Developing a Flood Inundation Forecasting System for India.

Flood forecasts are a crucial component of flood hazard mitigation strategies and forecasted flood inundation maps are essential for transitioning from forecast information to decision-making to reduce flood risk. A national medium-range streamflow forecasting system has been developed with ILDAS as its physical modeling core and integrated with an AI-based postprocessor. The forecasting system consists of integrated Noah-MP and mizuRoute that produce daily streamflow forecasts with 1-5 days lead time at more than half a million streams across the country. While the system is computationally very efficient, extending it to generate inundation forecasts remains a drawback, as vector-based routing models do not produce inundation maps. To bridge this gap, we integrate TRITON, a GPU-accelerated 2D hydrodynamic model with improved terrain representation, into the system to simulate flood inundation. A GPU-based hydrodynamic model has computational superiority over a CPU-based model and hence reduces forecast generation time, which is a major bottleneck in inundation forecasting systems. In this framework, streamflow forecasts from the existing system are taken as input to the GPU-based model, which generates inundation forecasts with lead times of 1-5 days across India, along with streamflow and water level forecasts. Initial results show reasonable forecast skills when compared with observed water level data and SAR-based flood maps. The system demonstrates its potential to support operational flood preparedness and disaster risk management, and is also an important step towards building an impact-based flood forecasting system for India.