AI could help predict floods that are difficult to predict using traditional methods

Artificial intelligence (AI) models can improve the accuracy of flood predictions, a new study published in 2016 has found. Nature. The system has been shown to be as accurate as or better than current leading methods and has the potential to provide early warning of large-scale flooding events.

Human-induced climate change is increasing the frequency of floods in some regions. Current forecasting methods are limited by their reliance on flow meters (observation stations along rivers) that are not evenly distributed around the world. Therefore, predicting ungauged rivers is difficult, and their negative effects are mainly felt in developing countries.

Gray Nearing and colleagues developed an AI model trained using 5,680 existing instruments to predict daily flow over a seven-day forecast period in a non-metered watershed. The AI ​​model was then tested against the Global Flood Awareness System (GloFAS), a worldwide software for predicting floods in both short-term and long-term scenarios.

The AI ​​model was able to provide five-day-ahead flood forecasts that were as reliable or more reliable than the current system’s same-day forecasts. Additionally, the AI ​​model’s accuracy when predicting extreme weather events with a five-year return window was as accurate as or better than his GloFAS predictions for events with a one-year return window.

These results demonstrate that AI models can provide flood warnings with longer notification periods than previous methods for both small and extreme events in ungauged watersheds, improving confidence in developing regions. This suggests that access to flood forecasting could be improved.