Ai And Weather: How Artificial Intelligence Improves Forecasting
Everyone has noticed it at some point: meteorology is not an exact science! The reliability of weather forecasts sometimes leaves much to be desired, but it could improve thanks to artificial intelligence. In less than two years, the rise of AI has disrupted global meteorological services. We invite you to discover how artificial intelligence is enhancing weather forecasts and climate modeling.
Can AI predict the weather?
In November 2023, an article published in the journal Science caused quite a stir in the media: GraphCast, the artificial intelligence tool developed by Google Deepmind, succeeded in surpassing almost all existing weather forecasting tools.
Since then, the use of AI has appeared to the general public as the miracle solution for obtaining much more reliable weather reports. Moreover, Google is not the only company working on new AI weather models. Microsoft, IBM, and Huawei have also joined the race for the best weather forecasts.
However, experts from Météo France have tempered the promises of artificial intelligence in the field of meteorology. The national meteorological service acknowledges that AI is a valuable tool. In fact, they have been using it for years! Since the 1990s, meteorologists have employed artificial intelligence methods such as machine learning to correct imperfect forecasts. But the forecasters at Météo France also highlight the limitations of this tool...
What are the limitations of AI for weather forecasting?
Even with new artificial intelligence resources, it is still very difficult to predict the weather in the medium and long term. Artificial intelligence does not allow us to see further than traditional models.
If you want to know the weather trends for the next 3 months, to plan your next vacation for example, be aware that even AI models are not yet capable of making perfectly reliable seasonal forecasts.
How can AI improve weather forecasting?
Despite its limitations, artificial intelligence already provides valuable assistance to forecasters on a daily basis.
Combined with traditional forecasting models, which rely on physical equations, AI-based tools enable:
• making the daily weather forecast increasingly accurate.
• conducting ever more detailed analysis of ongoing climate change and performing more precise climate simulations.
• better anticipating extreme weather events.
By comparing reality with previous forecasts, artificial intelligence is capable of correcting forecasts retrospectively to make them increasingly accurate. AI tools do not rely on atmospheric physics but on statistical learning of the relationships between today's weather and tomorrow's.
Artificial intelligence can also help weather forecasters process the massive amount of data they receive each day, including hundreds of forecasts and real-time weather observations.
Finally, one of the objectives of Météo France meteorologists is to use AI to better predict natural disasters and weather phenomena that could cause damage, such as severe storm structures. One of their main missions is indeed to issue weather reports and alert messages for citizens and state services.
Will AI replace weather forecasters?
In the face of the current rise of AI-based tools, should Météo France forecasters fear being replaced by artificial intelligences? Even if the weather forecast could one day be produced by an AI, this is not going to happen in the near future.
For now, artificial intelligence is far from being able to replace traditional forecasting models. AI models still have a lot to learn to refine their forecasts, especially on a local scale, and this will take time.
According to Marc Pontaud, Director of Higher Education and Research at Météo France, the two weather forecasting models, by physical equations and by artificial intelligence, will likely coexist over the next ten years.
In the longer term, everything will depend on the progress that AI models will have made. But, for weather forecasting as in other fields, the complementarity between human expertise and the computing power of machines will undoubtedly be the most effective approach.
