Zihan Ren (任子汉)

Thomas Edison’s lasting contribution was not the lightbulb — it was the electrical grid that made illumination scalable. Steve Jobs’s legacy is not the iPhone — it is the integrated ecosystem that unified an entire digital life in one device. I hold the same conviction for subsurface science: no isolated model, however sophisticated, solves the problem. What matters is the system — data, simulation, surrogate, and optimization, designed to function as one.

Subsurface systems are defined by what we cannot directly observe — sparse measurements, indirect signals, and empirical priors standing in for ground truth. Yet every decision built on this incomplete picture, from where to drill to whether carbon is truly sequestered, carries enormous stakes and irreversible consequences. The fundamental challenge is not prediction alone, but decision-making through uncertainty.

Nothing should work in isolation. I want to be part of the generation that builds integrated systems — where data acquisition, physical modeling, machine learning, and decision optimization are designed from the start to function as one coherent framework. If you find these ideas or my works interesting and would like to collaborate — or simply chat — please don’t hesitate to reach out via email or LinkedIn.

During my PhD, I built generative virtualization environments for pore-scale rock microstructures — sparse, expensive to scan, yet accurate source of single- and multi-phase transport properties. These environments synthesize controllable 3D structures conditioned on spatial rock properties, coupled with physical simulators via pore network modeling, and connected to field-scale heterogeneity. I also mentored graduate researchers on physics-informed neural networks (PINNs) — both soft-constraint and hard-constraint formulations — for predicting velocity and pressure fields in porous media.

Post-PhD, I began exploring a more systematic view on predictive modeling. The Bakken formation study is one step in that direction: data engineering, surrogate modeling, optimization, and decision support designed as one pipeline. Currently, I develop probabilistic inference methods for Enhanced Rock Weathering carbon removal — Bayesian unmixing of geochemical soil mixtures where each source endmember shifts under its own uncertainty, combined with optimized sampling to deliver defensible carbon credit auditing.