Return to Colloquia & Seminar listing
Machine learning for optimal tuning of the Simple Cloud-Resolving Earth Atmosphere Model (SCREAM)
Student-Run Research Seminar| Speaker: | Jennifer Paige, UC Davis |
| Location: | 2112 MSB |
| Start time: | Wed, May 7 2025, 12:10AM |
The climate group at Livermore National Lab are part of the team that contribute to DOE's climate model (the Energy Exascale Earth System Model, or E3SM), with their focus on advancing the atmospheric component. A high-resolution version of the Earth Atmosphere Model in C++ (called SCREAM) advanced the frontiers of climate computation by allowing global modeling at a scale fine enough to resolve important cloud dynamics. However, with the novelty of this implementation comes several challenges--one of these challenges is parameterizing the sub-grid micro-physical processes that the model relies on. I will present work on taking a dataset of perturbed-parameter run outputs to build a machine learning surrogate of the SCREAM model. Using the surrogate to inform our understanding of different possible parameterizations, we then use this to find an optimal parameterization and test with runs of the SCREAM model.
Free pizza as always. This is her practice qual, so questions are welcomed! :)