Research Projects

Warm, shallow convective clouds play a central role in shaping Earth’s climate through their impact on radiative balance and the hydrological cycle. Yet, predicting how these clouds evolve into rain bearing systems remains a long standing challenge in atmospheric science, largely due to what is known as the “cloud climate uncertainty” problem: our limited understanding of the microphysical processes that govern how cloud droplets grow, interact, and coalesce within a turbulent environment. Classical condensation theory predicts that droplets require more than two hours to grow beyond 40 microns under weak supersaturation, but field observations consistently show rain initiation in under 20 minutes. This discrepancy indicates that condensation alone is inadequate to explain the rapid broadening of droplet size distributions (DSDs) observed in nature. Instead, processes such as collision, coalescence, and droplet breakup—each strongly influenced by turbulence, gravity, and microphysical variability—are thought to drive the formation of precipitation. These processes are nonlinear, multiscale, and embedded in atmospheric motions that range from micron scale droplets to kilometre scale eddies and planetary circulations. This vast range of interacting scales poses a serious challenge for numerical models. Limited Area Models (LAMs), Numerical Weather Prediction (NWP) systems, and Global Circulation Models (GCMs) cannot resolve fine scale turbulent motions or the complex droplet scale interactions essential for accurate rainfall prediction. As a result, they rely on oversimplified parameterisations for key processes such as autoconversion, collision coalescence, breakup, and cloud albedo, which introduce large uncertainties into climate models and rainfall forecasts. In response to these limitations, superparameterisation approaches embed cloud resolving models (CRMs) within each grid cell of a GCM to explicitly simulate cloud scale processes, but even these remain computationally demanding and fall short of fully resolving droplet level physics. This proposal aims to address these limitations through a unified, multiscale modelling framework that blends high fidelity simulations with data driven approaches. Specifically, we propose a combination of Direct Numerical Simulations (DNS), super droplet cloud resolving models embedded within Large Eddy Simulations (SDM LES), and machine learning (ML) to build accurate, efficient representations of microphysical processes. ML algorithms will be trained on outputs from DNS and SDM LES to learn surrogate models for collision kernels, breakup rates, and other microphysical tendencies, which can then be integrated into larger scale models at reduced computational cost. By combining physics based simulations with ML based upscaling, this framework will bridge the gap between microscale interactions and mesoscale dynamics, ultimately leading to improved predictions of rainfall rates and DSD evolution in warm clouds. The long term goal is to reduce uncertainty in cloud microphysics across modelling platforms, improve our ability to forecast rainfall, and contribute to a better understanding of warm cloud feedbacks in the Earth’s climate system.