Dissertation Defense: Griffin Mooers

Title: Improving Modeling and Analysis of Tropical Convection and Precipitation through Machine Learning Methods

Abstract: Our knowledge of the atmosphere has increased immensely in the last few decades thanks to high-resolution "storm-resolving" climate models. With these models, we can simulate atmospheric processes including deep, moist convection with detail previously not possible. However, limits continue to constrain our understanding of the dynamics of the atmosphere. We presently lack the ability to run these new storm-resolving models (SRMs) for the durations we need to understand the cloud-climate feedback. Meanwhile, running these SRMs for any amount of time produces very large volumes of data which are difficult to objectively quantify. This work leverages disparate machine-learning approaches in an attempt to break through these deadlocks. First, we implement feed-forward neural networks to augment the "Super-parameterized Community Atmospheric Model" (SPCAM) allowing us to run the model at a fraction of the original computational cost but with the same accuracy even when realistic geographic boundary conditions are included. Second, we use deep generative models to analyze and organize SPCAM output. This allows us to identify unique types of convection as well as convective storm anomalies within the data. A third outcome involves expanding on this unsupervised learning work to compare different SRMs and quantify which have similar representations of the dynamics of the atmosphere. We find that even among high-resolution SRMs there are substantial differences in the type, proportion, and intensity of convection in representations of atmospheric dynamics. Fourth, we leverage these deep, generative modelsto measure climate change objectively and better understand the physical mechanisms driving changes in extreme precipitation. We capture anticipated signals of global warming with minimal human intervention while showing the importance of "Deep Convection" shifts in future changes in heavy rainfall.