top of page

Clouds - cloud modeling

  Cloud modeling is very challenging. In AR5, two classes of cloud modeling are assessed - cloud process modeling and parameterization of clouds in climate models. The major distinction between these two types of modeling strategies lies in their representative spatial and temporal scales. The former simulates with explicit clouds in high-resolution small domains; the latter involves parameterization of explicit clouds in General Circulation Models (GCMs) over global scale. Figure 7.8 (below) illustrates these differences, followed by a brief introduction on each modeling strategy.

Figure 7.8 | Model and simulation strategy for representing the climate system and climate processes at different space and time scales. Computational power prevents one model from covering all time and space scales. Since the AR4, the development of Global Cloud Resolving Models (GCRMs), and hybrid approaches such as General Circulation Models (GCMs) using the ‘super-parameterization’ approach (sometimes called the Multiscale Modelling Framework (MMF)), have helped fill the gap between climate system and cloud process models. 

1. Various Cloud Modeling Strategies

 

1.1 Cloud Process Modeling

 

    Major challenges of cloud process modeling arise from the large range of scales associated with various cloud formation processes, ranging from sub-micrometer (e.g. formation of cloud condensation nuclei (CCN)) to thousands of kilometers (e.g. cloud system). In AR5, it is pointed out that this large range of scale "is impossible to resolve with numerical simulations on computers, and this is not expected to change in the foreseeable future."

 

    Currently, two popular schools of modeling exercise fall on this category - one is the cloud-resolving models (CRMs), the other the large-eddy simulation (LES). CRMs can resolve the deep cumulus motions. They have been used for characterizing individual storms (horizontal resolution hundreds of meters), for estimating boundary layer cloud fractions, and moisture/energy fluxes (horizontal res. 1 km or larger), and for simulating deep convective clouds (horizontal res. 2 km or finer). In contrast, LES models are good at resolving boundary layer eddies. With even finer resolutions (horizontal res. ~100 m, vertical res. ~40 m), LES models have been used to study the vertical thermodynamical (e.g. temperature, moisture, cloud fraction, vertical velocity) structures and turbulent fluxes for the shallow trade wind cumulus convection system and the stratocumulus-topped boundary layers. Due to their fine grid resolutions, CRMs and LES models cannot be used for long-term projection, but rather for case studies.

 

    Regardless of their relatively fine grid resolution, the greatest limitation of these cloud process models is their capability of simulating cloud microphysics, precipitation and aerosol interactions. This is mainly due to our inadequate knowledge on basic cloud microphysics, especially on formation of ice clouds and aerosol-cloud interactions. Reflected by model performance, researchers have found that in the current cloud process models:

 

  • Rain starts too early in the day for warm clouds (CRMs);

  • Simulated precipitation efficiency shows some sensitivity to microphysical parameterization (LES);

  • Modeled cloud ensemble properties (e.g. vertical distribution and optical depth of ice clouds) are sensitive to CRM microphysical parameterization assumptions;

  • Simulated entrainment rate and cloud liquid water path are sensitive to the underlying numerical algorithms.

 

Bearing these limitations in mind, the modern cloud process models still represent great advance in simulating cloud and precipitation characteristics, and in understanding how turbulent circulations within clouds transport and process aerosols and chemical constituents. These models are especially useful for:

  • interpretation of in situ remote sensing observations;

  • understanding the influences of small-scale interactions, turbulence, entrainment and precipitation on cloud dynamics;

  • prediction on changes of cloud system properties (e.g. cloud cover, depth, radiative effect) in response to the changing climate;

  • testing and improving GCM’s parameterizations of cloud feedbacks and adjustment mechanisms, for instance, cloud-controlling processes (e.g. cumulus convection, turbulent mixing, small-scale horizontal cloud variability, aerosol-cloud interactions) and interplay between convection and large-scale circulations.

 

1.2 Parameterization of Clouds in Climate Models

 

    There are also two schools of modeling strategies fall on this category – one being those GCMs with grid resolution fine enough (as small as 3.5 km) to resolve large individual cumulus clouds over the entire globe, the other called GCM using ‘super-parameterization’ or Multiscale Modeling Framework (MMF). Currently, GCMs with explicit clouds cannot be used for climate projection, also due to being computationally demanding.

 

    Results from the Coupled Model Intercomparison Project Phase 5 (CMIP5) show that GCMs with explicit cloud parameterization better simulated the interplay between convective circulation and large-scale dynamics than the conventional GCMs. The former produced better simulations of the cloud cover, the diurnal cycle of precipitation, the Asian monsoon, the Madden-Julian Oscillation, and the El Nino-Southern Oscillation. Since they begin to resolve cloud-scale circulations, GCMs with explicit clouds can be used for studying the aerosol-cloud interactions that are missing in conventional GCMs. Regardless of all these advances, no current GCMs with explicit clouds can fully resolve cloud processes, particularly for low cloud systems.

 

 

In summary, GCMs, even with explicit cloud parameterization, still have great limitations in representing small-scale cloud variability. In contrast, CRMs and LES models have much more realistic simulation of cloud-scale variability. However, limited by computational power, they cannot extend to spatial and temporal scales large enough for determination of interactions between different cloud regimes, as well as evaluating their planetary radiative effects. In the near future, the best approach might be using CRMs or LES models combined with observations to better understand the uncertain cloud processes. Then, use the improved knowledge to better parameterize cloud system, and their rapid adjustments and feedbacks to climate change in GCMs.

2. Assessment of Cloud Parameterization in GCMs

 

    Clouds are an important climate forcing. Compared with the incoming solar irradiance at the top of atmosphere – 340 W m^-2, the net cloud radiative effect is ~ -20 W m^-2. Clouds of different type, at different height, with different droplet size distribution or liquid water / ice content, will have different radiative effects. This makes parameterization of cloud formation and evolution very important.

 

    In the meantime, cloud also influence climate through cloud processes, such as cloud dynamics and aerosol-cloud interactions. For instance, cumulus convection affects not only the surface heat and moisture fluxes and consequently the precipitation patterns in tropical and subtropical regions, but also the transport of aerosols and other atmospheric constituents. It has been recognized that parameterization of cumulus convection causes the largest uncertainties in estimate of the climate sensitivity [Sherwood et al., 2014].

 

    As discussed above, representing clouds in climate models is a great challenge. One reason is due to so many unknowns in our knowledge on cloud processes; the other is because the large grid spacing of GCMs (100 – 200 km horizontally and 100 or 1000m vertically) makes them hard to discern sub-grid small-scale variability of cloud field.

 

    Since AR4, results from cloud process models have significantly advanced our knowledge on cloud processes. Using cumulus convection parameterization as an example, CRMs and LES have helped people better understand entrainment/detrainment of cloud liquid and ice, and better distinguish turbulence and convective motions in cumulus updrafts. Better knowledge further inspired better convection parameterization in GCMs [Donner, 2014; Holloway et al, 2014; Hourdin et al., 2013]. Also since AR4, more applications of super-parameterization have improved GCM’s capability of simulating small-scale cloud variability.

 

      Below is a brief summary of improvements and remaining issues of AR5 cloud parameterization, in terms of microphysics of warm and mixed-phase/ice clouds, convective parameterization, and parameterization of cloud radiative effect.

 

2.1 Liquid clouds

 

​   Progress:

  • predict both mass and number mixing ratio of cloud, rain, and snow on more solid physical basis;

  • more realistic treatment on cloud scale water variability and cloud droplet activation [Morrison & Gettelman, 2008; Saltzmann, et al., 2010]

 

   Remaining issue: still unable to fully represent turbulence

 

2.2 Mixed-phase & ice clouds

 

   Progress:

  • more realistic treatment of mixed-phase processes and ice super saturation [Liu et al., 2007; Tompkins et al., 2007; Gettelman et al., 2010; Saltzmann, et al., 2010]  

   Remaining issue: Some real world complexities still missed, e.g. influence of pre-existing ice crystals in depletion of surrounding water vapor, impact of mixed-phase processes on storm strength and electrification, etc.

 

2.3 Convection parameterization

 

  Progress:

  • diagnose the vertical velocity in cumulus updrafts, in order to more completely represent aerosol activation, cloud microphysical evolution and gravity wave generation by cumulus convection;

  • couple shallow cumulus convection more closely to moist boundary layer turbulence

  Remaining issues: 

  • more realistic climate variability simulation at the expense of degraded mean state simulation;

  • some simulate a premature deep convective initiation over land.

 

2.4 Cloud radiative effect

 

   Progress:

  • improved sub-grid cloud variability, by using advanced computing methods, e.g. use of probability density functions of thermodynamic variables [Sommeria & Deardorff, 1977; Watanabe et al, 2009], stochastic approaches for radiative transfer [Baker et al, 2008], new treatments of cloud overlap [Shonk et al., 2012]

   Remaining issue: ‘too few, too bright’ low clouds

bottom of page