How do aerosols affect marine clouds?
The research aims to better understand aerosol-cloud interactions (ACI) in marine low and warm clouds. These clouds are said to be marine boundary layer clouds since they form at the interface between the boundary layer and the free troposphere over the oceans. ACI remains the largest source of uncertainty in the assessment of anthropogenic climate change, and especially for ACI in marine boundary layer clouds because of their strong cooling effect on Earth's climate, and their large spatial and temporal coverage, notably over the midlatitude oceans.
The challenge is that the aerosol-cloud interactions are not explicitly resolved in global climate models (GCMs) because they occur at a microphysical scale. High-resolution cloud simulations do better resolve the relevant ACI and have always carried the promise to offer better ways to parameterize ACI in GCMs. Offering better parameterization means that we improve the description of the physical processes that govern aerosol-cloud interactions in the climate models. This goal has been elusive, in part because of the complexity of ACI but also because climate involves a wide range of conditions that make a generalization extremely challenging. Data-driven methods however offer the ability to capture the dynamics embedded in these datasets and point towards simplifications and parameterization. Here we look at different cloud fields that are the liquid water, the cloud droplet number concentration and the raindrop number concentration and we try to find a relationship with the occurrence of in-cloud updrafts and downdrafts. This would give us a better representation of cloud microphysics, essential to reduce the uncertainty in the future climate projections.
To do so, we use the δ-Maps method that is a machine learning-based network analysis. δ-Maps first detects domains and secondly builds a network that embodies the essence of the datasets that have been generated with Large Eddy Simulations (LES). A domain is a set of grid cells of the dataset that share a high temporal activity. The network is a graph with nodes and edges that represent the domains and their teleconnections. These teleconnections are weighted and directed so that we have the strength of the statistical relationship and the temporal ordering of the events. In other words, δ-Maps captures the dynamical structure of the cloud fields predicted by LES. This complexity reduction then points to distilling the essence of the simulations and lead to a true improvement of our knowledge on the effect aerosol particles have on marine boundary layer clouds.
