Cloud-Radiation Parameterizations as a Scaling Problem
Richard C. J. Somerville
Scripps Institution of Oceanography
University of California, San Diego
Cloud-Radiation Parameterizations as a Scaling Problem
Richard C. J. Somerville
Scripps Institution of Oceanography
University of California, San Diego
In modeling the global climate, scaling is an especially critical issue, because small-scale processes, such as the formation of clouds and the role of clouds in the transfer of both solar and terrestrial radiation, have important effects on the climate system. Clouds are much too small to be modeled explicitly in a global simulation, and the physical processes involved in clouds are still imperfectly understood, but clouds and their radiative effects are too important to be ignored. The typical solution is to parameterize these effects. In this context, a parameterization is a rule or algorithm for representing the statistical effects of an ensemble of small-scale processes on the large-scale climate, with the understanding that these effects must be able to be prescribed explicitly as a function of conventional large-scale climate variables, such as wind, temperature and humidity.
Climate models are solved numerically on global grids with a typical resolution of a few hundred kilometers horizontally and about one kilometer or less vertically. When the model values of climate variables at this resolution are used to prescribe clouds, the resulting rules are often highly simplistic. These parameterizations may have some intuitive appeal, but they are rarely justifiable theoretically and have almost never been tested empirically in any thorough and satisfying way. For example, a typical parameterization might make the cloud amount in a model grid volume proportional to some simple function of the excess relative humidity above some prescribed threshold critical value. Both the functional form and the critical value can be "tuned" to reproduce crude observed climatic properties, such as the global planetary reflectivity, which is mainly due to clouds. However, parameterizations of this sort need not bear any resemblance to the way clouds actually vary in space and time in the present atmosphere, let alone as to how they will change and feed back on any future climate which differs from the present one. To put these highly idealized parameterizations on a firmer footing, a major direction in current research is to test them against observations.
Cloud-radiation interactions and feedbacks are at the top of every recent list of high-priority research topics in climate modeling because they dominate the response of climate models to imposed forcings.
It is important to realize that cloud-radiation interactions and feedbacks are not simply fascinating but ultimately unimportant playthings of climate modelers. On the contrary, they are at the top of every recent list of high-priority research topics in climate modeling, for the compelling reason that they dominate the response of climate models to imposed forcings such as changes in the atmospheric carbon dioxide concentration. Until a much better understanding of cloud-radiation processes is obtained and incorporated in models, it is a simple fact that large uncertainties will necessarily characterize all model predictions of climate change. Thus, the climate modeling community has been driven by the results of its own research to the realization that there is now a crucial need for careful comparisons between products of model cloud algorithms and observations of cloud-radiation processes in the actual atmosphere.
Somerville and colleagues have used an atmospheric general circulation model, or GCM, typical of those at the heart of contemporary climate modeling efforts, in inverse climate change simulations to study how climate sensitivity is affected by different cloud-radiation parameterizations. They have also used observations from several field programs to begin to test these same parameterizations for realism. In addition to a relative-humidity-based cloud scheme, of the type outlined above, they have tested several types of parameterizations incorporating prognostic cloud water, both with and without interactive cloud radiative properties. Their inverse climate change simulations involve forcing the model by prescribed global changes in sea surface temperature. Such simulations are far less ambitious than attempts to mimic the way the climate will be affected by gradual changes in greenhouse gas concentrations, but they are much easier to interpret and can thus be highly insightful.
In their GCM, the increase in cloud water content in a warmer climate leads to optically thicker middle and low clouds and in turn to negative shortwave feedbacks for the interactive radiative schemes, while the decrease in cloud amount produces a positive shortwave feedback for the schemes with specified cloud water path. Put in everyday terms, this means that in a warmer climate, clouds contain more water and are thus more reflective and tend to act like a thermostat to reduce the warming, according to the parameterizations tested that include varying water amounts and cloud radiative properties which depend on cloud water content. However, when the parameterizations were altered so that they did not include varying water amounts, the clouds simply decreased in amount in the warmer climate, thus decreasing their role in reflecting sunlight, and thereby amplifying the warming. In short, two plausible approaches to parameterization give feedbacks of opposite sign in terms of the cloud contributions to planetary reflectivity.
Two plausible approaches to parameterization give feedbacks of opposite sign in terms of the cloud contributions to planetary reflectivity.
A similar dilemma occurs in considering the cloud effects on the transfer through the atmosphere of infrared radiation emitted by the Earth. This is essentially the cloud contribution to the greenhouse effect. For these so-called longwave feedbacks, the decrease in high effective cloudiness for the schemes without interactive radiative properties leads to a negative feedback, while for the other parameterizations, the longwave feedback is positive. In other words, cirrus clouds, which are important contributors to the natural greenhouse effect, change with a climate warming so as to occupy a smaller fraction of the sky in the warmer climate than they do today, thus opposing the warming, according to the tested parameterizations of clouds without varying water. However, when the cloud water content is allowed to vary, the higher water content in the warmer climate adds to the greenhouse effect by making the clouds more opaque to infrared radiation, thereby amplifying the warming.
Thus, as in the case of cloud effects on solar radiation, two classes of parameterization give feedbacks of opposite sign in computing how clouds affect the transfer of terrestrial radiation through the atmosphere. Furthermore, in each class of parameterization, the sign of the feedback is different for solar radiation than for terrestrial radiation. It is also important to keep in mind that the quoted GCM results are all for global average circumstances. The global averaging obscures potentially serious local differences in sensitivity to clouds and cloud feedbacks. Whether a given region of the world experiences a positive or negative net cloud-climate feedback will depend on the types of clouds present there, which will certainly vary, not only from place to place, but with season, synoptic meteorological regime, and many other factors. Modeling of regional climate variability is still in its infancy. These sensitivity studies demonstrate how crucial it is that observational evidence be brought to bear to determine which, if any, of the tested parameterizations realistically represents how clouds behave in the actual atmosphere. Only recently have appropriate observations and theoretical tools begun to be available to tackle this task (Lee et al., 1997).
Somerville and colleagues' comparisons with observations are made using a theoretical diagnostic tool which they refer to as a single-column model, or SCM (see Figure 1.34). The model output includes temperature and moisture profiles, clouds and their radiative properties, diabatic heating terms, surface energy balance components, and hydrologic cycle elements. These comparisons of model versus measurement demonstrate clearly that it is inadequate to treat cloud amount as a simple function of relative humidity and to regard cloud optical properties as prescribed. Instead, the more realistic schemes are the more physically complete ones, i. e., those with explicit cloud water budgets, comprehensive treatments of cloud micro physics and interactive radiative properties which are based on the calculated cloud water amounts and detailed microphysics.
Sensitivity studies demonstrate how crucial it is that observational evidence be brought to bear to determine which, if any, of the tested parameterizations realistically represents how clouds behave in the actual atmosphere.
The fundamental concept of their SCM is to force and constrain an isolated time-dependent atmospheric GCM column with estimates of observed advective flux convergences, by which they mean the rates at which the wind advects heat, momentum and moisture into the column. These flux converges can be estimated accurately from modern measurements. The critical step is then to compare the model output with observations to judge the realism of the parameterizations. Because the SCM has only one space dimension (vertical), it is computationally very fast, and so it is practical to explore large ranges of all the relevant parameters by making hundreds or even thousands of numerical integrations, which is impossible with a full three-dimensional GCM.
The single column model is used to make the scaling link between observations and parameterizations.
Figure 1.34
Diagram of a Single Column Model (SCM)
Their SCM contains switch-selectable parameterizations based on current GCM practice. Their approach involves validating parameterizations directly against measurements from field programs, and then using this validation to tune existing parameterizations and to guide the development of new ones. They use the SCM to make the scaling link between observations and parameterizations. Surface and satellite measurements are used to provide an initial evaluation of the performance of the different parameterizations. The results of this evaluation are then used to develop improved cloud-radiation and precipitation schemes, and these schemes can then be tested in GCM experiments.
The SCM is diagnostic rather than prognostic. Its input is an initial state, plus the time-dependent advection terms in the conservation equations, provided from measurements at all model layers. Its output is a complete heat and water budget for the study site, specified as a function of altitude and time. The SCM thus may be thought of as a way of asking the parameterization in question how it would behave if it were forced and constrained by the fluxes that are actually observed at a given site in nature, rather than the fluxes that a GCM might compute under artificial circumstances.
In the hierarchy of climate models, this diagnostic model is intermediate between a physically comprehensive, fully three-dimensional model and an idealized treatment of an isolated physical process. Somerville and colleagues' recent research includes the following elements: incorporation of model improvements, particularly in the cloud-radiation formulations; testing and validation of the model through diagnostic analyses of observational data sets; and use of the diagnostic model to interpret the results of multi-dimensional models (i. e., GCMs), and to determine the sensitivity of model results to alternative parameterizations of physical processes.
The SCM is a way of asking the parameterization in question how it would behave if it were forced and constrained by the fluxes that are actually observed at a given site in nature, rather than the fluxes that a GCM might compute.
Figure 1.35
The SCM applied at three ARM sites: the Southern Great Plains, North Slope of Alaska, and Tropical Western Pacific sites.
Because the essence of the diagnostic use of single-column models involves comparing warming model output with intensive observations of an atmospheric volume representative of a single GCM grid cell, this type of research could not have been undertaken until modern observational technology was developed and deployed at appropriate sites. Suitable recent observations from field programs such as the Atmospheric Radiation Measurement program (ARM) of the U. S. Department of Energy are now available and give researchers the large data sets they require (see Figure 1.35). These observations provide invaluable information for the development of improved parameterizations. Such measurements permit the validation of GCM parameterizations using actual physical conditions rather than hypothetical ones. Independent measurements of quantities such as precipitation and net surface solar irradiance provide sensitive tests of the realism of parameterizations. In brief, this methodology allows direct observational validation of physical process parameterizations ( e. g., Randall et al., 1996).
Somerville and co-workers continue to develop and directly validate improved parameterizations of cloud-radiation and precipitation processes, using a diagnostic SCM together with observational data. It is now well-recognized, in both the GCM and numerical weather prediction (NWP) communities, that SCMs are useful tools for testing and improving parameterizations by validating them empirically against field observations. Their group has recently begun to emphasize precipitation as a crucial validation parameter. The extension to precipitation algorithms is a relatively new thrust, which should help accelerate progress with the cloud-radiation algorithms, because of the close connection via cloud microphysics. In one sense, precipitation is simply an additional validation route for the work on cloud microphysics. At the same time, improvements in the cloud water budget calculation can not only benefit the radiation schemes through improved specification of cloud radiative properties, but can also lead to improved precipitation simulations.
They are currently concentrating on improved physical process parameterizations for the treatment of summer convective precipitation and cloudiness. The proposed work combines analysis of observational data, theoretical process studies, diagnostic modeling, and GCM experimentation (Lee and Somerville, 1996). Their principal geographical regions of interest are mid-continent North America, where the main ARM sites in Oklahoma and Kansas are located, and the western tropical Pacific, which was the scene of the TOGA-COARE (Tropical Ocean Global Atmosphere Combined Ocean Atmosphere Response Experiment) field program and will also be an ARM site. They are pursuing a three-track strategy: (1) utilize stochastic radiative transfer theory (Malvagi et al., 1993; Byrne et al., 1996) to develop improved parametric representations of cloud-radiation interactions for atmospheric models; (2) validate and improve these parameterizations by using single-column models (Iacobellis and Somerville, 1991a, b; Randall et al., 1996) to make direct diagnostic comparisons with field observations; (3) test the parameterizations in their GCM to determine the sensitivity of model results to all aspects of the physical parameterizations (Lee et al., 1997).
This methodology allows direct observational validation of physical process parameterizations.
References
Byrne, R. N., R. C. J. Somerville and B. Subasilar, 1996. Broken-cloud enhancement of solar radiation absorption. J. Atmos. Sci., 53:878-886.
Iacobellis, S. F., and R. C. J. Somerville, 1991a. Diagnostic modeling of the Indian monsoon onset, I: Model description and validation. J. Atmos. Sci., 48:1948-1959.
Iacobellis, S. F., and R. C. J. Somerville, 1991b. Diagnostic modeling of the Indian monsoon onset, II: Budget and sensitivity studies. J. Atmos. Sci., 48:1960-1971.
Lee, W.-H., S. F. Iacobellis, and R. C. J. Somerville, 1997. Cloud-radiation forcings and feedbacks: General circulation model tests and observational validation. J. Climate, 10:2479-2496.
Lee, W.-H. and R. C. J. Somerville, 1996. Effects of alternative cloud-radiation parameterizations in a general circulation model. Annalae Geophysicae, 14:107-114.
Malvagi, F., R. N. Byrne, G. C. Pomraning, and R. C. J. Somerville, 1993. Stochastic radiative transfer in a partially cloudy atmosphere. J. Atmos. Sci., 50:2146-2158.
Randall, D. A., K.-M. Xu, R. C. J. Somerville, and S. Iacobellis, 1996. Single-column models and cloud ensemble models as links between observations and climate models. J. Climate, 9:1683-1697.
SCMs are useful tools for testing and improving parameterizations by validating them empirically against field observations.
Forward to Next Section//Back to Table of Contents//AGCI Homepage//Comments: agcimail@agci.org