Abstract: Systems and methods applicable, for instance, to AI-driven weather prediction and modification. Early detection and prediction functionality can perform actions including using deep learning models to identify emerging cyclonic systems, and using ensemble-based prediction frameworks to generate potential trajectory scenarios. Further, multi-scale modeling functionality can perform actions including combining models operating at different physical scales. Also, intervention optimization functionality can perform actions including applying reinforcement learning models to intervention strategy optimization, and employing differential evolution algorithms to identify atmospheric intervention points. Additionally, physical intervention functionality for atmospheric modification can involve actions including activating machine learning models when an atmospheric threat is detected, running counterfactual what-if ensembles for various candidate intervention classes, and evaluating rewards.