AI Weather Models Lag Behind Traditional Models in Predicting Record-Breaking Events, Study Finds

AI Weather Models Struggle with Record-Breaking Extreme Weather Forecasts

Computer models using artificial intelligence (AI) fail to match the accuracy of traditional physics-based climate models when predicting record-breaking weather events, according to a new study published in Science Advances.

While AI models have outperformed traditional models in some weather forecasting aspects, researchers found they still "underperform" when simulating extreme events that shatter existing records. The study authors tested both AI and traditional models against thousands of record-breaking hot, cold, and windy events recorded in 2018 and 2020.

The analysis revealed that AI models consistently underestimate both the frequency and intensity of these unprecedented weather events. Prof. Sebastian Engelke, a study author and professor at the University of Geneva, told Carbon Brief that the findings serve as a "warning shot" against rapidly replacing traditional models with AI alternatives for weather forecasting.

Why Accurate Extreme Weather Forecasts Matter

Extreme weather events—including floods, heatwaves, and storms—cause hundreds of billions of dollars in annual damages worldwide. These disasters destroy cropland, damage infrastructure, and tragically result in loss of life. Governments rely on early warning systems to alert the public and mobilize disaster response teams, which have proven effective in minimizing damage and saving lives.

For decades, scientists have used numerical weather prediction models to simulate weather days or weeks in advance. These models are built on complex equations derived from fundamental physics laws and decades of climate research, earning them the label "physics-based" models.

How AI Models Differ from Traditional Forecasting

AI-based climate models have gained traction as an alternative to physics-based approaches. Instead of relying on physics, these models use a statistical method:

• They are trained on large datasets of historical weather data.
• The training process teaches the model to recognize patterns and make predictions based on past events.
• For new forecasts, the AI model applies these learned patterns to generate predictions.

Advantages of AI weather models include:

• Lower computing power requirements (no need to run thousands of equations).
• Demonstrated superior performance in some weather forecasting tasks compared to traditional models.

Key limitations of AI models:

• They depend heavily on training data and are constrained by its range.
• They struggle to simulate entirely new weather patterns.
• They tend to forecast events similar in strength to those already observed.

"AI models depend strongly on the training data and are relatively constrained to the range of this dataset. In other words, AI models struggle to simulate brand new weather patterns, instead tending to forecast events of a similar strength to those seen before."

Can AI Models Predict Unprecedented Extreme Events?

Record-breaking extreme weather events are becoming more frequent and intense due to climate change. These events, by definition, have never occurred before, making them particularly challenging for AI models to predict accurately.

The study highlights a critical gap: while AI models excel in many areas, their reliance on historical data may hinder their ability to forecast truly unprecedented weather phenomena. This limitation raises concerns about fully transitioning to AI-based forecasting systems without further research and validation.