This project addresses the challenge of explaining the added value of machine learning in weather and climate models, particularly for complex deep learning models. By constructing a hierarchy of Pareto-optimal models along an error-complexity plane, the researchers, including Sara Shamekh, provide insights into model development and performance. Through three applications—cloud cover parameterization, shortwave radiative transfer, and tropical precipitation modeling—it demonstrates how machine learning can uncover nonlinear relationships, improve parameterization, and capture key physical processes. This hierarchical approach aims to improve understanding and trust in machine learning models for atmospheric science.