Researchers develop AI model to support long-range flood forecasting

Sydney (ANTARA) — An international team of researchers has developed an artificial intelligence (AI) model to support long-range flood forecasting, including by helping communities receive disaster warnings several weeks before an event.

The researchers used two decades of historical precipitation and river-flow data from the DuPage River watershed in Illinois, the United States, to develop a deep-learning model incorporating an attention mechanism to identify rainfall and river-flow patterns, according to a statement released by the University of Technology Sydney (UTS) in Australia on Tuesday (13 October).

By combining predictions from several models, the team improved forecast accuracy. The ensemble deep-learning model reduced prediction errors by about 38 per cent, while the proposed fusion framework improved predictive performance by around 47 per cent, according to the study published in the Journal of Hydrologic Engineering.

The approach remained highly accurate for forecasts up to 12 weeks ahead, said Professor Amir H. Gandomi of UTS, one of the study’s authors, emphasising that the additional warning time could help communities prepare for floods.

The findings could have implications for flood-risk management, including disaster preparedness, infrastructure resilience and water-resource management, the research team said.

“Ultimately, the goal is not simply to predict floods more accurately, but to turn better predictions into better decisions,” Gandomi said.