STORM Research Group

Research

Our work is motivated by open questions in atmospheric and climate science: how to better observe and characterize atmospheric processes, how to reason about their dynamics and predictability, and how to model them more accurately across scales. To address these questions, we develop and apply a range of scientific tools, combining observations, physical modeling, and data-driven methods.

Three research pillars

Observe

We work with ground-based and satellite observations to produce reliable atmospheric datasets. This includes denoising ARM observatory retrievals, and developing high-resolution precipitation and temperature datasets over Africa for use in convection research and health applications.

ARM retrievals Downscaling Africa climate Remote sensing

Reason

We study atmospheric processes, particularly convection and the planetary boundary layer, using a combination of observations, large-eddy simulation, reanalysis, and data analysis. We are interested in what controls the transition from shallow to deep convection, and what sets the predictability of precipitation across time and space.

Shallow-to-deep convection PBL dynamics Predictability Process studies

Model

We develop data-driven tools for modeling atmospheric processes, including convective parameterizations, dynamical system emulators, and parameter estimation frameworks. A recurring theme is how to incorporate physical constraints into learned models, so that they remain consistent with known conservation laws.

Convective param. Dynamical emulation Physics-informed ML Parameter estimation

Methods

We draw on a range of tools, including numerical simulation, statistical inference, remote sensing, and machine learning, depending on what the problem calls for. Generative modeling approaches such as normalizing flows, flow matching, and diffusion models appear across several projects, particularly where uncertainty quantification or probabilistic output is important.