Upper-tropospheric cirrus influence the climate through local radiative heating. These cirrus extend throughout the tropics and can be advected up to 1,000 km during their long lifetimes (Luo & Rossow, 2004). Jensen, Toon, Pfister, and Selkirk (1996) found that very thin cirrus can warm the surrounding atmosphere by a few Kelvins per day. Due to the prevalence of the cirrus, this heating alters the top-of-atmosphere radiation balance (Haladay & Stephens, 2009;Lee et al., 2009).Tropical cirrus in the upper troposphere are strongly related to deep convection. Areas with a high occurrence of cirrus clouds are often collocated with frequent convection (e.g., Lee et al., 2009;Sassen et al., 2009;Schoeberl et al., 2018). Near the tropopause, cirrus can form through convective anvil detrainment as well as in situ ice nucleation . Some of the cirrus formed in situ may also be related to convection if the ice nucleation results from cooling caused by gravity wave perturbations (Chang & L'Ecuyer, 2020;Jensen et al., 2016;Krämer et al., 2016). Pervasive, mostly optically thin cirrus characterize the transition region between the upper troposphere and lower stratosphere, known as the tropical tropopause layer (TTL; see review article by Fueglistaler et al., 2009). Several definitions for the TTL boundaries have been proposed in

High clouds in the tropics have been difficult to reproduce in global climate models (GCMs) because of their complex microphysics and radiative properties (e.g., Del Genio, 2012;Stephens, 2005). Proper representation of the properties of tropical cirrus, especially cloud amount and hydrometeor distribution, is a key issue for improving GCMs (e.g., Inoue et al., 2010;Stephens, 2005;Zelinka et al., 2012). GCMs generally have a low spatial resolution, which is unable to explicitly represent convection and the subsequent tropical cloud life cycle. Diverse convective and ice microphysical parameterizations lead to large differences between GCMs in the ice cloud population (Del Genio, 2012). This is the second of two papers comparing the formation and properties of tropical cirrus in relation to deep convection in several high-resolution global storm-resolving models (GSRMs). Nugent et al. (2022) (hereafter Part I) focuses on deep convection and its role as a source of ice and vapor for cirrus formation. Here, we compare the simulated ice cloud populations with satellite observations, interpreting them in terms of an aggregate cirrus life cycle.As noted in Part I, GSRMs are attractive for modeling tropical cirrus. Unlike conventional climate models with horizontal grid spacings of 25-200 km, GSRMs have sub-5 km grid spacing that enables them to explicitly simulate deep convection and its detrainment of ice into the upper troposphere and better represent the mesoscale

We present a machine learning based emulator of a microphysics scheme for condensation and precipitation processes (Zhao-Carr) used operationally in a global atmospheric forecast model (FV3GFS). Our tailored emulator architecture achieves high skill (≥94%) in predicting condensate and precipitation amounts and maintains low global-average bias (≤4%) for 1 year of continuous simulation when replacing the Fortran scheme. The stability and success of this emulator stems from key design decisions. By separating the emulation of condensation and precipitation processes, we can better enforce physical priors such as mass conservation and locality of condensation, and the vertical dependence of precipitation falling downward, using specific network architectures. An activity classifier for condensation imitates the discrete-continuous nature of the Fortran microphysics outputs (i.e., tendencies are identically zero where the scheme is inactive, and condensate is zero where clouds are fully evaporated). A temperature-scaled conditional loss function ensures accurate condensate adjustments for a high dynamic range of cloud types (e.g., cold, low-condensate cirrus clouds or warm, condensate-rich clouds). Despite excellent overall performance, the emulator exhibits some deficiencies in the uppermost model levels, leading to biases in the stratosphere. The emulator also has short episodic skill dropouts in isolated grid columns and is computationally slower than the original Fortran scheme. Nonetheless, our challenges and strategies should be applicable to the emulation of other microphysical schemes. More broadly, our work demonstrates that with suitable physically motivated architectural choices, ML techniques can accurately emulate complex human-designed parameterizations of fast physical processes central to weather and climate models.

No abstract

No abstract