Advances in artificial intelligence (AI) are flooding the field of weather forecasting and natural disasters. While 2024 was marked by many spectacular dramas related to climate change, 2025 promises to be a year of major innovations for much better predictions.

2024: when climate change challenges prediction

Cyclone Chido in Mayotte, spectacular flash floods in Valencia, historic drought in Colombia, large-scale floods in West Africa, Southeast Asia and China, and now a devastating fire in Los Angeles… 2024 appears to be a pivotal year in the dramatic effects of global warming which, let us remember, has resulted in an average increase in temperatures across the globe of 1.5 degrees. And 2025 is off to a bad start.

Predicting weather trends accurately and in advance is therefore becoming a major challenge, including on an economic level, given that the damage caused by disasters has amounted to hundreds of billions of dollars in recent months.

Artificial Intelligence has demonstrated its potential in terms of forecasting/anticipation in recent years. It is therefore natural that it comes to the aid of weather and climate experts, with the hope of saving thousands of lives and reducing the gigantic bill of climate damage.

How AI is improving weather forecasting

Overall, it is thanks to its rapid processing, collection and analysis capabilities of massive and complex data that artificial intelligence (AI) can contribute to a turning point in the weather prediction of natural disasters. Here is why precisely:

  • Increased accuracy and speed

Traditional weather forecasting models are based on complex physical equations, based on the collection of multiple data, with variable reliability, giving rise to gigantic volumes of calculations. Laure Raynaud, researcher at the National Center for Meteorological Research (CNRM), explained at the summit: Big data and AI ", organized in October 2024 in Paris, that the use of AI could clearly "change the game":

“AI would make forecasting calculations faster and more accurate, especially on local scales. Where a current model takes an hour to make a forecast, AI could do it in seconds!”

  • Deep learning to help with medium and long-term forecasting.

AI is now able to bridge the gap between immediate forecasts (from a few minutes to 24 hours) and longer-term ones, or even in the very long term (the decade). In integrating real-time data (satellite imagery, sensors), it produces Hyperlocal forecasts with unmatched accuracy.

Of the learning models have already been tested integrating an ever-increasing and ever-more-updated volume for forecasts that are both more reliable and more local.

Able to exploit “convolutional neural networks” ", which are notably capable of analyzing images, AI will be able to quickly model more precisely the birth and expansion of extreme phenomena, such as cyclones or storms. Thanks to deep learning algorithms, already defined, it becomes possible to simulate rare but critical scenarios, such as extreme heat waves or tropical cyclones, to better anticipate their occurrence and intensity.

  • For the populations, the use of AI must enable the development of effective warning systems which warn populations in real time of imminent dangers such as earthquakes, floods or forest fires, allowing better preparation and even their evacuation.
  • Cost reduction.

This is not a trivial subject. One of the problems with weather forecasting is its cost, its continuous need for investment: personnel, data capture equipment, satellites, stations, drones, radars, etc. and calculation capacities have an astronomical cost. For Météo-France alone, the expenditure flirts with 300 million euros per year!

Because it is based on a deeper data mining, which she is constantly learning and consuming less computing resources than traditional physical models, AI makes forecasts faster and less expensive without this expected saving yet being evaluated.

Four major innovations already deployed

Innovations in the forecasting sector are already numerous. Here are four examples of how AI has contributed to weather and natural disaster forecasters that have emerged since 2023:

  • Flood Hub, from Google:

Deployed in 80 countries, this tool uses combined hydrological and flood models to predict floods up to 7 days in advanceIt has helped protect around 460 million people, particularly in Africa and South-East Asia, where forecasting infrastructures are often poorly performing.

  • GenCast, from DeepMind:

This AI model unveiled at the end of 2024 has pushed the limits of weather forecasting by achieving a rate of accuracy of 97 % on tests carried out with historical data, at 15 days. It can generate detailed forecast in just 8 minutes, compared to several hours for traditional systems.

  • Earth-2, from NVIDIA:

Thanks to its microservices like FourCastNet NIM, this platform created by Nvidia, the computing giant, has accelerated climate simulations by up to 500 times, thanks in particular to the accelerated exploitation of 20 times more data, allowing more precise anticipation of natural disasters such as hurricanes or forest fires, with interactive visualization.

  • Specific applications

   – Japanese researchers have developed a device capable to observe typhoons from the eye of the storm, significantly improving forecast accuracy.

   – In Australia, a Spatial AI was used to track the evolution of wildfires, providing better planning of interventions.

Why you need to act quickly

In 2024, several major natural disasters hit the world and Europe, including France, causing considerable human and material losses. Here are some striking examples that justify the urgent use of AI for better predictions and a more effective warning system.

  • Floods in Valencia, Spain : This event was the most serious in Europe, with torrential rains reaching 600 mm in one day. These floods caused the at least 200 people died and damages estimated at $11 billion.
  • Cyclone Chido in Mayotte (France) : This cyclone devastated the French island in December 2024, destroying infrastructure and causing significant human losses: around forty dead and 2100 injured at least. Reinsurance companies estimate the cost of material damage between 650 and 800 million euros.
  • Floods in Germany and neighboring countries : In June and September 2024, these floods caused financial losses of more than 9 billion dollars. Information on human losses has not been aggregated.
  • Los Angeles fire (USA) : the most devastating fire in the history of the United States, which appeared on January 7, 2025, caused at least 24 dead and no less than $275 billion in damages.

One challenge remains for now: the mass and quality of available dataThey could quickly curb or even stop the effectiveness of the contribution of AI and therefore require both more collection and international and institutional cooperation.

visual: Weather icon with hands typing on a laptop with cloud environment change technology icons floating above, indicating the use and management of AI in data visualization design for web development. source ADOBE stock