Climate Physicists Face the Ghosts in Their Machines: Clouds
Quanta Magazine • • science
Key Points:
- Climate scientists are increasingly using artificial intelligence (AI) to improve climate models, particularly to better represent clouds, which are a major source of uncertainty in predictions of global warming.
- Traditional physics-based climate models rely on equations like Navier-Stokes but struggle to directly simulate clouds due to their small scale, leading researchers to use parameter estimates; AI is helping automate and refine this parameterization using extensive cloud simulation data.
- An alternative AI approach bypasses physics equations entirely, using neural networks trained on historical atmospheric data to forecast weather and climate, showing promising accuracy and much faster computation than traditional models.
- Despite advances, there is skepticism about AI models’ long-term reliability for climate forecasts because neural networks approximate rather than fully replicate physical laws, and they may
