Buyu Deng 邓卜瑜

CO₂ Storage Modeling and Rock Physics Agent Development

At Stanford SCERF, I compared five grid resolutions in ECLIPSE 300 CO₂STORE simulations using fixed-time plume geometry, fixed-location saturation histories, space-time plume metrics, and computational cost. I selected a 0.12 ft vertical grid for subsequent modeling.

I mapped geology and CO₂ saturation to elastic properties with a facies-based rock physics workflow, then generated baseline and time-lapse seismic responses for 20 years of injection and 80 years after shut-in.

I also designed an agent workflow using the Stanford Rock Physics and Borehole Geophysics (SRB) model library. Across 45 independent runs, I compared agents with no references, textbook guidance, and textbook guidance plus SRB tools. Without references, agents made unsupported model assumptions. Textbook guidance reproduced reference calculations. SRB MCP tools enabled direct model calls and matched the references when used correctly, but some runs misinterpreted tool outputs and produced incorrect predictions.

[Download final presentation (PDF)]

All-in-One 3D Seismic Denoising

Degradation-representation learning for robust suppression of random noise, acquisition footprints, and migration artifacts in 3D seismic data. [Code] [Download noise addition slides (PDF)]

PHY-to-BGC Ocean Biogeochemical Emulation

A 2.5D full-image U-Net emulator mapping global physical ocean fields to coupled biogeochemical variables, with evaluation under anomalous ocean-climate conditions. Manuscript in preparation.

GNSS–NISAR Ionospheric Reconstruction

Aligned GNSS TEC-derived phase fields with processed NISAR interferometric phase and trained a residual U-Net to test cross-modal reconstruction. The model followed broad GNSS trends but did not recover NISAR-scale structure from the available sparse observations.

[Download English project report (PDF)]