Weather shapes everyone's decisions and safety. Knowing how far ahead it can be forecast would show how much warning is possible for extreme events, which can help officials save lives, avoid damage and save money, and help plan renewable energy.
What we know
AI models now match or beat the best traditional forecasts up to about 15 days ahead.
What blocks us
The chaos of the atmosphere seems to limit forecasts to around two weeks.
Current leads
AI ensemble forecasts, monthly and seasonal predictions.
Sub-questions
Can AI weather models ever work without the traditional physics-based models that supply their training data and starting conditions?
How can forecasts best describe the uncertainty that grows quickly beyond a few days ahead?