This lecture explores how hybrid and causal machine learning can combine physical knowledge with data-driven models to improve understanding and inference in Earth system science. It highlights recent advances toward more consistent, explainable, and physically grounded AI methods for geoscientific analysis.