My research sits at the intersection of computational science and Earth observation, spanning two complementary directions. On one side is a 15-year program in numerical methods for subsurface fluid dynamics — work that began during my postdoc at RERI and has produced tools for CO₂ sequestration, natural hydrogen exploration, and fractured reservoir engineering. On the other is a newer direction, born during my 2020–2021 sabbatical, that applies deep learning to multi-modal satellite imagery at planetary scale. Both directions share a commitment to high-fidelity physics and a focus on problems with real environmental and energy implications.
This program started during my 2008–2013 postdoc at the Reservoir Engineering Research Institute (RERI) in Palo Alto, working with Abbas Firoozabadi. For over 15 years I have been developing and applying higher-order finite element methods for compositional multiphase flow in geologically complex, fractured porous media. The work spans fundamental numerical methods, CO₂ sequestration monitoring, shale gas, and — most recently — natural hydrogen systems, for which my group holds 4 patent families with 10 fully issued patents.
Mixed-hybrid and discontinuous Galerkin finite element methods (MH-FEM and DG-FEM) for multicomponent, multiphase, compressible flow on unstructured 3D grids (hexahedral, prismatic, tetrahedral). These higher-order methods achieve orders-of-magnitude lower numerical dispersion on coarse grids compared to standard finite-difference simulators, enabling accurate compositional modeling on realistic geological geometries.
Explicit discrete fracture network (DFN) modeling of multiphase flow incorporating capillarity, Fickian diffusion, and gravitational effects. Development of an alternative cross-flow equilibrium model for fractured media. Applications include CO₂ injection in fractured heavy oil reservoirs, water coning, and CO₂-enhanced recovery in naturally fractured systems validated against core-flood experiments.
Viscous and gravitational fingering in multiphase compressible compositional flow; dissolution-driven convection in geological CO₂ sequestration; scaling relations for onset and growth of instabilities. Fundamental studies of hydro-thermodynamic mixing across phases in porous media, with implications for CO₂ trapping efficiency and enhanced oil recovery.
Reactive transport modeling of CO₂ injection in brine formations; heterogeneity-assisted storage; simulation of the Cranfield (Mississippi) sequestration pilot using high-resolution static models and accurate CPA equation of state. Monitoring via stable isotopes and perfluorocarbon tracers — DOE-funded work (2013–2019) with field applications to Gulf Coast saline aquifers.
Experimental and theoretical characterization of gas adsorption/desorption in nanoporous organic shales under geological pressure and temperature conditions. Canister desorption testing; supercritical gas adsorption theories; pore structure characterization via NMR and nitrogen adsorption. Applications to unconventional resource assessment in the Ordos Basin and Appalachian plays.
Vapor-liquid equilibrium modeling of H₂–water mixtures using a Cubic-Plus-Association Equation-of-State (CPA-EOS); reservoir simulation for natural and stimulated hydrogen production; serpentinization and carbonization of mafic/ultramafic rocks. ARPA-E-funded work (2021–2025) in partnership with Koloma Inc. Four patent families with 10 issued patents in the US, Japan, Saudi Arabia, and South Africa; 22+ additional applications pending worldwide.
4 patent families · 10 issued patents · ARPA-E funded
This direction began during my 2020–2021 sabbatical, when I first trained fully convolutional neural networks on very-high-resolution commercial satellite imagery. It has since grown into my primary research focus, bringing together deep learning, physics-informed modeling, and multi-sensor satellite data to track Earth surface changes at global scale. In August 2025 this work was institutionalized through the BuckAI Observatory.
Fully convolutional neural networks (FCN) applied to 30 cm panchromatic WorldView/QuickBird commercial imagery for binary river classification. Models achieve >90% precision and recall. Enables tracking of Arctic river channel width changes and snowmelt-driven discharge variations from space at unprecedented resolution.
Remote Sensing of Environment, 2022
Deep neural networks for water classification at 2 m effective resolution from freely available Sentinel-2 multispectral imagery (native 10 m). Super-resolution approach recovers sub-pixel spatial detail for water body delineation — enabling global-scale hydrology without expensive commercial data.
Journal of Hydrology, 2023
Combining river channel geometry (hypsometry derived from ArcticDEM) with remote sensing observations to improve estimates of river discharge in data-sparse Arctic regions. Collaboration with Mike Durand (OSU), Ian Howat, and the NASA Terrestrial Hydrology program.
Remote Sensing of Environment, 2024
Semantic segmentation of "fairy circles" — semicircular surface depressions (SCDs) indicative of subsurface hydrogen degassing — using the Segment Anything Model (SAM) combined with Sentinel-2 multispectral imagery and Copernicus DEM. Global mapping of potential natural hydrogen reservoirs, covered by New Scientist and 10+ international outlets. Industry partner: Koloma Inc.
AGU Fall Meeting 2023 · industry-funded
Physics-informed neural networks (PINNs) combined with ICESat-2 satellite altimetry and Sentinel-2 multispectral imagery for shallow-water depth mapping in coastal and lacustrine environments. Peer-reviewed work on Sentinel-2 bathymetry decisions published at IEEE/CVF WACV 2025. NSF CAIG proposal ($1.5M, 2026–2029) pending.
IEEE/CVF WACV, 2025
AI-driven satellite remote sensing to quantify forest loss and gain, biomass change, and agroforestry dynamics in Uganda and the Sahel. Collaboration with economists (Leah Bevis, SPIA/CGIAR) to link satellite-derived land cover changes to food security outcomes. Funded by TDAI, Sustainability Institute, and CGIAR's Standing Panel on Impact Assessment ($1.25M, 2025–2028).
CGIAR/SPIA funded · in review, Comm. Earth & Environ.
Founded in August 2025 with a $1M OSU seed grant (2025–2030), the BuckAI Observatory is an interdisciplinary center within the College of Arts and Sciences dedicated to AI-driven Earth observation. The Observatory hosts weekly research seminars, a BuckAI Innovation and Data Visualization Space, and shared GPU infrastructure through Ohio Supercomputer Center's Unity HPC cluster and cloud resources. We maintain an active GitHub organization for open-source research code, and a certificate program in Geospatial AI is currently in development.
The Observatory's Scientific Advisory Board includes: