistical model based on a non-linear regression technique. Data for 38 years from 1958 to 1995 has been used for the development of the model with a margin of error of ± 5 per cent.

According to the meteorological department, the probabilistic model is based on the 'linear discriminant analysis' (LDA) technique, which gives the probability of monsoons for the country as a whole in 5 broad rainfall categories. The probabilistic model was developed using 40 years of data from 1958 to1997.

''India has a pronounced monsoon, unique in the world,'' says S R Ramanan, director, Area Cyclone Warning Centre, Indian Meteorological Department, Chennai. Though the meteorological department has been a butt of jokes in the past, its predictions proved right last year, after having gone awry in 2002. It is expected that this year's forecast, too, will prove right.

The anxiety about the monsoons in India is understandable, given the extent of the economy's dependence on agriculture. The Indian farm sector, with small, fragmented landholdings, too miniscule for mechanisation and totally dependent on timely rainfall, generates about 40 per cent of the country's gross domestic product. A good south west monsoon between June and September means higher farm incomes and enhanced spending capacity which in turn lead to stronger demand for consumer goods, and higher savings.

India witnessed its fullest monsoon in more than a decade last year, which brought in a bumper harvest of food grain and commercial crops leading to a 16.9 per cent growth in the agricultural sector. This raised overall economic growth by an impressive 10.4 per cent in the third quarter (September to December) of 2003-04.

Meteorological forecasts can be classified into four broad categories:
Current forecasts: predictions for next few hours. This data is mostly used by the aviation sector.
Short term forecasts: prediction for the next 24-48 hours.
Medium term forecasts: the forecast span is ten days. These forecasts are conducted by the National Centre for Medium Range Weather Forecasting (NCMRWF) located in New Delhi, and;
Long term forecasts: predicting the onset of the south-west monsoon.

The meteorological department's forecast on April 15, 2004, of a near normal south-west monsoon (June - September) using the global weather data available up to March, was based on the new model. After the forecast last year, the new model was subjected to peer review and found satisfactory. The results were published in Current Science (vol.86, no 3 on February 10, 2004), a leading international research journal published by the Indian Academy of Sciences, Bangalore.

According to Ramanan, the new model represents the combined brainwork of the department. The earlier model, based on 16 parameters, had been in use since 1988 and had provided sufficiently accurate forecasts, till 2002 when it failed to predict the severe drought that the country experienced that year. The actual rainfall was deficient by 19 per cent of the forecast and July itself, which sees plentiful rains, experienced severe shortfall.

With the new model, the drought of 2002 could have been predicted. This provided the impetus to switchover to the new model last year and go in for the two-stage forecast now.

Based upon the new model, the Indian Meteorological Department's long range forecast for the 2004 south-west monsoon season is that the rainfall for the country as a whole is likely to be 100 per cent of the long period average (LPA) with a model error of ± 5 per cent.

The new model's indication (for whole of India) for 2004, based on the new model are:

• 16 per cent probability of below normal rainfall (90 to 97 per cent of LPA)
• 58 per cent probability of near normal rainfall (98 to 102 per cent of LPA)
• 18 per cent probability of above normal rainfall (103 to 110 per cent of LPA)
• 4 per cent probability of excess rainfall (more than 110 per cent of LPA)
  

There is only a four-per cent probability of the south-west monsoons being deficient (rainfall less than 90 per cent).

After the first forecast in mid-April, the next update will be made during the last week of June 2004.

The factors that are taken into account in the new formula are:

• Arabian sea surface temperature
  
• The El Nino (refers to the warming of eastern Pacific ocean surface temperatures that is seen to cause disruptions in weather patterns worldwide) effect in the previous year (July, August and September)
  
• Eurasian Snow Cover of December
  
• North West European temperature during January
  
• January-February pressure radiant in Europe
  
• East Asia Pressure
  
• South Indian Ocean Temperature in March
  
• 50 hecto pascal (height in metres above sea level) wind pattern during January-February
  

According to Ramanan, weather prediction is a correlation of different phenomenon that prevail in different parts of the world and their impact on our weather. For instance, wide snow cover will make the atmosphere cold and inhibit the monsoon from setting in. On the other hand, warm temperatures around the Arabian sea would aid the onset of monsoons in India.