Forecasting a storm in India

The Indian Department of Science and Technology’s directive that only the India Meteorological Department can publish annual monsoon forecasts could set a dangerous precedent, says R. Ramachandran.

In June, India’s Department of Science and Technology took an unacceptable step. It stated that only one of its agencies, the India Meteorological Department (IMD), had the mandate to make public the annual long-range monsoon forecast and that no other agency or institution could do so, even in the research mode.

Accordingly, it issued a directive to the Bangalore-based Centre for Mathematical Modelling and Computer Simulation (CMMACS), an institution under the Council for Industrial and Scientific Research (CSIR), not to make public the forecasts that its scientists have been coming out with every year.

The problem lay in the fact that, since CMMACS began its research exercise a few years ago, these forecasts have disagreed with official operational forecasts issued by the IMD. In the Department of Science and Technology’s perception, such divergent forecasts by different agencies caused ‘confusion’ in the public mind (read market).

It must be emphasised that, unlike the IMD forecasts, the CMMACS results are only posted on the institution’s website, and not publicised through press conferences and the like. The media, however, has been taking these forecasts from the website and putting out stories.

The directive from the Department of Science and Technology implies that CMMACS, or any other Indian institution for that matter, cannot publish such research results, even on their website. As long as CMMACS or the CSIR do not go to town with the forecasts, there can be no valid objection to disseminating research results electronically or otherwise.

A dangerous precedent

Indeed, in physical sciences, there has been a long-standing tradition of disseminating research findings in the form of ‘preprints’ long before they are published. The preprint tradition is very healthy. Circulation of preprints serves to validate research findings by a scrutiny within the scientific community, a sort of informal peer review.

The Department of Science and Technology has argued that since the forecasts are not published in any peer-reviewed journal, they should not be publicised. This goes against the spirit of the preprint or electronic-preprint tradition. The implication of the department’s move, therefore, is that any research that could serve public interest should not be disseminated in advance of publication.

If today the directive is against monsoon research findings, tomorrow it could be against, say, earthquake predictions. Therefore, irrespective of the merits of the CMMACS model, which uses ‘neural networks’ as opposed to the IMD’s statistical approach, and the predictions based on it, the Department of Science and Technology’s move sets a dangerous precedent.

More pertinently, forecasts put out by the IMD are also not peer-reviewed. The IMD’s journal Mausam, in which its forecasts are sometimes published, can hardly qualify to be called a peer-reviewed journal.

Furthermore, while the Department of Science and Technology may exercise its authority over Indian researchers and institutions, it cannot stop research institutions abroad from making monsoon forecasts, which, in this electronic age, become instantly available on the web. Such forecasts, more often than not, disagree with the IMD’s and the media has been carrying stories based on these as well.

Moreover, since 1988 when the IMD forecast was made operational, the predictions have had a high failure rate (nearly 65 per cent). Therefore, the IMD can hardly lay any claim to the primacy or supremacy of its forecasts based on the multi-parameter `power regression’ statistical model that it has devised.

Indeed, there has been widespread criticism of the IMD’s statistical model from the standpoint of its scientific soundness; and this notwithstanding the modifications made recently in the wake of the gross failure of its forecast during the 2002 drought year. Indeed, at a recent conference on monsoon forecasts, J. Shukla, a well-known US-based atmospheric scientist, asked why the IMD does not take the help of statisticians from the Indian Statistical Institute (ISI), who are held in high esteem worldwide.

Statistics and development

29 June marked the anniversary of the birth of the great Indian statistician, planner and visionary, Prasanta Chandra Mahalanobis, who founded the ISI at the Presidency College, Calcutta, in 1931. And it is likely to be declared by the government as National Statistics Day.

Mahalanobis’s approach to the subject of statistics is summed up in this remark: "Statistics must have a clearly defined purpose, one aspect of which is scientific advance and the other, human welfare and national development." His 50 years of work in statistics up to the late 1960s grew from this perspective that statistics are a tool for real life applications, particularly for vital national problems.

One of the early areas of research that Mahalanobis undertook was meteorology. In the early 1920s, his statistical studies in anthropological measurements of people in Bengal attracted the attention of Gilbert Walker, then director-general of the IMD. He asked Mahalanobis to look at certain meteorological problems involving upper air variables. One such problem that Mahalanobis solved in 1923 was that the real seat of origin of meteorological changes is about four kilometres above the Earth’s surface. German scientist Franz Bauer rediscovered this much later from physical considerations.

If he were alive, Mahalanobis would have been appalled by the move of the Department of Science and Technology. In the context of the need for cross-validation of data, he once said: "Anything which is supplied or published by a government office is accepted as reliable. To have any doubt would be a challenge to established authority. The very idea of having cross-checks is frightening as conflicting results arising from independent checks would be ‘confusing’ and must be resisted […]

"A meeting of statisticians, not so long ago, recommended that statistics of one type of information must be collected by only one single government agency so that ‘confusing’ discrepancies could never arise. In this situation, the statistical servicing is bound to be weak in spite of a great deal of knowledge in theoretical statistics."

Today’s statisticians at ISI, despite their extensive knowledge of theoretical statistics, seem to be least interested in such mundane problems as weather forecasting. With a problem so crucial to the country’s economy and IMD’s operational statistical forecasts being the basis of agricultural estimates, economic projections and the market behaviour, it is strange that the ISI has not considered it important enough to look at the problem and correct the basic flaws in the IMD model.

In fact, as became evident during a conversation with this correspondent, the director of ISI, Kalyan B. Sinha, was not even aware of the fact that IMD predictions are based on a statistical model. He thought that the forecasts were based on ‘dynamical’ models, which depend on actual numerical solutions of complex non-linear equations of atmospheric dynamics using high-performance computers.

Glaring disconnect

Indeed, this disconnect is more glaring in the case of technical inputs needed by the national statistical system which comprises, in the main, the Central Statistical Organisation, the National Sample Survey Organisation (NSSO) and, the main user of the data generated by these, the Planning Commission. Considering that Mahalanobis established the national sample survey and was also the architect of the country’s five-year plans based on quantitative economic models, and in his vision of things, ISI was an integral component of this system of applying statistical techniques to address national issues, this disconnect is an irony.

"These institutions have moved away from issues of direct interest to the government to issues driven by research per se," points out Pronab Sen, advisor (perspective planning) in the Planning Commission. "And a meeting of minds has become more difficult than we thought," he adds.

Sinha says, however: "We have moved away from routine problems but we are seeking more challenging problems."

"Even in the so-called routine survey data, there are many challenging problems, provided you are willing to look at [the] data and work with it. But most of them are just high on algebra," says B. S. Minhas, former professor of ISI and former chairman of the NSSO Council. He believes the separation of the national sample survey from the ISI in 1972 (shortly before Mahalanobis’ death) is the cause for declining interest in applications at the ISI.

"Statistics at present is going through a very rapid phase of growth in diverse areas that involve quite intricate theoretical concepts and techniques and these are becoming more and more popular because they get you immediate papers and recognition in the academic world," says Suresh Tendulkar, the present chairman of NSSO Council and professor emeritus at the Delhi School of Economics.

More than merely declaring National Statistics Day to mark the great professor’s birth date, a great deal needs to be done to bring back to centre-stage Mahalanobis’ vision of "statistics as a key technology" to solve national problems.

R. Ramachandran is science editor of Frontline, a fornightly magazine published by The Hindu group. This article originally appeared in The Hindu newspaper and has been reproduced with permission.