In And Now All This, a 1930s textbook parody, Sellars and Yeatman wrote that agriculture in India depended on the “rainy season, which is known locally as the mongoose. It is often a wrongoose when the rainfall is deficient. Sometimes it is a bongoose”.
Jokes aside, not much has changed. The monsoon still dictates India’s food production, along with its urban water supply. Less than 35 per cent of agricultural land is under irrigation. Roughly 75 per cent of annual rain falls between June and September.
Official monsoon predictions are not only inaccurate. They are also insufficiently granular and released too late to be of practical use. The first seasonal forecast is issued by the Indian Meteorological Department (IMD) in April, and this is updated in June. It relies on statistical analysis of historic data and gives projections for India as a whole. In June, after the monsoon sets in, predictions are updated for four regions.
This April, IMD predicted that total rainfall between June-September would be 99 per cent of the long-period (50-year) average. In June, rainfall was downgraded to 96 per cent of average. June rain has been about 30 per cent lower than average.
The assumed error range for IMD predictions is five per cent (plus or minus). A comparison of IMD predictions between 1994 and 2009 with the actual rainfall of these 16 years shows much larger errors. Actuals deviated by over five per cent (plus or minus) from predictions in 12 years. Errors exceeded 10 per cent in seven years and ranged from minus 20 per cent (2002, 2009) to plus 18 per cent (1994).
More reliable predictions, released earlier, would optimise planting times, fertiliser use and crop rotation. A granular break-up would also help administrations tackle floods and droughts at district level. This is only possible if IMD can upgrade its forecasting models and that, in turn, means massive induction of technology.
The National Monsoon Mission (NMM), launched earlier this year, aims to build the infrastructure and skills necessary to move from the old regression statistical methods to dynamic forecasting models. Coordinated by the Indian Institute of Tropical Meteorology (IITM), Pune, the NMM will pull in academics as well government organisations.
The maths of weather-forecasting is complex. The best results come from multi-variable dynamic models that can mimic the physics of weather. Chaos theory – the study of dynamic non-linear systems – has been largely driven by a need to understand weather.
Fast supercomputers are required. Thousands of differential equations are programmed and solved to simulate rainfall and temperature patterns, with a broad range of assumptions of the value of many variables. Weather is influenced by many variables that interact with each other in a non-linear and complex manner, creating multiple feedback loops.
The El Nino effect for example, arises when Pacific Ocean temperatures on the other side of the world rise by sometimes as little as 0.5 degrees Celsius. When it occurs, El Nino has a big negative influence on the monsoon. Dynamic forecasting models must account for very local factors as well as global effects. Data collection has improved dramatically with remote sensing satellites and new methods of analysing radar signals also offers insights.
IMD has a long track record. It issued its first predictions in 1886. By the 1920s, it had built a regression model, statistically correlating historical data. In 1989, Dr V R Gowarikar, then the secretary of the Department of Science and Technology, developed a 16-variable model. But the 16-V model disastrously missed out on predicting the 2002 drought.
In 2004, IMD started using experimental dynamical models for long-range predictions, but it lacked the silicon muscle to do this for shorter-terms. The official predictions still use the old statistical methods. Several other government departments have tried their hand at dynamic forecasting. The Centre for Mathematical Modelling and Computer Simulation at CSIR is a major research hub along with the National Centre for Medium Range Weather Forecasting in Noida.
The NMM will start with the Climate Forecast System of the National Centre for Environmental Prediction, USA, as a base-model. It hopes to adapt the CFS-NCEP to develop dynamic models that fit Indian conditions. The aim is to get dynamic seasonal and extended range predictions (16 days to one season) as well as short to medium range predictions (up to 15 days). A 38 km x 38 km grid will be applied for granularity. NMM has been allocated Rs 400 crore over the next five years. Much of that investment will go in creating the required computing resources. Once the models are “back-tested” for reliability, they will replace the old regression-based statistical methods. At that stage, the reliability and practical use of weather predictions should improve significantly.
