Since March, Google's artificial intelligence-based flood forecasting system has been in operation, providing timely warnings of potential flooding in almost every corner of the globe, including the current situation in Hungary. According to a network researcher at HUN-REN SZTAKI, it would also be beneficial to link precipitation patterns predicted by medium- to long-term climate models with this artificial knowledge, thereby enhancing preparedness for future floods.
Artificial intelligence is making life easier for nearly half a billion people by predicting river floods, sometimes up to 5–7 days in advance. This gives us more time and opportunities to respond effectively. Google has put AI to work in the service of flood protection and prevention, as flooding is the most common natural disaster.
This forecasting system is a tremendous help to a significant portion of the world, especially in developing countries where hydrological forecasts are of much lower quality than in more developed regions. Google's algorithm learns from data collected from approximately 6,000 measurement points. "If we add up the lengths of the available time series at each measurement point, the total amounts to 152,000 years. Thus, an exceptionally large data collection process preceded the training and development of the model," said Ferenc Béres, a network researcher at HUN-REN SZTAKI.
The data were derived from atmospheric and near-surface measurements, including temperature, precipitation, irradiance, and barometric pressure. In addition, satellite images were scanned using artificial intelligence to extract as much accurate data as possible. Other important input data included the slope of certain areas and regions, the type of rock and soil quality, the agricultural activity of the area, and its proximity to a nearby city. Google's process takes all these factors into account to predict the expected water flow in a given area, which also helps forecast the severity of the flood protection situation.
The HUN-REN SZTAKI researcher added that in more developed regions, where there is high-quality infrastructure for recording and monitoring data, artificial intelligence provides much more accurate forecasts than in areas with less data available. However, predicting floods is not an easy task, not least because many rivers lack flow gauging stations.
According to Ferenc Béres, if precipitation patterns predicted by medium- and long-term climate models are fed into such a machine learning process, artificial intelligence can provide guidance on how the risk of flooding might increase in specific areas and settlements in the future. For example, if increased precipitation is projected for certain regions over a 50-year period, we could receive feedback on how much the severity of those floods might rise within that timeframe.
Google's artificial intelligence development currently assists nearly 460 million people across 80 countries, including Hungary. For information on flood forecasting in Hungary, click here.