Scientific AI for Cloud Processes and Earth System Models
I develop physics-informed machine learning and agentic AI methods to improve understanding of atmospheric processes and their influence on climate. I use observations (from laboratory, airborne, and remote sensing) and high resolution modeling, integrating data-driven discovery with process-based physical understanding. My current research focuses on cloud microphysics, differentiable modeling, and automated parameterization development, with applications to precipitation, aerosol–cloud interactions, and climate interventions.
- Associate Research Scientist, Dept. of Earth and Environmental Engineering, Columbia University
- Adjunct Assistant Professor, Climate School, Columbia University
- Scientific Machine Learning Research Thrust Lead, NSF LEAP Center
- AI for Science Director, Executive Committee, NSF LEAP Center
Current research
Agentic AI for Earth system model development
We develop multi-agent systems that help design, translate, test, and optimize Earth system model parameterizations. This work combines automated scientific reasoning, differentiable modeling, benchmark development, and rigorous online evaluation to make Earth system model development more systematic and reproducible.
Learning cloud microphysics with physics-informed machine learning
We combine laboratory and airborne field observations, high-resolution simulations, and scientific machine learning to discover interpretable cloud microphysical processes and develop improved model parameterizations. Current topics include using generative AI and physics-informed machine learning to improve ice microphysical process representation and using reduced order modeling to parameterize warm rain microphysical processes.
AI for precipitation and climate interventions
We use large-scale historical data, causal inference, and physically grounded machine learning to evaluate interventions affecting precipitation and climate risk. Current work examines the climatological impacts and optimization of cloud seeding, with connections to water resources, drought, and wildfire risk.
Current highlights
- I’m leading the DOE Genesis Mission project “Cloud Microphysics Multi-Scale Modeling Moonshot (CM4)” with collaborators at Columbia, PNNL, NCAR, and UW–Milwaukee.
- I’m collaborating on the Brookhaven-led DOE Genesis Mission project “An Automated, Multimodal-AI-Enabled Cloud Chamber for Constraining Cloud Microphysical Processes.”
- I recently received a Google Award for Machine Learning and Education with TPUs to explore how agentic AI and differentiable programming can accelerate Earth system model development.
- I’ll be an invited speaker for the session “Developments in Machine Learning Across Earth System Modeling” at AGU 2026 in San Francisco in December.
- I’m co-organizing the Representations for the Physical Sciences workshop at NeurIPS 2026 in Paris in December.
Selected publications
- Discovering How Ice Crystals Grow Using Neural Ordinary Differential
Equations and Symbolic Regression
[preprint]
Recovers interpretable governing equations for depositional ice growth directly from cloud chamber experiments. In press, 2026. - Perspectives on Systematic Cloud Microphysics Scheme Development with
Machine Learning
[link]
Lays out a framework for building microphysics schemes systematically rather than piecemeal. Journal of Advances in Modeling Earth Systems, 2026. - Structured Dataset of Reported Cloud Seeding Activities in the United States
(2000–2025) Using a Large Language Model
[link]
The first large-scale accessible data set of historic U.S. cloud seeding operations. Scientific Data, 2025. - Reduced Order Modeling for Linearized Representations of Microphysical
Process Rates
[link]
Shows that microphysical process rates admit compact, physically meaningful low-dimensional representations. Journal of Advances in Modeling Earth Systems, 2024. - Implicit Learning of Convective Organization Explains Precipitation
Stochasticity
[link]
Demonstrates that a learned measure of convective organization accounts for much of the apparent randomness in precipitation. PNAS, 2023.