The devastating floods that claimed lives and displaced thousands of people are a grim reminder of how unpredictable and destructive water can be. This vulnerability led researchers at the Indian Institute of Technology (IIT) Bombay to develop a sophisticated two-pronged artificial intelligence (AI) system designed to accurately predict not only where the next major flood will occur, but also how deep the water will rise.
By combining satellite radar data and advanced machine learning, the researchers created a high-resolution mapping system that identifies flood-prone zones with more than 93% accuracy. It is earmarked to cover an area of 55,000 square kilometers from Tadri in Uttara Kannada district of Karnataka to Kanyakumari along the coast of the Western Ghats in south India. This new system promises to be a powerful tool to protect millions of people living in India’s most vulnerable coastal areas.
Traditionally, flood forecasting has relied on extensive historical rainfall data and physical sensors. A team from IIT Bombay, researchers Dr. Kashish Sadhwani and Professor TI Eldo, turned to pattern recognition and analyzed several key conditioning factors.
Interestingly, this study found that surface runoff was a more important predictor than rainfall itself.
Dr. Sadhwani explains, “Rainfall is the main driver of flood events, but it does not directly lead to inundation at a particular location. Surface runoff represents the integrated hydrological response of a landscape, capturing the combined effects of rainfall intensity, soil moisture, land use, infiltration capacity, and drainage characteristics.”
To process this data, the team used a two-step process. First, a classification model identifies whether an area is at risk. The regression model then estimates the depth of water that is estimated to accumulate and creates a continuous map of potential inundation. To train the model, the team used the European Space Agency’s Sentinel-1 synthetic aperture radar (SAR) satellite imagery. This image can effectively pass through monsoon clouds to record observations. By comparing historical pre-flood and during-flood images, the model was able to recognize dark shades in images that indicate standing water.
Ultimately, the model will provide high-resolution mapping down to a 30-meter grid, but currently operates with a margin of error (RMSE) of approximately 0.99 meters. While variations of about 1 meter are important in urban planning, Dr. Sadwani says the current value of the system lies in its breadth and speed.
“This model is designed for rapid regional-scale flood assessments, offering high computational efficiency and the ability to rapidly generate flood extent and depth information over large areas,” says Dr. Sadhwani. “This makes it particularly valuable for early stage planning, prioritization of vulnerable zones, and emergency response support.”
The system is also currently focused on terrain with slopes less than 7%. As stated in the study, this was a deliberate methodological choice: “To account for the possibility of water movement during image acquisition and to ensure accurate flood inundation mapping with SAR images, inundation depth calculations were limited to areas with slopes less than 7%.”
Additionally, in steeper terrain, radar signals are subject to geometric distortions such as shadows and layovers, and water movement during acquisition can lead to inaccurate depictions of flood extent and depth. Therefore, applying a slope threshold ensures that the derived inundation depth is reliable and physically consistent.
For Kerala and Karnataka, this tool is potentially transformative. In areas where water collects in clay soils and is dominated by low-lying coastal plains, 30-meter resolution maps can help local authorities know exactly which hospitals, schools and roads are most likely to be submerged.
Dr. Sadhwani emphasized that the framework can “identify areas at risk of flooding and guide urban planning and land use management” and “plays an important role in disaster preparedness and response by enabling authorities to effectively allocate resources and prioritize vulnerable areas in evacuation and relief operations.”
Although the current study focuses on the southern west coast of southern India, the researchers believe the framework is ready to be scaled up to complex urban hubs such as Mumbai and the east coast. However, moving into these areas requires consideration of new variables and may require recalibration and retraining of the model.
“In coastal environments, additional complexities arise, including the effects of tidal fluctuations, storm surges, sea level changes, and drainage backflow,” explains Dr. Sadhwani. “By incorporating these coast-specific parameters into the existing framework, this methodology can be effectively adapted.”
As climate change increases the frequency of extreme weather events, this AI system offers more than just predictions. This will help disaster management teams prepare and plan to reduce flood-related losses and improve community resilience.
