Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Explore how artificial intelligence can bridge weather prediction and climate modeling in this conference talk from IPAM's Mathematics and Machine Learning for Earth System Simulation Workshop. Discover how deep learning successfully emulates atmospheric reanalyses with high fidelity, enabling well-calibrated ensemble weather forecasts at progressively longer lead times. Learn about a proposed minimal framework for extending AI prediction systems to climate-relevant horizons, which requires explicit representation of external forcings like greenhouse gases and land-use change, while restricting them to physically appropriate state tendencies. Examine the importance of stress-testing AI model robustness in out-of-distribution regimes, including extreme weather events and counterfactual climate trajectories. Analyze comparative studies using leading climate emulators and hybrid physics-AI models that identify coupling and development challenges while comparing computational scaling with resolution and effective complexity. Understand why AI models may not be intrinsically more efficient than GPU-ported dynamical models when complexity is properly accounted for, yet discover their unique advantage in directly predicting target variables at desired grid resolutions without integrating the full high-frequency, multivariate state. Investigate diverse machine learning downscaling strategies that can partially substitute for explicit fine-scale resolution when observational data is available, potentially enabling inexpensive, local risk assessment across multiple prediction horizons from weather to climate timescales.
