Pedram Hassanzadeh, University of Chicago

Reducing the uncertainties associated with extreme weather forecasting and climate change projections is essential for formulating mitigation and adaptation strategies. This requires “accurate” weather and climate models that are “computationally cheap”, thus capable of producing large ensembles and probabilistic predictions. However, in traditional numerical weather and climate models, accurate models are computationally demanding. In recent years, advances in artificial intelligence (AI) have opened doors for the possibility of building AI-based weather and climate models that are 100,000 times faster than numerical models while providing comparable or even better accuracy. Weather forecasting at the time scales of 1 to 10 days has been already transformed by these AI-based models, and sub seasonal to seasonal forecasting is at the verge of a revolution too. I will briefly discuss some of these advances and suggest ideas for how AI-based models can even revolutionize climate change projections, at the decadal and longer time scales. Such AI-based climate models can potentially play a critical role in building the next generation of integrated assessment models.