Oral Presentation

Machine learning (ML) is reshaping Earth science, offering new ways to extract information from data and simulate complex physical systems. In earthquake science, ML has significantly enhanced our ability to build high-resolution seismic catalogs and is emerging as a key tool for analyzing emerging data sources, such as DAS. However, its impact on forward modeling has been more limited. By contrast, climate science has seen rapid progress in developing machine learning-based emulators that approximate expensive simulations, enabling fast predictions and interactive exploration of climate scenarios. These emulators, when paired with uncertainty quantification and explainable AI, open new avenues for how both scientists engage with model-based predictions. In this talk, I will share insights from recent projects in my group, including machine learning methods for climate emulation and statistical downscaling, and intrinsically interpretable ML architectures for spatiotemporal data. These projects represent steps toward a future of AI for Earth Science in which scientific foundation models and human-AI interfaces expand the reach, trust, and utility of AI in both climate and earthquake science.