Why Can’t An El-Nino Play Spoilsport In 2017? Rains After July Matter Too
I am a great admirer of the Indian Meteorological Department (IMD) and our statisticians. They do a thankless job and those who don’t understand the difference between an initial forecast and a final number, make their life miserable. The statisticians also don’t always get ministers to stand up for them in Parliament to explain that the best forecasts can go wrong. As things stand today, Skymet Weather Services, the private sector company which competes with the IMD, doesn’t have the IMD’s capability in physical modelling of the monsoon.
The IMD has a multi-equation statistical model which has assimilated a large amount of daily weather data over the course of nearly a century. In fact, India borrowed its first supercomputers from the United States in the late-1980s to build these models. When the Americans insisted that they station their own men there to keep a watch that we were not using the supercomputers for other purposes, Indian officials made it clear that we had signed an undertaking, are not a banana republic, and must be trusted. The Americans did not. This meant India had to build its own supercomputers. Vijay Bhatkar, the architect of supercomputing in India, did so in three years flat in the early 1990s. And he was soon displaying India's first supercomputer – the PARAM – in global meets.
Private weather forecasters tend to contradict the IMD. But a look at past forecasts and the monsoon outcome shows that the IMD is right, more often than not.
The Department began to share its medium and long-term forecasts after I, as minister for science and technology, defended the team against attacks made in Parliament. When the Members of Parliament were given a lecture on probability, the Lok Sabha quickly got to the ‘next question’.
Peculiar Early Pronouncements
In February 2016, we began to hear comments about a ‘good monsoon’. Coming after two consecutive years of deficient rains, this commentary made most people happy, but here I was skeptical, as we still had two months to go before the first forecast was officially released in April. For the 2016 Kharif, the IMD did not get the delay in rainfall right, which means they missed out on something that led to agricultural losses. Late rainfall means a crop grown over sixty or seventy days has a lower yield and is delayed by a fortnight.
This is not a decline in output but a decline in achievement as compared to the potential. The disruption in the crop schedule and rain distribution also leads some farmers to grow lesser preferred crops. These are variables that are fairly important for the farmer.
In March 2017, the IMD chief said in an interview that they “are not worried about El Nino at the moment because this weather pattern is likely to emerge only after July.” The El Nino is one of the significant variables in the monsoon forecast model. It occurs when unusually warm ocean temperatures build up in the eastern and central Pacific equatorial region that impact global weather patterns.
Skymet’s forecast that the 2017 monsoon will be below normal due to an El Nino has caught the IMD off guard.
If the IMD is saying something as important as El Nino does not matter, the least they owe the scientific community and the country’s farmers, is an explanation of the science behind their stand.
The IMD’s and Skymet’s projections in 2016 were important because India was coming off two consecutive droughts, and policy makers couldn’t be too careful. In fact, irresponsible talk would have serious consequences in an open trade economy that is largely dominated by the private sector and is globalised at the margin.
Getting Tied Up In Probability
The IMD’s final forecast for the 2016 south-west monsoon, interestingly for the first time, included a probability number to the decision rule, attached to the numerical forecast made. Neither the first forecast for 2016, nor any of the 2015 Kharif forecasts had such a number.
The number “94” created quite some confusion, and since a lot of numbers, averages, and probabilities were floating around, a word of explanation was needed.
The model used is the Experimental Coupled Dynamical Model Forecasting System. In 2016, the main forecast was that Kharif Rainfall will be 106 percent of the long-period average – in other words, instead of 89 centimetres it would be 94 centimetres, with a +/- 5 percent range of error. This was the first ‘94’ number that was put out. But there was another one, which needs a minute of explaining.
Suppose you need to forecast a number from a model. Depending on the regularities and variation seen in the past, there will be a probability associated with each forecast. This gives the accuracy estimate for your forecast if the model is to be believed in. Now that year’s Kharif rainfall would be either 94 cm or it won’t be so. This probability number was meant to give you an idea of the accuracy you can expect on the decision. This probability number that the IMD volunteered in a press conference was, again… 94 (percent). Now, this was strange. Normally, much higher accuracy can be attempted on such numbers. For example, in operations research or war games, 99.5 percent decision rules are common. The decision-making rule is normally 99 percent. A 5 percent variation from 100 is seen as quite shaky. So, at 94 percent you are really out there.
There will always be a probability which will justify any number. Numbers with adverse consequences will become ‘acceptable’ as the probabilities of them being correct decline. At 50 percent, you can toss a coin and see if a number will hold or not.
The 94 percent probability to the decision was a new rule the IMD gave us. The least they could have done is to tell us what this number was in their past predictions.
The reason I get into all this is the fervent hope that nothing of this kind is happening this year. The statement that the El Nino is not important this year cannot be justified by the IMD’s Experimental Coupled Dynamical Modelling Forecasting System. The El Nino is a significant variable there. If the present readings of other variables counter this effect, we need a statistical explanation. Not a sarkari pronouncement by a scientific department.
Yoginder K Alagh is Chancellor of the Central University of Gujarat. He was Union Minister of State (Independent Charge) for Planning and Programme Implementation, Science, Technology and Power from 1996 to 1998.
The views expressed here are those of the author’s and do not necessarily represent the views of BloombergQuint or its editorial team.