Abstract

The scientific evidence is now clear that human activities are influencing climate change. These activities include adding carbon dioxide and othergases to the atmosphere that increase the natural greenhouse effect andlead to global warming. They also include adding small particles and otherpollutants to the atmosphere. Climate models now tell us with great confidencethat, because of these human activities, the 21st century will be muchwarmer than the 20th century, with significant increases in sea level andchanges in weather patterns, among many other serious consequences forpeople and ecosystems.
 

Climate models are limited by incomplete understanding of key physicalprocesses, to a far greater extent than by computer limitations or othertechnical barriers.  The improvement of climate models depends ondeveloping new physical understanding and then finding physically realisticalgorithms incorporating it.  Today, clouds are believed to be thesingle most uncertain source of uncertainty in climate model forecastsof future climate change, because their interaction with solar and terrestrialradiation is such an important aspect of the climate system.  A currentdilemma of climate modeling is that model results are strongly sensitiveto the treatment of several poorly understood physical processes, especiallycloud radiation interactions. Thus, different models with alternative plausibleparameterizations often give widely varying results. Yet, we typicallyhave had little basis for estimating which parameterization is more realistic,although most of the global differences in results between leading climatemodels, as measured by their sensitivity to greenhouse gases, can be tracedto different model treatments of cloud radiation interactions.

Cloud radiation interactions are regarded today as one of the most criticalareas in global change research. In particular, when climate models areintercompared, cloud radiation parameterizations are responsible for mostof the global mean differences in sensitivity to greenhouse gas increases.This fact has become well established through parameterization transplantexperiments.  In such computational experiments, transplanting thecloud radiation algorithm from one model to another typically causes therecipient model to closely replicate the climate sensitivity of the donormodel.  The uncertainty in model responses is directly due to a lackof fundamental understanding of the physical processes involved. A majorresearch effort is underway worldwide in response to this challenge. Furthermore,closely related research areas, such as the role of atmospheric aerosolsin climate, are also beginning to receive the attention they deserve.

On even the simplest theoretical grounds, it is not surprising thatclimate is extremely sensitive to cloud amount. Similar arguments can bemade to show that climate also ought to depend strongly on other cloudproperties, such as cloud height and cloud liquid water or ice content.In fact, an easily stated but still unsolved major problem is to understandwhy the global cloud cover is now about 60% and why the planetary albedois now about 30%. Were these quantities the same during the ice ages? Whatmechanisms maintain the system at the present values of these key parameters,and how stable are these mechanisms to perturbations, such as those dueto changing greenhouse gas concentrations? We simply do not know. Untilwe find out, the "confidence limits" or "error bars" on the results ofclimate model greenhouse simulations will be much too large.

For many years, virtually all global climate model or general circulationmodel (GCM) treatments of clouds were based on simple algorithms relatingcloud amount to relative humidity. Such parameterizations usually producedpositive global average cloud radiation feedbacks in numerical experimentssimulating greenhouse-induced climate change. For example, in a typicalintegration performed with a GCM developed in the 1970s, a climate warmingdue to increased atmospheric carbon dioxide concentrations would lead toincreased average cloud heights and/or decreased average cloud amounts.It is easy to understand qualitatively why such feedbacks were positive.First, higher clouds are colder and so less effective infrared emitters,and they generally have lower albedos than lower clouds, so the cloud heightfeedback was positive (i.e., the change in clouds produced by the warmingtended to amplify the warming). Second, average model clouds, like averagereal clouds, contribute more strongly to the planetary albedo than to theplanetary greenhouse effect (in technical terms, the short-wave cloud forcingis larger than the long wave cloud forcing by about 20 W m-2). Hence, areduction in cloud amount reduces the short wave effect more than the longwave effect of clouds. Thus, the cloud amount feedback is also positive.

Climate models are now more numerous and more complicated, however,and model responses to increased greenhouse gas concentrations are morevaried. GCMs today attempt to take into account a broader range of physicalprocesses involved in cloud radiation feedbacks. The climate modeling communitynow realizes clearly that cloud feedback processes are not limited to macrophysicalcloud properties, such as cloud amount and cloud altitude. In recent years,many GCMs have begun to include cloud parameterizations which include explicittreatments of cloud physics.

Clouds have a powerful effect on the radiation budget of the earth.The reasons are obvious: clouds contribute to both the greenhouse effect,which warms the planet, and to the earth's reflectivity, or albedo, a competingcooling effect. Studies in which alternative cloud parameterizations canbe tested against observations show great promise for improving our understandingof clouds and their influence on climate. Ultimately, this improved understandingwill find its way into the representations of cloud used in general circulationmodels.

Improving the realism of this aspect of models is a key to improvingthe model simulations of climate change. We know that the leading modelsof today differ by a factor of three among themselves when they are comparedin terms of their simulation of the global average surface temperatureincrease due to a prescribed climate forcing, such as doubling the atmosphericconcentration of carbon dioxide. Most of this factor of three is due tothe different ways in which the models represent clouds and cloud radiationinteractions. We know, for example, that in many climate models, cloudamounts in a warmer climate tend to be somewhat less than in the present-dayclimate, and cloud altitudes tend to be somewhat greater. These changeslead to positive feedbacks, increasing the apparent sensitivity of themodel climate to the imposed forcing which led to the original warming,such as an increase in carbon dioxide.  However, the nature and strengthand especially the regional aspects of these feedbacks differ greatly frommodel to model. For this reason, many scientists regard cloud radiationprocesses as the most critical area of climate modeling research, deservinghighest priority for climate research resources.

Simple black-body radiative equilibrium calculations suggest that changesin cloud amount by only one or two percent might double or halve the modelsensitivity to carbon dioxide. This feedback occurs because clouds, whichcover about half the earth's surface, are responsible for about two-thirdsof the planetary albedo. At present, the albedo is about 30%. An albedochange of only 1% would cause a change in the blackbody radiative equilibriumtemperature of about 1°C. This is about the same black-body temperatureresponse as would occur in response to adding 4 W m-2 to the earth's surfaceradiation budget, which is approximately the direct radiative forcing equivalentto doubling the atmospheric carbon dioxide concentration.
 
Much research is now underway exploring the role that microphysicalproperties of clouds might play in affecting climate change. These processesmight well lead to strong feedbacks. For example, as the climate warmsbecause of an increase in the greenhouse effect, the entire hydrologicalcycle may accelerate. More water may evaporate from the oceans, and theatmospheric concentration of water vapor may increase. Because of the greateravailability of water vapor, some clouds in the warmer climate may havemore liquid water or ice than their counterparts in today's climate. Ingeneral, a higher liquid water or ice content is thought to lead to a higheralbedo, hence a negative feedback. However, for thin clouds, particularlycirrus, the cloud greenhouse effect may also increase. In addition, itis not at all clear that cloud water content will change in any systematicway as climate alters. It may also be too simplistic to look on temperatureas a dominant controlling factor for cloud microphysics.

Furthermore, the way in which cloud water or ice content depends ontemperature, even in the present climate, is not well understood. Simpletheory and aircraft data and some modeling studies support the idea ofhigher cloud water contents in warmer clouds. Some recent interpretationsof satellite data, however, suggest that even the sign of the temperaturedependence may be in doubt. It seems unlikely that any simple universalrelationship is valid. Additionally, the radiative properties of cloudsalso depend on factors such as the size distribution of cloud droplets,the shape of ice particles, and other factors. Despite much observationaland theoretical work in recent years to explore these issues, we are stillfar from a comprehensive physical understanding of them.

Nevertheless, several leading atmospheric general circulation modelinggroups in different countries have now incorporated this class of cloudfeedback mechanisms in one way or another. In a typical approach, cloudliquid water or ice content is included in the model as an additional prognosticvariable, just like temperature, wind velocity and water vapor. The physicalprocesses which act as sources and sinks for cloud water or ice, such asevaporation, condensation and precipitation, are simulated parametrically.In other words, the effects of these processes on the cloud water and icebudget are represented by simple formulas relating these processes to thelarge-scale variables which the model predicts explicitly.

The results of climate change simulations with these models confirmthe strong sensitivity of climate to cloud microphysics. In one strikingset of numerical experiments, a British general circulation modeling group(Senior and Mitchell, 1993) produced global average surface temperaturechanges (due to doubled carbon dioxide) ranging from 1.9 to 5.4°C,simply by altering the way in which these cloud-climate feedback mechanismswere treated in the model. They tested four different parameterizations,successively incorporating relative humidity cloud, prognostic cloud water,phase changes from water to ice, and interactive radiation dependent oncloud microphysics. Their cloud water algorithm was that of Smith (1990). It is somewhat unsettling that the results of a complex model can be sodrastically altered by what amounts to changing a few lines of code, essentiallyreplicating the factor-of-three difference in global sensitivity betweenGCMs that has been revealed by extensive model intercomparisons (Cess etal., 1989).  Clearly, further research is urgently required to understandthis class of physical processes better and to incorporate this understandingin models.

One particularly promising avenue of research is to combine processmodeling with intensive field observations and with research using generalcirculation models. For too long, research in this field has been characterizedby too many plausible cloud radiation parameterizations and too littleeffort to test them empirically. Now that appropriate observations andnovel modeling tools are at last becoming available, we may anticipaterapid progress in this critical area of climate research.

Recent research has led to a greatly increased understanding of theuncertainties in today's climate models.  In attempting to predictthe climate of the 21st century, we must confront not only computer limitationson the affordable resolution of global models, but also a lack of physicalrealism in attempting to model key processes.  Until we are able toincorporate adequate treatments of critical elements of the entire biogeophysicalclimate system, our models will remain subject to these uncertainties,and our scenarios of future climate change, both anthropogenic and natural,will not fully meet the requirements of either policymakers or the public. The areas of most-needed model improvements are thought to include air-seaexchanges, land surface processes, ice and snow physics, hydrologic cycleelements, and especially the role of aerosols and cloud radiation interactions.

Only about 70% of the sunlight intercepted by the Earth is availableto drive the climate system. The other 30% (the planetary albedo) is simplyreflected to space, mainly by clouds, which cover some 60% of the surfaceof the planet. On average, clouds reduce the global average absorbed solarradiation by about 50 Wm-2. Clouds also help to trap terrestrial radiation,contributing about 30 Wm-2 to the greenhouse effect. The net cloud radiativeforcing is the difference, approximately 20 Wm-2. Thus, the albedo effectdominates, and the net effect of clouds at present is to cool the Earth.However, the plain fact is that we lack a basic understanding as to whythe global cloud amount is about 60%, why the planetary albedo is about30%, how these and other fundamental quantities may have changed as climatechanged over geological time, and how they may change in the future.

One simple way to appreciate the climatic significance of clouds isto compare the cloud radiative forcing magnitudes given above with thedirect radiative effect of doubling the concentration of atmospheric carbondioxide, which is only about 4 Wm-2. Thus, if a climate change caused byincreased CO2 were to result in even a small change in the cloud amount,the cloud feedback effect might well be important. Furthermore, even ifglobal average cloud amount did not change appreciably in response to achanged climate, the spatial and temporal distribution of clouds mightwell be altered, as might other critical quantities, such as cloud altitudeand cloud radiative properties. All of these changes in clouds could leadto significant feedback effects. Until cloud processes are much betterunderstood, and until this understanding is incorporated in our models,the model results will always be subject to major uncertainties.

A serious dilemma of climate modeling today is that model results areextremely sensitive to parameterizations of several poorly understood physicalprocesses. As a result, models with different plausible parameterizationsgive very different results. Unfortunately, we have no firm basis for knowingwhich parameterization is more nearly "correct." Perhaps the most dramaticexample of this dilemma is the mystery of cloud radiation interactions.

Unfortunately, it is not known which of these parameterizations is themost realistic, or even if any of them captures the essential feedbackprocesses of actual clouds. Furthermore, other GCM groups have obtaineddifferent results by trying other ways of incorporating cloud microphysicalprocesses and their radiative interactions (e.g., Le Treut and Li, 1988,1991; Roeckner et al., 1987), in contrast to the approach which Seniorand Mitchell (1993) followed. There is thus a clear need to intercomparethese approaches with one another, and, even more importantly, to validatethem against observations, so as to evaluate the strengths and weaknessesof each.

It is noteworthy that the sensitivity of model-simulated climates tochanges in atmospheric carbon dioxide concentration has undergone majorfluctuations in recent years. The equilibrium global average surface temperaturechange in response to a carbon dioxide doubling, based on GCM results frommodels developed in the mid-1970s, was typically between 2 and 3 deg C.By the middle to late 1980s, the range of typical GCM sensitivities wasbetween 4 and 5 deg C. Nearly all of the increase in sensitivity couldbe traced to cloud radiation interactions. More recently, several GCMsincorporating more complex cloud algorithms, including some feedbacks arisingfrom cloud microphysical processes, have shown reduced sensitivity to changinggreenhouse gas concentrations (Senior and Mitchell, 1993).

In the earlier models, clouds were treated in a very simplistic way,and their ability to undergo changes, and thus to influence climate variability,was limited. In some GCMs, in fact, clouds and their radiative propertieswere prescribed once and for all and then held constant, so that no feedbackswere possible. Later models, by contrast, featured clouds which could anddid change their absolute amount and their height distribution in responseto changes in atmospheric water vapor content (e.g., Slingo, 1987). Asthe simulated clouds changed, so did their ability to contribute to bothplanetary reflectivity, or albedo, and to the greenhouse effect.

One type of problem is to characterize clouds, once they are formedin GCMs, i.e., to determine their radiative properties. Many unansweredquestions are tied to this type of problem. For example, does carryingcloud liquid water as a prognostic variable offer real advantages in termsof being able to specify cloud radiative properties realistically? Or isit feasible to specify these properties directly from the other large-scaleGCM fields? Some recent work (e.g., Tselioudis et al., 1993) suggests thatit may be difficult or impossible to infer cloud radiative properties assimple functions of temperature and other variables carried explicitlyby GCMs, but much research remains to be done on cloud characterization.

Another class of problems involves cloud formation. Here the goal isto  develop parametric treatments which enable GCMs to simulate whenand where clouds occur. In current GCMs, typical algorithms relate cloudamount to GCM variables such as relative humidity. Then partial cloud coveris handled by weighting clear sky and overcast radiative calculations bythe predicted cloud fraction. One promising alternative approach is tospecify clouds stochastically, i. e., to develop parameterizations whichyield probability distributions for variables such as the size and spacingof clouds (e. g., Malvagi et al., 1993).

New theoretical tools have been developed to aid in validating parameterizationsagainst observational data. One such tool is the single-column model orSCM (Somerville, 2000). An SCM is a computationally efficient and economicalone-dimensional (vertical) model, resembling a single column from a GCMgrid (e. g., Iacobellis and Somerville, 1991a, b). The model contains afull set of modern GCM parameterizations of subgrid physical processes.To force and constrain the model, the advective terms in the budget equationsare specified observationally (Randall et al., 1996).

The trend of increased reliance on observational field programs whichcan provide both satellite and in-situ data, together with the developmentof SCMs and other means of using these data, is a powerful combination(e. g., Iacobellis and Somerville, 2000).   This trend holdsgreat promise for improving our understanding of cloud radiation processesand for the future improvement of the treatment of these processes in climatemodels (e.g., Lee et al., 1997).
 
 

REFERENCES

Cess, R. D., and 19 others, 1989: Interpretation of cloud climate feedbackas produced by 14 atmospheric general circulation models. Science, 245,513-516.

Iacobellis, S. F., and R. C. J. Somerville, 1991a: Diagnostic modelingof the Indian monsoon onset, I: Model description and validation. J. Atm.Sci., 48, 1948-1959.

Iacobellis, S. F., and R. C. J. Somerville, 1991b: Diagnostic modelingof the Indian monsoon onset, II: Budget and sensitivity studies. J. Atm.Sci., 48, 1960-1971.

Iacobellis, S. F., and R. C. J. Somerville, 2000:  Implicationsof microphysics for cloud-radiation parameterizations:  Lessons fromTOGA-COARE. J. Atm. Sci., 57, 161-183.<

Lee, W.-H., S. F. Iacobellis, and R. C. J. Somerville, 1997:  Cloud-radiationforcings and feedbacks:  General circulation model tests and observationalvalidation. J. Climate, 10, 2479-2496.

Le Treut, H., and Z.-X. Li, 1988:  Using meteosat data to validatea prognostic cloud generation scheme.  Atmos. Res., 21, 273-292.

Le Treut, H., and Z.-X. Li, 1991:  Sensitivity of an atmosphericgeneral circulation model to prescribed SST changes:  feedback effectsassociated with the simulation of cloud optical properties. Climate Dynamics,5, 175-187.

Malvagi, F., N. Byrne, G. Pomraning and R. C. J. Somerville, 1993: Stochasticradiative 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 observationsand climate models. J. Climate, 9, 1683-1697.

Roeckner, E., U. Schlese, J. Biercamp and P. Loewe, 1987:  Cloudoptical depth feedbacks and climate modelling. Nature, 329, 138-140.

Senior, C. A., and J. F. B. Mitchell, 1993: Carbon dioxide and climate:the impact of cloud parameterization. J. Climate, 6, 393-418.

Slingo, J., 1987: The development and verification of a cloud predictionscheme for the ECMWF model. Quart. J. Roy. Meteor. Soc., 113, 899-927.

Smith, R. N. B., 1990:  A scheme for predicting layer clouds andtheir water content in a general circulation model. Quart. J. Roy. Meteor. Soc., 116, 435-460.

Somerville, R. C. J., 2000: Using single-column models to improve cloud-radiationparameterizations.  General Circulation Model Development:  Past,Present and Future, D. A. Randall (ed.), Academic Press, 641-657.

Tselioudis, G., W. B. Rossow, and D. Rind, 1992: Global patterns ofcloud optical thickness variation with temperature. J. Climate, 5, 1484-1495.