Predicting the weather of the future could be in the hands of artificial intelligence. A study by the European Center for Medium-Range Weather Forecasts (ECMWF) has shown that Google DeepMind's weather AI is the most accurate system in the world in its 10-day calculations.
The AI model has already been named GraphCast and promises medium-term weather forecasts of "unprecedented precision" thanks to its rigor and speed, two factors tremendously valued in the world of weather forecasting and High Resolution Forecasting. (HRES).
One of the clearest examples of this effectiveness of GraphCast is that the AI predicted nine days in advance in September that Hurricane Lee would make landfall in Nova Scotia, something that traditional forecasting methods only indicated with about six days of margin and less. consistency regarding the time and place of landfall.
The data collection and analysis capabilities of this artificial intelligence can identify dangerous weather phenomena without being trained to find them and collect information that could save lives and livelihoods, even more so as the weather becomes more extreme and unpredictable.
Matthew Chantry, machine learning coordinator at ECMWF, considers the arrival of GraphCast a change of era: "There is probably more work to be done to create reliable operational products, but this may be the beginning of a revolution," explains the expert.
There is still room for improvement for this AI weather system in measuring cyclone intensity: “At the moment, it is an area where GraphCast and machine learning models are still a little behind physical models” , Chantry assured, although it is clear that a new era has arrived in weather prediction.
