Why predictions of national weather forecaster go awry
Early morning fog, bone-chilling cold nights — as North India reels under one of its coldest December winters, there are fears it could get worse as the harshest period is yet to begin.
The winter was preceded by monsoon rains that devastated at least 13 states, including Bihar, Uttar Pradesh, Maharashtra, Karnataka and Assam.
The two seasons have a common thread — awry predictions by the India Meteorological Department, the country’s apex weather forecasting body. The IMD was so off mark on both occasions that the weather has been opposite of what it predicted.
In November, IMD predicted a winter warmer than normal for Northern India, especially in the month of December. New Delhi, however, is witnessing its worst winter in over 100 years.
The IMD had also predicted a normal June-September monsoon with rainfall of 96% of the Long Period Average (LPA).
The season saw an excess rainfall of 110% that flooded several cities and was the highest rainfall in the country since 1994.
This is consistent with forecasts that have been inaccurate when it comes to temperatures, droughts, heat waves, cold waves, and the monsoon.
They, however, aren’t an anomaly. Getting the weather wrong isn’t unheard of, even in other nations. And it comes down to how weather is modelled and how much we understand it.
There are traditionally three types of forecasts: a short-term forecast for a week to 10 days, including daily temperatures and time periods for rainfall; seasonal forecast for a period of 60 to 90 days, including the amount of rainfall; and a long-term forecast that deals with climate changes 30 to 40 years in the future. Of late there is also a fourth: decadal forecast for a period of 10 years.
What creates inaccuracies in short-term weather modelling is input data. For short-term forecasting, scientists collect data from weather stations across the country, which transmit information about local temperature, precipitation, humidity, wind speed, atmospheric pressure and more. The information is captured every six to 12 hours and fed into a weather model.
The model then extrapolates data and churns out a forecast.
IMD uses a model developed in the US called the Weather Research and Forecasting (WRF) Model, which assimilates data and processes.
The IMD model adapts this by uploading Indian data and calibrating it for an Indian setting. They have been uploaded on the Prithvi High Performance Computers at the Indian Institute of Tropical Meteorology, Pune.
The system, however, isn’t always perfect.
The IMD does fairly well in forecasting weather for five to seven days but in some locations, the numbers can be off. The data for all models come from IMD sensors which aren’t enough.
Like in the Himalayas, there is a sparse network of weather stations, some as far away as 80 to 90 km from each other. There are just 77 stations across Jammu and Kashmir, Himachal Pradesh and Uttarakhand. There are 8,000 stations in the rest of the country, and several are non-functional.
“Because of the complex topography of the Himalayan region, there are serious challenges in short-term prediction,” said Sarita Azad, assistant professor at IIT-Mandi. Azad’s ongoing research, an analysis of 18 cloudbursts in the region, revealed that only six locations had weather stations in their vicinity. With such a deficit, models cannot be calibrated accurately.
Azad says the interpolation to grid is too basic. While it works well for plain surfaces, it fails to take into consideration varying terrain between stations.
There is also a lack of co-ordination between agencies tracking the weather, scientists say.
The resolution of forecasts is also a factor in accuracy. IMD forecasting typically generates predictions with a resolution of 12 km, meaning the system is sensitive enough to offer differences in forecast for regions 12 km apart.
Mrutunjaya Mohapatra, IMD director general, says the difficulty in forecasting is due to the nature of the country’s weather.
“Unlike countries like the US and UK, weather in tropical countries like India is like an undisciplined child,” said Mohapatra.
Mohapatra says qualitatively, IMD predicted the monsoon trends correctly. “If you look at the statement issued in April/May, we said that weak El Nino condition is prevalent and it is likely to become weaker, and will approach neutral condition in the second half,” Mohapatra said. “Based on this, we said June will be deficient, then rainfall will gradually increase. The second half will have better conditions.”
“What happened is that June became deficient, July was normal, August was excess and September was large excess. Which means that qualitatively what we predicted occurred,” he said.
The IMD claims that while it can predict large-scale parameters that affect seasonal monsoons, it still struggles with variations within a season.
Other scientists say that predicting weather within a season is a difficult task across the globe.
“There is uniform accuracy across the world in predicting the weather for five to 10 days. We are good at predicting long-term climate. But I can tell you that there is very little prediction skill for seasonal forecasts,” said Govindasamy Bala, professor at the Centre for Atmospheric and Oceanic Sciences at IISC.
For seasonal forecasts, the concept of internal variability plays a role. As there are parameters that cannot be controlled and understood, the chaotic nature makes prediction hard.
This is despite the fact that IMD makes use of two kinds of modelling mechanisms. One is the statistical model that uses datasets of over 100 years to find correlations in weather patterns.
A new model, developed by IITM-Pune built on top of the American Climate Forecast System, is a coupled ocean-atmosphere modelling system that takes in data for long-range seasonal forecasting. “There is no theoretical basis for improving accuracy with our current understanding of weather,” Bala said.
One major problem is that the underlying math behind climate modelling is a method of approximations. Thus a probability of errors is inherent. “If we manage to locate the right parameter and see how it behaves, we could get new information,” said Bala.
Weather scientists are looking for patterns similar to the Eurasian snow cover, which can hold a cause-and-effect relationship with seasonal weather. Generally, if there’s a high amount of springtime snow in Eurasia, the monsoon is weaker in summer.
But this can’t be taken as a rigid indicator. There are other phenomena whose understanding could improve seasonal forecasting. These include formation and evolution of El Nino Southern Oscillation, where surface temperatures in Pacific Ocean rise and fall for long periods.
Another is the Indian Ocean Dipole (IOD) where the Arabian Sea becomes alternately warmer and cooler than the Bay of Bengal.
A third is the Atlantic Zonal Mode (AZM) that has an inverse relationship with monsoon’s intensity.
This year’s excess rainfall is thought to have occurred due to either the IOD or lack of weakening of El Nino or both, while the excessively cold winter likely occurred due to Western Disturbance (WD). “There is some variable in these or other similar phenomena that affects seasonal rainfall in a specific way, but science hasn’t managed to locate it yet,” Bala said.
One of the areas that the IMD has earned credit is in the prediction of cyclones. The loss of lives in 1998 Gujarat Cyclone and 1999 Odisha Cyclone prompted the government to modernise the IMD and set up the Ministry of Earth Sciences in 2006.
“It took 18 hours to make a prediction. By that time the cyclone would have moved and completely changed course,” Mohapatra explained. “Data was plotted by hand. Telegrams would reach in six hours. Someone would write bulletin by hand.”
“There has been a paradigm shift in all aspects of forecasting, from observations, modelling, forecasting, and dissemination.”
India’s cyclone preparedness ahead of Cyclone Fani earned worldwide praise as IMD’s forecast helped authorities evacuate people.
The Ministry of Earth Sciences plans to develop a ‘Centre of Excellence’ by 2024 to address the needs of the North Western Himalayan region. A major augmentation of the observational network in the region has been envisaged, according to a statement by the ministry.
Mohapatra said the aim is to strengthen the observational network and make improvements to dissemination. “The entire country is not covered by radar, so we are planning to have a total of 62 radar systems in the next five years,” Mohapatra said.
IMD is also planning a mobile app for better dissemination of ‘nowcast’ — very short term, localised weather predictions.
There is also the problem of pollution. Experts have been calling for a dynamic action plan in response to the prediction system at System of Air Quality and Weather Forecasting and Research (SAFAR) at IITM. However, the system’s forecasts are dependent on IMD’s modelling.
“All the weather-related input data comes from the IMD, but the information related to air quality is collected by SAFAR,” said Gufran Beig, Project Director of SAFAR.