With a degree of accuracy that is largely on-point and a range that is ahead of traditional weather forecasting, Google’s AI-centric brain trust DeepMind is putting the old methods to shame with its advanced capabilities.
The machine learning model, known as GraphCast, is capable of predicting weather conditions of up to 10 days in a fashion that is better, quicker, and more energy-efficient than the tools responsible for bringing the same information to our devices today.
In a study published on Tuesday, Google’s researchers wrote that the discovery “marks the turning point in weather forecasting.”
The current forecasting model involves “numerical weather forecasting” (NWP), which processes weather data based on the principles of thermodynamics, fluid dynamics, and other atmospheric sciences. Not only is the process known to be intricate, but the cost and requirements for the procedure are resource-intensive as well.
READ: How to help improve weather forecasting using your smartphone
While the traditional method has its eyes set on the dynamics of molecules in the atmosphere, GraphCast puts more importance on historical data. This means that the AI’s predictions take into account information from the past in making present and future predictions.
Although using GraphCast will still require many sciences that relate to computers, the procedure is comparably simpler in the degree and the number of computations required to generate results.
By tapping on both present and past data, GraphCast can predict the future. For instance, by using present weather data and data from six hours ago, the AI can calculate what the weather conditions would be like six hours into the future. Each calculated result will then be fed back into the model to subsequently produce longer-term forecasts.
In comparison to the present model used in medium-range weather prediction, or HRES, the Google team said that GraphCast outperformed 90 percent of the test target.