Understanding the underlying causal structure of complex systems is essential for advancing scientific insight beyond mere statistical associations. In this talk, I introduce the foundations of causal models and the task of causal discovery—learning causal relationships directly from data. I then present DCDI, a differentiable causal discovery method capable of leveraging interventional data and integrating seamlessly with modern machine learning tools. Finally, I show how causal representation learning can be applied to real-world climate science data to uncover interpretable latent structures.