TaiJi: State-Conditional Adaptive Combination of Machine-Learning and Physics-Based Global Weather Forecasts

The paper observes that data-driven machine-learning weather models now rival and sometimes surpass operational numerical weather prediction, yet no single model dominates across variables, pressure levels, lead times, or regions. TaiJi is presented as a spatiotemporal adaptive ensemble framework that combines machine-learning and physics-based global weather forecasts in a state-conditional manner. It targets the atmospheric/geophysical weather-prediction domain and addresses the diminishing marginal returns of building ever-larger individual models.