When was the last time you were outside watching the sky carefully? If you have done so it becomes immediately visible: Clouds appear in endless variations, reaching from low-level shallow cumulus, which provide spots of shade on a hot summer day, to towering cumulonimbus, which can cause severe damage in the course of a thunderstorm. In short, their natural variability is manifold and, consequently, the same is true for their interaction with the atmosphere around them.
In the last decades, many efforts have been made to understand these highly dynamical objects better. Some questions have been solved but a lot of open questions remain. Starting from the seemingly simple question of how to define the boundary of a cloud, to the more
complex questions, e.g., how does the solar radiation interact with clouds and how does a potential feedback change the radiative transport (RT), the cloud properties, and the cloud evolution?
My previous research tried to answer some of these open issues and focused on the development and application of airborne remote sensing methods to investigate cloud optical properties. In addition, observations of upward and downward irradiance, were used to evaluate the appropriate treatment of radiation in current operational numerical weather prediction (NWP) models.
Much too often, NWP and cloud retrieval methods suffer from one particular caveat - They treat clouds as homogeneous layers. As mentioned, one look at the sky and it becomes obvious that this is not a sufficient representation of clouds, especially in the case of heterogeneous small shallow-cumulus cloud fields.
This intrigued me to better understand the development of clouds and to advance their three-dimensional (3D) appearance and 3D RT effects in models.
Unfortunately, simulations of 3D RT is based on computationally expensive calculations, which limits their application to individual case studies and does not allow for use in operational NWP models. To circumvent the necessity of costly calculations, it is one of my research goals to develop parameterizations, which correct the simulated 1D RT of clouds to account for 3D effects. These parametrizations are based on machine learning techniques, i.e., pattern recognition of cloud clusters, which link them with cloud cover, cloud distribution, and cloud microphysical and macrophysical properties. Previous research, e.g., by Gristey et al.  and Schmidt et al , showed promising results, which provides encouragement to promote the development of models in this direction.
In contrast to the work done by Gristey et al. , who focused on shallow cumulus cloud fields over the Great Plains, it is intended to investigate trade-wind cumulus fields in the North Atlantic region, leeward to Barbados. In course of the Elucidating the Role of Clouds-Circulation Coupling in Climate (EUREC4A; Bony et al., 2017) campaign, measurements of downward spectral irradiances as well as radar and microwave observations onboard the research vessel Meteor, among many others, were performed. The radiation measurements will be used to calculated distribution functions of irradiance and compared with distributions obtained by 1D and 3D RT simulations. The analysis is supported by radar and microwave estimates of LWC from the ship. In conjunction with measurements from the research aircraft HALO the extensive dataset allows for a broad overview of the atmospheric state and to constrain the LE simulations better.
The development of machine learning-supported models fits well in the objectives of NOAA's Atmospheric Science for Renewable Energy (ASRE) project, which aims to better understand the evolution of shallow cumulus clouds, their representation in NWP, and to improve the forecast of solar energy from photovoltaic fields. Accurate predictions of the power output from such power plants is vital as the number of installed plants will increase in the coming years.
The development of the aforementioned 3D cloud radiation parameterization is a joint project of NOAA and CU’s Laboratory for Atmospheric and Space Physics (LASP).