This IIT Team Built an AI System That Can Predict Floods Before the First Drop Falls |
What if cities could rehearse the monsoon before the first drop of rain falls?
What if a municipal officer could see, in real time, which neighbourhood would flood first, how deep the water would rise, where traffic would choke, and which intervention could reduce the damage?
Every monsoon, roads disappear under water, traffic comes to a standstill, and officials rush to respond once the damage is already done. For years, this has been how urban flooding has been handled in India.
On a campus in Gujarat, a team is bringing together physics, artificial intelligence and the everyday realities of Indian cities to help officials anticipate these problems before they unfold.
At IIT Gandhinagar, the AI Resilience and Command (ARC) Centre is trying to help cities prepare for floods before the damage begins.
Why cities need more than another flood report
Every monsoon, disruption feels inevitable. Roads turn into rivers. Underpasses vanish. Commute times double. The scramble that follows is routine.
For Prof Udit Bhatia, Principal Investigator of the Urban Flooding Module under the AI Centre of Excellence (AICoE) for Sustainable Cities at IIT Gandhinagar, that repetition raised a question: in an era defined by artificial intelligence, why were cities still reacting instead of anticipating?
“We are in the age of AI, and while India is positioning itself as a leader, we chose to focus on one of our most fundamental challenges: urban flooding,” Prof Bhatia tells The Better India.
His Machine Intelligence and Resilience (MIR) Lab had already published in journals such as Nature Cities, including a study modelling how Surat’s flood dynamics would shift under riverfront development. The research also informed planning discussions.
The team now wanted to take this research beyond academic validation and make it usable for cities.
“We need to demonstrate these solutions. Develop algorithms and stress-test them. Make them usable for cities,” he says.
The need for such a system became clear during the team’s fieldwork on flooding linked to the Tapi River in Surat. City stakeholders wanted a tool that could show flood risks in real time and help them decide what action to take, rather than another report explaining the problem after the damage.
That realisation led to the ARC Centre, a 2,000-square-foot command space at IIT Gandhinagar’s Research Park. The idea was simple: research needed to move beyond papers and presentations to support people making difficult decisions on the ground.
The centre brings researchers and city stakeholders together to build AI tools suited to Indian cities and their realities, rather than relying on standardised models developed elsewhere.
As Prof Bhatia puts it, “Indian cities are very different in their typology and architecture. We cannot just take models from the West and apply them here.”
That need for contextual intelligence shaped the Centre’s defining innovation: physics-guided AI.
How physics keeps the AI grounded
Flood models have traditionally used physics-based simulations to show how water moves through rivers and across land. These models can be accurate, but they often take a long time to run and require considerable computing power.
“In a flooding situation, you cannot wait for days,” says Dr Vivek Kapadia, former Secretary, Government of Gujarat, and Professor of Practice at IIT Gandhinagar. “You need a solution in a jiffy.”
AI models offer speed, but their answers may overlook the physical laws that govern how water actually moves. The ARC team decided to combine the strengths of both........