Global prediction of extreme floods in ungauged watersheds
In this paper, the authors evaluate the extent to which artificial intelligence (AI) trained on open, public datasets can be used to improve global access to forecasts of extreme events in global rivers. On the basis of the model and experiments described in this paper, they developed an operational system that produces short-term (7-day) flood forecasts in over 80 countries.
The authors recommend that the best way to improve flood forecasts from both data-driven and conceptual modelling approaches is to increase access to data. Hydrological data are required for training or calibrating accurate hydrology models, and for updating these models in real time (for example, through data assimilation).
Explore further
![Cover](/sites/default/files/styles/por/public/2024-03/Cover_Nature.png.jpg?itok=NbJLOCJ6)