IMD taking baby steps in using AI, but apprehensions over accuracy remain

The Indian Meteorological Department is testing AI-based forecasting models, but has found that physical models still offer higher accuracy, especially regarding intensity of specific weather events

The India Meteorological Department (IMD), which is looking to actively integrate artificial intelligence (AI) models into weather forecasting, has launched small pilot programmes but remains cautious about their accuracy across all aspects of forecasting, IMD Director-General Mrutyunjay Mohapatra said.

 

The Ministry of Earth Sciences, which oversees IMD, has constituted a high-powered panel of experts drawn from IMD, the Indian Institute of Tropical Meteorology in Pune, the National Centre for Medium Range Weather Forecasting, and other agencies to deliberate on the use of artificial intelligence (AI) in forecast modelling.

 

Internally, IMD has also formed a panel to work on AI-driven forecasting and explore broader applications of AI. 

Meanwhile, the department is using AI to develop a ChatGPT-like communication tool, and the multilingual weather forecasting app Bhashini already employs AI to translate forecasts into regional languages.

 

“Not only that, we have also partnered a host of private and public research institutions such as Indian Institutes of Technology, Indian Institutes of Management, and other universities to jointly develop AI-based models and applications,” Mohapatra told Business Standard recently.

 

He added that IMD is studying various AI models and exploring ways to adapt those used by European agencies such as the European Centre for Medium-Range Weather Forecasts (ECMWF). Currently, IMD is using AI for smaller, day-to-day tasks such as calculating heat indices and tracking the paths of tropical cyclones, among other applications.

 

“But these are very small uses of AI,” Mohapatra admitted. “An all-India, global, or even regional AI-based model for predicting rainfall, temperature, wind speed, or any other parameter — comparable to what we achieve through physical models — has not yet been developed. But it is under development by the ministry.”

 

He added that there is no firm time frame for when a fully AI-based weather forecasting model will be ready in India, only that the work is ongoing.

 

According to Mohapatra, this limitation is not unique to India. Globally too — whether in Europe or China — no country or agency has yet fully relied on AI to develop weather forecasting models. “It has not replaced physical models anywhere so far,” he pointed out.

 

“People are still experimenting with AI, and, most importantly, for now, it cannot replace all existing physical models,” Mohapatra said.

 

He cautioned that AI-based forecasting models are not always completely reliable and sometimes show lower accuracy than physical models.

 

Mohapatra cited recent tropical cyclone Montha as an example. He said IMD saw the AI model of ECMWF. However, the track forecast of IMD did not depend upon any individual model. "We are checking the performance of different models. It will take time," Mohapatra said.

 

“There are many such examples of AI not performing uniformly well in all aspects. Especially for extreme weather events, our experience shows that AI has not been able to capture their intensity accurately,” he said. “So yes, there are still many issues with AI, though it continues to evolve both in India and globally.” 

“If you ask whether it will rain or not at a particular place, AI may perform well,” he said. “But if you ask whether it will rain heavily or not, AI will fail.”