India's weather forecasting infrastructure has a well-known blind spot. Measurement systems are largely concentrated at airports, seaports, and major public buildings, designed for an era when a 100-kilometre grid was good enough. That era is ending.
The Indian AI Research Organisation (IAIRO) is now working in close coordination with the India Meteorological Department (IMD) to bridge that gap using artificial intelligence, and the ambition is significant: weather data personalised not just to your city, but to your street.
Prof. Dev Niyogi, an IAIRO founding member and UNESCO Chair at the University of Texas at Austin, believes that the convergence of AI, expertise, and institutional readiness offers India a crucial window it cannot afford to miss.
"What we needed was an authentic range of AI researchers working with authentic physics and dynamics and health and meteorology and climate experts," Niyogi tells AIM. "Together, when they bring the best of their knowledge, what comes out of it is magic."
Why India's Weather Is Uniquely Hard to Predict
Before understanding what IAIRO and IMD are building, it helps to understand the fundamental difficulty of the problem. Unlike Western nations, where most weather is driven by frontal systems — predictable, linear, trackable — India's rainfall is largely convection-based. Rather than a wave rolling in from a distance, convective rainfall behaves more like bubbles rising in a pot of boiling water: chaotic, non-linear, and resistant to precise prediction.
"Even if you have the best model, it still has a degree of randomness and uncertainty because you can't predict where the bubble will come up," Niyogi explains. The phenomenon is visible to anyone who has lived in Bangalore. It can pour on one side of a street while remaining bone-dry on the other, a product of topography, urban heat islands, and traffic-driven pressure differentials interacting simultaneously.
This is the baseline challenge IMD works against every single day, and Niyogi is emphatic that the department deserves more credit than it typically receives. IMD has already begun using AI and machine learning to downscale temperature data and produce experimental thermal maps.
"It is one thing for a research organisation to just say, I have done it," Niyogi points out. "Another thing entirely for an operational entity to deliver a product that people rely on, that has to be available 24/7, quality controlled, with protocols people can interpret." IMD, he says, has the technology ready. What follows is the harder work of operationalisation.
From Coarse Grids to Your Neighbourhood
The technical journey from continent-scale forecasts to neighbourhood-level intelligence involves solving a fundamental data scarcity problem. Ground sensors are expensive to maintain and suffer from representativeness issues. A temperature sensor at the front of a building will read differently from one at the back. Satellites offer spatial breadth but are blocked by cloud cover and can only revisit a location every few days.
IAIRO's approach is to fuse these imperfect inputs using AI, and the toolkit being deployed is sophisticated. For temperature hyperlocalisation, the primary workhorse is the Super Resolution Convolutional Neural Network (SR-CNN) — the same class of algorithm that sharpens low-resolution images — applied to coarse weather maps. Reinforced by land-use data (green cover lowers temperature; concrete raises it), the model iteratively refines its output to produce high-resolution thermal maps from sparse inputs.
Where satellite imagery has cloud gaps, IAIRO is turning to diffusion models, a technique that learns the statistical structure of an image well enough to reconstruct missing data from the surrounding context. Think of it as the AI equivalent of restoring a photograph from partial information.
For deeper physical problems, such as understanding how surface temperature measurements translate to subsurface soil behaviour, the organisation is using Physics-Inspired Neural Networks (PINNs).
These models don't just find data correlations, but they're constrained by physical laws, enabling physically coherent data interpolation rather than just statistical fits. IAIRO is exploring quantum AI to optimise these equations, but Niyogi notes this is likely in one to two years, not immediately.
The City Digital Twin: Infrastructure for a Smarter India
The longer-term vision ties all of this together in the concept of a city digital twin — a live, physics-informed computational model of an urban environment that ingests real-time data from traffic cameras, radiation sensors, and weather instruments to simulate scenarios before they happen.
The use cases are tangible. Which canals should municipal workers prioritise clearing before the monsoon? Which roads are likely to flood? Where should heat shelters be positioned to serve the most vulnerable people during a heat wave? These are decisions currently made on experience and intuition. A functioning digital twin could make them data-driven.
"You can still have limited information, feed it some rules, and it can learn and create an excellent dashboard of evidence scenarios," Niyogi says.
The Ecosystem Gap
Despite progress, the main obstacle isn't algorithms or data but the research ecosystem. India lacks a culture of long-term investment in basic research, unlike countries like the US. Private industry favors quick, superficial AI, labeled "two-minute noodle AI" by Niyogi, prioritizing rapid returns over depth.
"A lot of information we have is useful, but very little of it is usable," he says. "The translation from useful to usable requires an intense understanding of why data is being created and how people are going to use it."
The near-term roadmap for IAIRO, therefore, focuses on partnerships with disaster management authorities, urban municipal bodies, the agriculture sector, and energy planners to co-develop the "data to decision" pathways that convert atmospheric intelligence into actionable guidance. The organisation is not yet delivering live products to these sectors, but the groundwork is being laid.
"We have confidence in the technology and science that is being built," Niyogi says. "Our path currently is to build these partnerships. As the implementation pathway solidifies, we will start seeing results on the ground."
With India's cities heating up and its monsoons growing increasingly erratic, the pressure to move from potential to delivery has never been greater. The science, it seems, is ahead of the system built to receive it. Closing that gap is now the real work.
Your reaction
Discussion
What's your take on this story?
No comments yet — be the first to share your view.