Abstract
Abstract. The climatic relevance of aerosol–cloud interactions depends on the sensitivity of the radiative effect of clouds to cloud droplet number N, and liquid water path LWP. We derive the dependence of cloud fraction CF, cloud albedo AC, and the relative cloud radiative effect rCRE=CF⋅AC on N and LWP from 159 large-eddy simulations of nocturnal stratocumulus. These simulations vary in their initial conditions for temperature, moisture, boundary-layer height, and aerosol concentration but share boundary conditions for surface fluxes and subsidence. Our approach is based on Gaussian-process emulation, a statistical technique related to machine learning. We succeed in building emulators that accurately predict simulated values of CF, AC, and rCRE for given values of N and LWP. Emulator-derived susceptibilities ∂lnrCRE/∂lnN and ∂lnrCRE/∂lnLWP cover the nondrizzling, fully overcast regime as well as the drizzling regime with broken cloud cover. Theoretical results, which are limited to the nondrizzling regime, are reproduced. The susceptibility ∂lnrCRE/∂lnN captures the strong sensitivity of the cloud radiative effect to cloud fraction, while the susceptibility ∂lnrCRE/∂lnLWP describes the influence of cloud amount on cloud albedo irrespective of cloud fraction. Our emulation-based approach provides a powerful tool for summarizing complex data in a simple framework that captures the sensitivities of cloud-field properties over a wide range of states.
Highlights
Aerosol perturbations can lead to changes in cloud brightness and amount via the influence of aerosol on cloud formation and various aerosol–cloud interaction (ACI) processes
While previous applications of emulators, e.g., Lee et al (2013) and Johnson et al (2015), have explored how the behavior of the system varies across the parameter space, we explore how it varies across the resulting state space
This transformation leads to a shift in the isolines: cloud fraction isolines become approximately vertical as τ > 1 is controlled by cloud water path (CWP) and hardly influenced by the additional contribution of rain water path (RWP) to LWP
Summary
Aerosol perturbations can lead to changes in cloud brightness and amount via the influence of aerosol on cloud formation and various aerosol–cloud interaction (ACI) processes. Our contribution addresses an increasing interest in machine learning approaches within the atmospheric sciences, especially in the context of parameterizing shallow clouds (Krasnopolsky et al, 2013; Schneider et al, 2017; Brenowitz and Bretherton, 2018; Gentine et al, 2018; O’Gorman and Dwyer, 2018) This interest in utilizing modern computational statistical methods illustrates a community need to address a certain mismatch between traditional process-based cloud research and synthesizing approaches, especially for representing clouds in climate models. Emulation has so far been used to investigate the relationship between model response and uncertain parameters associated with physical parameterizations and to a lesser extent boundary conditions, e.g., Lee et al (2011, 2013), Johnson et al (2015), and Posselt et al (2016) We adapt this method to quantitatively derive relationships between cloud-field properties that evolve over the course of numerical simulations, namely rCRE(LWP, N ), AC(LWP, N ), and CF(LWP, N ). We derive and discuss the partial susceptibilities in Eq (2) (Sect. 5) before we conclude (Sect. 6)
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
