Google DeepMind has developed an AI model called GenCast that can accurately predict the weather, competing with traditional forecasting methods. The model outperformed a leading forecast system in 2019 and showed promising results when tested on data from 2019. GenCast uses machine learning to recognize patterns in historical weather data and produces ensemble forecasts.
GenCast operates at a resolution of 0.25 degrees, which is comparable to the European Centre for Medium-Range Weather Forecasts’ (ECMWF) current system. However, traditional models like ENS require more computational power and time to produce forecasts. GenCast can produce one 15-day forecast in just eight minutes using a single Google Cloud TPU v5.
The model’s efficiency has raised concerns about its environmental impact, but further research is needed to determine its sustainability. Despite this, GenCast’s developers believe it holds significant promise for improving weather forecasting. The model’s predictions can be used to assess wind power availability and provide more accurate forecasts of tropical cyclone tracks and extreme weather events.
To verify the results, the code for GenCast has been released as an open-source model. The meteorological community is cautiously optimistic about AI’s potential in improving forecast accuracy. As GenCast continues to evolve, its developers aim to build widespread trust and confidence among practitioners.
GenCast’s achievements mark a significant milestone in weather forecasting, and further research will be needed to determine its full potential.
Source: https://www.theverge.com/2024/12/7/24314064/ai-weather-forecast-model-google-deepmind-gencast