My current research is supported by the U.S. Department of Energy, the NSF LEAP Science and Technology Center, the Zegar Family Foundation, and a Google Award for Machine Learning and Education with TPUs. The projects below are the ones I currently lead or co-lead; collaborative projects on which I am a co-investigator or named collaborator are listed separately.

Current projects

Cloud Microphysics Multi-Scale Modeling Moonshot (CM4)

Question: Can we develop consistent multi-scale representations for turbulence and cloud microphysical processes in Earth system models using AI and DOE ARM observations?

Methods: Agentic AI, differentiable single-column and Earth system modeling, automated parameterization development, benchmark design, online evaluation.

Collaborators: Columbia, PNNL, NCAR, UW–Milwaukee
Funding: U.S. Department of Energy, Genesis Mission

Connecting Laboratory Experiments and In Situ Observations of Depositional Ice Growth

Question: What controls the rate at which ice crystals grow by vapor deposition in cirrus clouds, and how should that process be represented in models?

Methods: Cloud chamber and levitation diffusion chamber experiments, aircraft cloud particle imagery, neural ordinary differential equations, symbolic regression, conditional diffusion models.

Collaborators: Pennsylvania State University, Columbia University
Funding: U.S. Department of Energy, Atmospheric System Research

Recent papers: ACP (2023), JGR: MLC (2025), ice growth discovery paper (in press).

Scientific Machine Learning and Ice Microphysics through LEAP

Question: How can scientific machine learning turn observations and high-resolution simulations into stable, interpretable cloud parameterizations in Earth system models?

Methods: Reduced-order modeling of particle-based microphysics simulation, self-supervised learning on cloud particle imagery, perturbed parameter ensembles.

Funding: NSF LEAP Center internal research awards

Recent papers: JAMES (2024), JAMES (2026).

Systematizing Earth System Model Development with Agentic AI

Question: Can we more systematically develop, integrate, and tune parameterizations in Earth system models?

Methods: Differentiable modeling, agentic AI, large-scale training on TPUs, benchmarking.

Funding: Google Awards for Machine Learning and Education with TPUs

Evaluating and Optimizing Cloud Seeding with AI

Question: Does operational cloud seeding measurably affect precipitation, under what conditions is it effective, and could optimized strategies alter conditions associated with drought or wildfire risk?

Methods: Vision language model information extraction, causal inference, large eddy simulations, reinforcement learning, denoising diffusion models.

Collaborators: Columbia University
Funding: Zegar Family Foundation

Recent papers: Scientific Data (2025) — a comprehensive structured dataset of reported U.S. cloud seeding activities, 2000–2025.

Collaborative projects

• An Automated, Multimodal-AI-Enabled Cloud Chamber for Constraining Cloud Microphysical Processes (DOE Genesis)
• NSF Science and Technology Center: Learning the Earth with Artificial Intelligence and Physics
• NASA Digital Twins for Climate Science (with NASA GISS)

---

Earth and Space Science Machine Learning research sprints

Frontier Development Laboratory Europe, 2022

I was the domain lead mentor for the Aerosols team. Over eight weeks our team developed a dataset of high-resolution geostationary satellite imagery and meteorological data to improve forecasting of pyrocumulonimbus events using machine learning and causal methods.

[1] Tazi et al., NeurIPS Tackling Climate Change with ML (2022)
[2] Diaz Salas‐Porras et al., NeurIPS Workshop on Causality for Real-world Impact (2022)

NASA Frontier Development Laboratory, 2019

I spent eight weeks in Mountain View, CA, as a domain science researcher on the Forecasting Geoeffectiveness team, developing a data-driven machine learning approach for predicting space weather phenomena.

[1] Lamb, Malhotra, Vlontzos, Wagstaff et al., NeurIPS ML4Physics Workshop (2019)
[2] Lamb, Malhotra, Vlontzos, Wagstaff et al., NeurIPS ML4Physics Workshop (2019)