Improving The Modeling and Analysis of Tropical Convection and Precipitation through Machine Learning Methods - Dissertation

Aug 5, 2026

This dissertation improves modeling and analysis of tropical convection and precipitation using machine learning methods. It develops neural-network emulations for sub-grid parameterizations, evaluates spatial and temporal variability, and performs formal hyperparameter tuning with attention to physical constraints such as the diurnal cycle. It further investigates generative modeling of atmospheric convection via architectures like VAEs, then compares storm-resolving models and climates using clustering and latent-space analysis to yield physically interpretable convection structures.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages198

Reading Time8 h 19 min

AccessFree to download

Introduction

Background

Outline

Neural-Network Emulation of Sub-grid Parameterizations

Methods

Results

Conclusion

Generative Modeling of Atmospheric Convection

Methods

Results

Conclusion

Comparing Storm Resolving Models and Climates

Methods

Results

Discussion

Understanding Extreme Precipitation Changes

Methods

UC Irvine UC Irvine Electronic Theses and Dissertations Title Improving The Modeling and Analysis of Tropical Convection and Precipitation through Machine Learning Methods Permalink [https://escholarship.org/uc/item/3sz018pc](https://escholarship.org/uc/item/3sz018pc) Author Mooers, Griffin Stuart Publication Date 2023 Copyright Information This work is made available under the terms of a Creative Commons Attribution License, availalbe at [https://creativecommons.org/licenses/by/4.0/](https://creativecommons.org/licenses/by/4.0/) [Peer reviewed|Thesis/dissertation](Peer reviewed|Thesis/dissertation) [eScholarship.org](eScholarship.org) Powered by the California Digital Library University of California UNIVERSITY OF CALIFORNIA, IRVINE Improving The Modeling and Analysis of Tropical Convection and Precipitation through Machine Learning Methods DISSERTATION submitted in partial satisfaction of the requirements for the degree of DOCTOR OF PHILOSOPHY in Earth System Science by Griffin Mooers Dissertation Committee: Associate Professor Mike Pritchard, Chair Associate Professor Stephan Mandt Assistant Professor Tom Beucler Professor Jin-Yi Yu Professor Jim Randerson © 2023 Griffin Mooers TABLE OF CONTENTS Page LIST OF FIGURES v LIST OF TABLES xiv ACKNOWLEDGMENTS xvii VITA xix ABSTRACT OF THE DISSERTATION xxi 1 Introduction 1 1.1 Background .................................... 1 1.1.1 Our Cloud-Climate Deadlock ...................... 1 1.1.2 The Tropical Atmosphere ........................ 3 1.1.3 Machine Learning ............................. 5 1.1.4 Outline .................................. 8 2 Neural-Network Emulation of Sub-grid Parameterizations 12 2.1 Abstract ...................................... 12 2.2 Introduction .................................... 13 2.3 Methods ...................................... 16 2.3.1 Climate Simulation Data ......................... 16 2.3.2 Neural Network Design .......................... 19 2.3.3 Performance Analysis and Postprocessing ................ 20 2.3.4 Formal Hyperparameter Tuning ..................... 24 2.4 Results ....................................... 27 2.4.1 Spatial Structures ............................. 28 2.4.2 Temporal Variability ........................... 32 2.4.3 Hyperparameter Optimization vs. Physical Constraints for Emulating the Diurnal Cycle ............................. 35 2.4.4 Towards Interactive Land Coupling ................... 40 2.5 Conclusion .................................... 42 2.6 Appendix A: Performance Comparison with Existing Literature ....... 46 2.7 Appendix B: Supporting Tables, Figures, and Movies ............. 46 3 Generative Modeling of Atmospheric Convection 66 3.1 Abstract ...................................... 66 3.2 Introduction .................................... 67 3.3 Methods ...................................... 69 3.3.1 Architecture ................................ 70 3.3.2 VAE Loss Implementation ........................ 70 3.3.3 Data & Preprocessing .......................... 72 3.3.4 Quantifying Reconstruction Performance ................ 74 3.4 Results ....................................... 77 3.5 Conclusion ..................................... 79 3.6 Appendix A: Additional Figures ......................... 80 4 Comparing Storm Resolving Models and Climates 82 4.1 Abstract ...................................... 82 4.2 Introduction .................................... 83 4.3 Methods ...................................... 85 4.3.1 Data and Preprocessing ......................... 85 4.3.2 Variational Autoencoders ......................... 88 4.3.3 Understanding Convection via Vertical Structure ........... 90 4.3.4 The Horizontal Extent of Convection .................. 90 4.3.5 K-Means Clustering of Tropical Convection ............... 91 4.3.6 Vector Quantization ............

FAQs

What is the dissertation’s main goal regarding tropical convection and precipitation?

To improve how tropical convection and precipitation are modeled and analyzed by using machine learning methods, including neural-network emulation and generative modeling approaches.

How does the work evaluate neural-network emulations of sub-grid parameterizations?

Through performance analysis and postprocessing, including assessments of spatial structures, temporal variability, and hyperparameter optimization while considering physical constraints for the diurnal cycle.

What modeling techniques are used to study atmospheric convection and its patterns?

The dissertation uses generative modeling, including VAE-based approaches, and compares model outputs using methods such as clustering and analysis in latent space to reveal differences and interpretable convection structures.

Aug 5, 2026

Upload by Ava Thompson