Ankur Paliwal is a journalist. He mostly writes about science, health and the environment.
In mid-July this year, a group of farmers in Anandgaon village in Beed district of Maharashtra filed a police complaint against the Indian Meteorological Department (IMD). They accused the department, which forecasts weather, of misleading them. Around that time, Swabhimani Shetkari Sangathan, a farmers’ body in Maharashtra, threatened to lock up IMD’s office in Pune. “If IMD cannot forecast accurately, it is better we shut down this department instead of keeping such white elephants,” Yogesh Pande, spokesperson of the farmers body, had said.
In early June, farmers across India heard from the IMD that the monsoon would ‘most likely’ be normal this year. That was followed by weekly weather and farming advisories issued by the central government’s 130 regional agrometeorological units spread across India. On June 12, an agrometeorological unit in Maharashtra’s drought-prone Marathwada region asked farmers to start sowing their kharif, or monsoon, crop. They did. In fact, four times more than in the last year in anticipation of good rains. And it rained well until June 20.
The following week was mostly dry. The same happened in the week after that. Staring at a long dry spell, the agrometeorological unit asked the farmers to undertake protective irrigation – a technique in which water available from rivers or reservoirs is thinly spread in an equitable manner over a large area. But agriculture in Marathwada is mostly rain-fed and most reservoirs didn’t have enough water. When the dry season didn’t seem to improve, Maharashtra’s chief minister Devendra Fadnavis, asked farmers on July 9 to delay sowing till July 20. But by then as much as 75% of the kharif crop had already been sown. This led to the farmers’ outburst.
Following the dry spell, which lasted from 20 to 45 days across Maharashtra, more dry days in the Marathwada region and reports of crop loss began to trickle in. In Latur, the crop yield of moong and urad was reduced by 40-60%. “We wouldn’t have suffered this loss if the forecasts had been accurate and communicated to us about 10-20 days in advance,” said Pande over the phone. But is India’s meteorological department capable of delivering this? Maybe not. The weather models that the IMD and its agencies use have limited capacity. They are being improved upon but at a snail’s pace. Then there is more that cripples the forecast.
Use of numerical models
The weather department issues five kinds of forecasts. Nowcast is for less than 24 hours. The short-range forecast is for up to three days. The medium-range is from three to 10 days. The extended-range is for 10-30 days. The long-range is on a seasonal scale (e.g. monsoons). These forecasts are used for various enterprises, such as agriculture, transport and water management.
The forecasts are generated with the help of weather models. Over the years, the meteorological department has moved from crude to a mix of crude and advanced models to predict the weather.
More than a century ago, when there were no computers, IMD’s forecasts depended only on snow cover. Lesser cover meant a better monsoon. Then in the early 20th century, the British physicist Gilbert Walker, who’d had no experience in meteorology, came to head the IMD. He designed a statistical weather model – an empirical way of predicting the weather – based on the relationship between two weather phenomena. For example, historical data between the monsoons and El Niño or unusual warming of the Pacific shows that they have a negative relationship. A strong El Niño brings on a weak monsoon. In fact, this is how the IMD dispensed its long-range forecasts until recently.
However, a lack of specificity and confidence in an observational model pushed scientists to find more scientific ways to predict the weather, something they could use to quantify weather phenomena. In 1994, the National Centre for Medium Range Weather Forecasting (NCMRWF) began to employ numerical weather prediction models for short- and medium-range forecasts. A numerical model, simply put, is a set of complex algorithms that solve certain atmospheric equations that will determine the weather’s ‘status’ in the future. For example, an algorithm might solve a thermodynamics problem to predict temperature values a couple weeks down the line. And in 2014, the IMD started to use numerical models to supplement statistical models for long-range forecasting as well.
Now, although the numerical models used by the IMD are state-of-the-art – developed by the US National Centres for Environmental Prediction – their forecast capacity is still weak. Longer the period of forecast, the more uncertainty there is.
“That is because the algorithms solve the equations for a small portion of time at a time, and then repeat, say, every two minutes. Some error creeps in at each step,” S.K. Dash, professor emeritus at the Centre for Atmospheric Sciences (CAS) at IIT Delhi, told The Wire. “So, longer the time input, more is the error.” This is why seasonal forecasts are often off the mark and also why it’s hard to predict extreme weather events in advance.
“Seasonal forecast is still in the research mode, I don’t think anybody will swear by it that all is right,” said Jayaraman Srinivasan, a professor at the Centre for Atmospheric and Oceanic Sciences, Indian Institute of Science (IISc), Bengaluru. The meteorological department officials have said that, over time, they have been able to achieve 60% certainty in forecasts for up to five days. The confidence drops after that.
Need for block-wise data
The officials also say that at the end of the season, the monsoon is still likely to be normal, with just 5% deviation from prediction. However, that is because some parts of India – like western Rajasthan, Saurashtra and Kutch – received more than normal rainfall (the average annual rainfall between 1951 and 2000). “But that information isn’t really of much use because a lot of central India and parts of Maharashtra and southern India had very long dry periods,” said an IMD official on condition of anonymity. “What the department needs to do is to divide India into different zones and issue long-range forecasts for each zone.”
While Pande and other farmers demand accuracy in long-range forecasts that could help them figure out whether or not to sow in the season, what they need more is forecasts at the block level. A block is an administrative unit within a district. “Knowing how much and when it will rain in Maharashtra or even in Marathwada [a cluster of eight districts in Maharashtra] doesn’t help us because rainfall varies so much even within the district,” Pande said.
At the moment, the IMD provides district-wise weather data but this is also inchoate. For example, when it says there will be scattered rainfall over a particular district, it means that 26-50% that district (by area) will receive rainfall. “That helps only to a certain extent,” said Mohan Gojamgunde, an agriculture officer in Latur. What it doesn’t tell us is the shape of that ‘26-50%’ region, he added. According to Ashis K. Mitra, a scientist with the National Centre for Medium Range Weather Forecasting (NCMRWF), Noida, this is difficult to predict. “The error in forecast increases as you decrease the area. Weather is a complex phenomenon and rainfall is an especially-so variable that is hard to get right,” said Srinivasan. “You should accept this with humility.”
But why aren’t we able to provide block-wise forecasts? There are many reasons.
Dynamical models won’t run until you feed data about current weather conditions – a.k.a. the initial values – to predict the values of the future. As Dash says, “If you need block-level forecasts, you need to put in block-level initial values.” The IMD collects weather data like temperature, humidity, wind and precipitation through 679 automatic weather stations, 550 surface observatories, 43 radiosonde or weather balloons, 24 radars and three satellites (which constantly share weather data with other global satellites). Sadly, these numbers are just not enough given the size of India. “If my observations are 100 meters out and I don’t know what is happening in between, I approximate those values in the model. Those approximations might not be very accurate,” said Dash. “That affects the quality of the forecast.”
Then, there are major data gaps, like those involving dust, aerosols, soil moisture and maritime conditions. Scientists say that India needs at least 20 more radars to collect precipitation data at the wider and vertical levels in the atmosphere.
Moreover, the lack of data is just one challenge. What paralyses models is its poor quality. A retired scientist from the Pune meteorological department, who wished to remain unnamed, said that some of the automatic weather stations are of substandard quality and that the government purchased them at lower prices to cut costs. “Substandard equipment will give you substandard data,” he said. He added that the upkeep of instruments was/is a major problem. “They need to be calibrated and cleaned regularly, which doesn’t happen often. That affects data.”
Mitra claims that the government has begun to replace the substandard instruments. “We are very conscious about improving our data and model’s capacity. We plan to give block-level forecasts in a year or two.”
Lack of software professionals
Some scientists say that one area where little progress has been made is the fine-tuning of weather models to suit Indian conditions. “The models that we have brought from the west have been developed by western scientists to forecast in their region,” Dash said. Models are fine-tuned by incorporating more Indian, or at least tropical, information of physical processes like how convection evolves over the subcontinent. However, providing that information is difficult because weather systems in the tropics aren’t understood very well. One reason for this is that weather systems destabilise faster in the tropics than they do in the extra-tropics, where they persist for longer durations.
This is also why processes like cloud-formation are neither well-understood nor well-represented. Similarly, we know little about why the monsoons vary as much as they do on the terms of their relationship with other weather phenomena. “But [the fine-tuning to fix this] is not something that you can do in one or two years,” said Srinivasan. By some estimates, it takes a decade of work to improve the forecast by one day. Put another way, today’s five-day forecast is as accurate as the four-day forecast was ten years ago.
Stirring this messy pot further is, of course, global warming. Although some scientists suspect that rising temperatures are responsible for recent changes in the monsoon’s behaviour, like increased intensity of rainfall in short durations, scientific consensus is awaited.
India realised the importance of meteorology much after the West did. After attaining independence, when the government had been grappling with a host of socio-political issues, meteorology was somewhere at the bottom of its to-do list.
When western weather scientists had started to move towards analytical methods of forecasting, their Indian counterparts were still stick to empirical methods. As a young scientist, Dash had access to a weather model from the European Centre for Medium-Range Weather Forecasts (UK) a copy of which he brought to India in 1984. However, he couldn’t find a computer in India that could run it. “It was frustrating,” he recalls. India’s first supercomputer itself would come online only five years later.
Ironically, now that India has more computing power and better models to work with, there is a lack of competent software professionals and scientists working with the IMD. Only a small fraction of the modellers and atmospheric scientists that graduate from institutes like the IISc are recruited by the IMD and its associated agencies.
When India launched a programme to improve the prediction of monsoon rainfall in 2012, the NCMRWF asked the Ministry for Earth Sciences for 20 more scientists. That hasn’t happened yet. Kanduri Jayaram Ramesh, the director general of the IMD, however, isn’t happy that he isn’t getting Indian scientists to work with either. “They either go abroad or don’t join us.”
There aren’t many dedicated groups of Indian scientists conducting long-term experiments examining the complexities of Indian weather systems and incorporating that data into models. But that’s changing gradually. Just a couple of months ago, CAS received a go-ahead from the Ministry of Science and Technology to launch a Centre of Excellence in Climate Modelling. Once it’s operation, it will support a group of ten scientists studying how we can improve models and reduce errors.
As heartening as this is, the pace of it all hasn’t pleased farmers. Pande says that the members of his community have tired of being told that the meteorological department can’t provide accurate forecasts because science hasn’t progressed fast enough. “You can save yourself by saying that. But what do we do?”