In the field of weather forecasting, new technological innovations are quietly emerging. The DeepMind team, a subsidiary of Google, has developed an artificial intelligence weather prediction program called GenCast, which has surpassed the performance of the current leading weather forecasting system - the European Centre for Medium-Range Weather Forecasts (ECMWF) ENS prediction system. According to research, GenCast improves the accuracy of daily and extreme weather event predictions by 20% compared to ENS.
Image Source Note: Image generated by AI, image licensed by service provider Midjourney
GenCast's main advantage lies in its fast and efficient predictive capabilities. Traditional weather forecasting relies on complex physical models that require hours of computation on supercomputers. In contrast, GenCast can complete a 15-day weather forecast in just eight minutes by training on 40 years of historical meteorological data from 1979 to 2018. It can predict global weather changes in a 28-kilometer square area, with updates every 12 hours.
In a comparative experiment, GenCast outperformed ENS in predicting tropical cyclones and their landfall locations, especially in the prediction of extreme events, which has significant reference value for energy and related industries. Although GenCast is currently used as an auxiliary tool for traditional weather forecasting, its accuracy and efficiency mark an important turning point in weather prediction technology.
Currently, GenCast is one of Google's latest achievements in advancing AI technology for weather forecasting. Last year, Google also launched NeuralGCM, which combines AI with traditional physical models, and GraphCast, which focuses on single best predictions. GenCast further enhances the reliability of forecasts by generating over 50 weather predictions and assigning probabilities to different weather events.
The meteorological community has expressed excitement about this technological advancement. A chief forecaster from the UK Met Office described it as "exciting work," while a spokesperson from the European Centre for Medium-Range Weather Forecasts acknowledged it as "an important development." However, some experts point out that despite GenCast's encouraging performance, attention must be paid to whether it possesses sufficient physical realism to handle the uncertainties in weather forecasting.
Although AI weather prediction technology shows great potential, experts indicate that there is still a long way to go before it can fully replace traditional physical models, and further research is needed to address related scientific issues.
Key Points:
🌦️ GenCast is an AI weather prediction program developed by Google, with accuracy surpassing the traditional ENS system.
⏱️ GenCast only requires 8 minutes for predictions, greatly enhancing the efficiency of weather forecasting.
🧪 Despite GenCast's outstanding performance, experts remain concerned about whether it can completely replace traditional physical models.