I once observed one of the most advanced machine climate models globally during a foggy evening at the European Centre for Medium-Range Weather Forecasts. The computer in the adjacent room processed trillions of variables every minute, displaying routine status messages on the monitor in front of me. It utilized physics equations to convert recent wind observations into a detailed image of the sky for the next 14 time intervals over two hours. This data is fundamental for worldwide scientific predictions aiding in activities such as selecting appropriate clothing, aircraft navigation, or seeking shelter during inclement weather. The current forecasting system, refined over five decades, showcases remarkable technological and scientific collaboration.
Recently, researchers from Google DeepMind released a report in Science hinting at a potential shift in the dominance of this system. The emergence of a new climate modeling approach driven by artificial intelligence could revolutionize the field. Google’s GraphCast system relies on a 39-year historical data analysis rather than traditional physics-based algorithms. This innovative system operates on a single computer, delivering results in less than a minute compared to the previous two-day processing time. It represents a shift towards “machine learning-based weather prediction,” showing promising early outcomes in contrast to conventional “numerical weather forecasting.”
While immediate improvements in weather forecasts may not be imminent, the prospect of Google surpassing the capabilities of established institutions like the Met Office looms on the horizon. The efficiency and speed of AI technology are surpassing existing standards, sparking excitement among experts. This advancement hints at enhanced accuracy and precision in weather predictions, potentially extending from short-term forecasts to longer-term climate models.
The methodology employed by GraphCast reflects a nod to the past. In 1922, English scientist Lewis Fry Richardson envisioned a future where meteorologists could rapidly solve weather equations, eliminating the need to manually compare past and present conditions for future predictions. Despite Richardson’s lack of access to modern technology like Google, GraphCast leverages a comprehensive dataset provided by ECMWF spanning from 1979 to 2017. While GraphCast can predict weather six hours ahead based on recent weather conditions, the ECMWF’s intricate system remains essential for data processing.
Furthermore, the collection of observational data through satellites, weather stations, and various monitoring devices by national meteorological organizations plays a crucial role in weather prediction. The recent Science publication funded by Google DeepMind highlights the evolving landscape of weather forecasting, with implications for the global climate research community. The reliance on personal data for weather predictions marks a departure from traditional methodologies, potentially reshaping the industry.
GraphCast’s dependency on this innovative system underscores a significant shift in weather prediction practices. While private entities have attempted similar ventures in the past, Google’s foray into weather forecasting signifies a notable departure from conventional approaches. The potential transition to this new system raises questions about prioritizing private interests over public welfare, particularly in a domain as critical as weather forecasting.
Undoubtedly, improved weather predictions hold immense value. However, the distribution of benefits from these advancements may not align with the collective interest of safeguarding the Earth’s atmosphere and mitigating the perils of environmental crises.
