Google has made AI-powered nowcasting available across Africa, helping users to understand exactly when and where they should expect precipitation over the next few hours.
The forecasts are made possible through advancements in Google Research’s AI-powered nowcasting model MetNet, which uses satellite data and ground observations to produce precipitation forecasts in data-sparse regions of the world.
Google’s technology can predict global precipitation with high accuracy within a 5km radius every 15 minutes for the next 12 hours — all under one minute.
One critical component for weather forecasts is dense ground radar, which is not available for the majority of the world where billions of people live. The need for new approaches and better weather forecasts is especially pronounced in Africa, where ground observations are limited and global weather models tend to exhibit low skill. For example, North America has 291 radar facilities to track weather, while Africa has 37.
To scale nowcasting globally, Google research teams came up with a new approach, using globally available satellite observations in our nowcasting models.
Using AI and satellite observations, the model was able to fill in the gaps, even in areas where radar was unavailable, overcoming previous limitations in data sparse regions.
It is also Google’s first model to directly incorporate observations from a multitude of satellites.
Nowcasting on Search is now live for users in Africa, displaying for weather searches across the continent with more precision.