In 2018, as Kerala was battered by one of the most devastating floods in a century, weather forecasts failed to provide timely warnings. For Prajeesh AG, one of the key developers behind the AI-powered Bharat Forecast System (BFS), the tragedy became a seed for thought. BFS was developed by the Indian Institute of Tropical Meteorology.
“There was no strong mandate from the ministry or the government,” he said. “However, after the floods in Kerala and flash floods in Himachal, it was clear we needed better systems.” So, Prajeesh and his team set out to build one.
IMD offers high-resolution forecasts tailored for India, potentially outperforming even the GFS model used globally. Yet, despite this technological leap, Prajeesh admitted that none of the technological advancements reach the actual stakeholders.
Why? Because the data and the systems are, in effect, locked up.
Reacting to criticism from private weather players and developers about the inaccessibility of official weather data, Mrutyunjay Mohapatra, director general of meteorology at IMD, told AIM that their data is not closed. “Any data is available upon request,” he said, adding that it is “free for research purposes and available on payment for commercial use”. He cited Reliance Foundation as one of their clients accessing data.
The AI Advantage and the Access Problem
Globally, AI is transforming weather prediction. Microsoft’s Aurora and DeepMind’s GraphCast have made headlines for outperforming conventional physics-based models in both speed and accuracy.
In India, however, access to large amounts of high-quality data is the biggest hurdle.
“In terms of quality, Indian forecasts are as good as any. But even if private users want to pay for this data, there’s no easy way to get it. No web interface, no access protocols,” Prajeesh said.
He isn’t alone in this assessment.
‘IMD Does Not Work With the Industry’
Jatin Singh, founder and chairman at Skymet Weather Services, is more direct. “As a national weather agency, IMD’s data collection is coarse, and it is not shared with people outside IMD. They do not work with the industry,” he said.
Despite India’s rapidly advancing climate risks from cyclones to cloudbursts, Jatin pointed out that IMD continues to operate in isolation.“IMD has been slow in installing radars. Until recently, they used Chinese-made ones, which brought cybersecurity issues. Even now, their radar rollout is slow.”
Jatin emphasised that while Skymet is building district-level AI forecasts and experimenting with latent radio frequency-based systems, IMD remains reluctant to engage with private innovators or offer granular public access.
Addressing specific criticism, Mohapatra said, “Skymet has never officially requested any data from IMD.”
Why This is a Big Problem
“Forecasting is not just about making models,” Prajeesh said. “It’s about transforming raw data into usable formats for real-world decisions. That’s where we fail.”
He pointed to platforms like Windy, a popular app that visualises GFS (US) model data. “Many weather enthusiasts in India rely on GFS data because it’s visual, accessible, and usable. But India’s own forecasts—arguably better for our region—don’t make it to public tools.”
The result? Even renewable energy companies in India buy data from the US and Europe, not because the forecasts are superior, but because IMD doesn’t make Indian forecasts accessible.
“Even if companies want to pay for Indian forecasts, there’s no mechanism to do that,” Prajeesh said.
Skymet’s Approach
While IMD stalls, Skymet is embracing AI with full force. The company’s CTO, Vivek Singh, explained, “We use GraphCast, Pangu-Weather, and Microsoft’s Aurora. But we don’t throw out physics, we create hybrid models that combine machine learning with physics-based forecasts.”
Currently, Skymet runs its models four times per day at a 28 km resolution, aiming to improve it to 9 km. “That’s a limitation of AI/ML models—they can’t always incorporate new or extreme changes without physical models. But with the right data and computation, AI sharpens the output,” Vivek said.
Still, even Skymet’s innovation hits a ceiling if public data, especially radar inputs, remains under lock and key.
For its part, the IMD admits that AI is still “in the developmental stage”.
“Physical models will continue, AI will complement. The physics will remain,” Mohapatra added.
While scientifically valid, this position is seen by private stakeholders as too conservative, especially given the speed at which AI tools are evolving globally.
Skymet focuses less on radars and more on automated weather stations and new sensing techniques. But without access to IMD’s radar data or public APIs, their forecasting capacity is limited to what they can independently observe and compute.
“It’s not about technology anymore,” Jatin added. “It’s about will and access.”
Until India cracks open its forecasting ecosystem by sharing data, building APIs and collaborating with private players, AI in weather forecasting will remain powerful, yet tragically underutilised.
The post Why AI Can’t Fully Replace Traditional Weather Forecasting Yet appeared first on Analytics India Magazine.
