Use of large-eddy simulations to design an adaptive sampling strategy to assess cumulus cloud heterogeneities by remotely piloted aircraft

Abstract

Trade wind cumulus clouds have a significant impact on the Earth's radiative balance due to their ubiquitous presence and significant coverage in subtropical regions. Many numerical studies and field campaigns have focused on better understanding the thermodynamic, microphysical, and macroscopic properties of cumulus clouds with ground-based and satellite remote sensing as well as in situ observations. Aircraft flights have provided a significant contribution, but their resolution remains limited by rectilinear transects and fragmented temporal data for individual clouds. To provide a higher spatial and temporal resolution, remotely piloted aircraft (RPA) can now be employed for direct observations using numerous technological advances to map the microphysical cloud structure and to study entrainment mixing. In fact, the numerical representation of mixing processes between a cloud and the surrounding air has been a key issue in model parameterizations for decades. To better study these mixing processes as well as their impacts on cloud microphysical properties, the paper aims to improve exploration strategies that can be implemented by a fleet of RPA. Here, we use a large-eddy simulation (LES) of shallow maritime cumulus clouds to design adaptive sampling strategies. An implementation of the RPA flight simulator within high-frequency LES outputs (every 5 s) allows tracking individual clouds. A rosette sampling strategy is used to explore clouds of different sizes that are static in time and space. The adaptive sampling carried out by these explorations is optimized using one or two RPA and with or without Gaussian process regression (GPR) mapping by comparing the results obtained with those of a reference simulation, in particular the total liquid water content (LWC) and the LWC distribution in a horizontal cross section. Also, a sensitivity test of length scale for GPR mapping is performed. The results of exploring a static cloud are then extended to a dynamic case of a cloud evolving with time to assess the application of this exploration strategy to study the evolution of cloud heterogeneities. While a single RPA coupled to GPR mapping remains insufficient to accurately reconstruct individual clouds, two RPA with GPR mapping adequately characterize cloud heterogeneities on scales small enough to quantify the variability of important parameters such as total LWC.