AI systems have become competitive with traditional methods for weather forecasting, but they produce blurrier forecasts over time due to error penalties in training.
DeepMind's GenCast outperforms the European ensemble model on long-term forecasts by merging computational approaches with generative AI methods, maintaining high resolution.
Traditional weather models are advantageous as they rely on atmospheric physics and empirical data, running multiple instances to produce a measure of uncertainty in forecasts.
Efforts to combine traditional weather models with AI demonstrate the potential for improved forecasting accuracy, leveraging physical rules alongside machine learning innovations.
Collection
[
|
...
]