STRATOCUMULUS CLOUD EVOLUTION IN A LAGRANGIAN FRAMEWORK: LARGE-EDDY SIMULATIONS, GCM EVALUATIONS, AND MACHINE LEARNING

Abstract

Marine stratocumulus (Sc) clouds, low-lying and extensive over global oceans, play a crucial role in Earth’s climate by reflecting solar radiation and exerting significant radiative cooling. However, their response to global warming remains a major source of uncertainty in climate projections. To improve our understanding of Sc cloud evolution and its representation in climate models (also called General Circulation Models, Global Climate Models, or GCMs), this dissertation employs a Lagrangian framework, combined with Large-Eddy Simulations (LESs) and Machine Learning (ML) approaches. These complementary methodologies allow for an in-depth examination of environmental and microphysical controls on Sc clouds.The study first investigates the impact of horizontal temperature advection, one of the least understood cloud-controlling factors, on Sc cloud evolution using idealized Lagrangian LESs. Sc cloud decks persist longer under warm-air advection (WADV) than cold-air advection, which is supported by observations. This increased persistence is driven by reduced entrainment of dry air due to WADV-induced decoupling, which outweighs the reduction in cloud-base moisture transport. This mechanism becomes more pronounced when the free troposphere is more humid. The role of aerosols in the stratocumulus-to-cumulus transition (SCT) is also investigated via a set of Lagrangian LESs. It is explored by injecting aerosols from the surface into marine boundary layers. We find that the influence of aerosols on the transition is not subject to injection timing, although the boundary-layer stability changes over time. Moreover, in clean environments, surface-injected aerosols can significantly delay the SCT via suppressing precipitation, leading to an overall cooling effect. In polluted environments, however, the SCT delay is not notable due to saturated aerosol effects. Simulating Sc cloud evolution in GCMs remains a challenge. Low-cloud fraction (LCF) is evaluated in a GCM, namely the Community Earth System Model 2 (CESM2), against satellite observations in a Lagrangian framework. We identify a too-rapid decline in LCF during the SCT in CESM2, which is attributed to overestimated drying effects from sea surface temperature changes. To improve LCF predictions, an ML model (XGBoost) is developed, leveraging key large-scale meteorological factors while circumventing the representation of unresolved small-scale processes. The ML model significantly improves the long-standing issues of “too few” low clouds and “too rapid” SCT in GCMs, highlighting the unique potential of ML in improving LCF representations in GCMs.