New Delhi: Researchers at IIT Delhi have developed India’s first national AI-enhanced river flow modeling system that significantly improves the accuracy of river streamflow predictions across the country, a critical advancement for flood forecasting and water resource management.

Published in the prestigious journal Water Resources Research (https://doi.org/10.1029/2024WR039792), the study led by PhD Scholar Bhanu Magotra and Prof. Manabendra Saharia from IIT Delhi's Department of Civil and Environmental Engineering and Yardi School of Artificial Intelligence demonstrates how integrating artificial intelligence with traditional hydrological models can overcome longstanding challenges in water cycle prediction.

Accurate river flow information is critical for water resources management, including irrigation scheduling, flood risk reduction, and reservoir operations.  However, large-scale hydrological models often produce significant uncertainties in streamflow estimates at local scales unless extensive basin-specific calibration is performed. Such calibration is computationally expensive and challenging to implement across a country as hydrologically diverse as India.

Addressing this challenge, the IIT Delhi team developed a novel integrated framework that combines the physical consistency of land surface models with the computational efficiency of Artificial Intelligence (AI). The system employs a two-stage Long Short-Term Memory (LSTM) neural networks, a type of AI particularly effective at recognizing patterns over time, to systematically correct river streamflow from the Indian Land Data Assimilation System (ILDAS). The AI model was trained on decades of streamflow data from hundreds of river gauge stations across India maintained by the Central Water Commission (CWC) under the Ministry of Jal Shakti.

By training on multi-decadal data and incorporating 27 different meteorological and geophysical attributes, the LSTM models learned to account for complex local conditions that traditional models struggle to represent. The model improved daily river flow predictions in nearly 95% of the basins, raising the national median value of Kling Gupta efficiency (KGE) from 0.13 to 0.60. Post-monsoon performance improved by 81.7% in error reduction, while flood peak predictions became 25% more accurate in both timing and magnitude.

The model was developed on IIT Delhi’s supercomputer. The research is particularly notable for its handling of India's hydrologically diverse catchments, which include heavily regulated rivers with reservoirs, arid regions, and monsoon-dominated basins.

This work presents a significant step forward in integrating traditional hydrological science with modern artificial intelligence. This technology can be used to develop river basin digital twins, supporting informed decision making in India.

                                                                                                  ******