Instituteof Physics

Climate Change PredictionA challenging scientific problem

2 Instituteof Physics 2005

This paper, produced on behalf of the Institute of Physics by Professor Alan J. Thorpe, explains how predictions of future climate change are made using climate models. It is hoped that the paper will increase believability in

these models and be persuasive that anthropogenic activity is likely to be causing global warming. It aims toconvince policy-makers, the general public and the scientific community that the threats posed by global climate

change are real.

Professor Thorpe is currently the Director of the Natural Environment Research Council (NERC) Centres forAtmospheric Science (NCAS), based at the Department of Meteorology at the University of Reading, where he

serves as a professor in Meteorology. In April 1999 he took a two year leave of absence to become Director of theHadley Centre at the Met Office. As of April 2005, he will take a leave of absence of four years to become

Chief Executive and Deputy Chair of NERC.

3Instituteof Physics 2005

Climate Change PredictionA challenging scientific problem

ByProfessor Alan J. Thorpe

Climate change is one of those issues where science is crucial in determining government andinternational policy-making. Like the weather, everyone has a view on climate change but, as willbe discussed, not all such views, such as the one reproduced below, are equally defensible onscientific grounds:

“The claim of man-made global warming represents the descent of science from the pursuit of truth into politicised propaganda.The fact that it is endorsed by the top scientist in the British government shows how deep this rot has gone.” Melanie Phillips, DailyMail, 12 January 2004.

Climate change is a fundamental problem involving basic science including physics. There is muchresearch still to be done before we get to a position of sufficient certainty about all the aspects ofclimate change that are required by society to plan for the future. Predictions of future climatechange, based on numerical global climate models, are the critical outputs of climate science.Whilst much has been written about the details of the predictions themselves, scepticism about theprediction models is rife and this is why this paper is devoted to de-mystifying the predictionmethodology. Consequently this paper focuses on the scientific basis of climate change prediction.As for all problems in science, uncertainty and its quantification are a fundamental part of thescientific process and thus they will figure largely in this paper. There is little doubt that a lack ofknowledge about how climate change is predicted and the associated uncertainties are amongst themain reasons for ill-informed comment on climate change.

4 Instituteof Physics 2005

What is climate? Evidence of change

Weather is the state of the atmosphere at a giventime whilst climate is the average weather over aperiod of time. The state of the atmosphere isusually given in terms of its temperature,atmospheric composition (for example, watervapour, liquid water or carbon dioxide content),wind speed and direction, pressure and density. Inaddition the intensity of solar and terrestriallyemitted radiation are fundamental determiningfactors. The characteristic timescale of thevariability of the atmosphere at various spatialscales is critical in deciding whether and how thefuture state of the weather and climate might beforecast. It is the presence of relatively slow timeand large space scale phenomena in the atmospherewhich means both that accurate forecasts can bemade and that society can benefit from them.

As the atmosphere interacts with the underlyingsurface – oceans, land, and ice – the term climatesystem is used to encompass both the atmosphereand the influence of the Earth’s surface on climate.Understanding and predicting both the climate andother properties of the atmosphere, the surface andsub-surface media are now referred to as EarthSystem Science reflecting a holistic view of thesystem. The long timescale of mixing and transportof heat in the oceans is a key factor in determiningthe timescale of climate variations.

Climate is a time-average of weather, but whatmakes it complex is that this average varies in time.Clearly if we average over the complete history ofthe Earth’s atmosphere there is a single climate state.However any finite average varies significantly onall longer timescales. The reason for these variationsis crucial in understanding the physics of climateand of climate change. It is commonplace to look atclimate averages over weekly, monthly, seasonal,annual, decadal, centennial, millennial and longertime periods and figure out how climate, so defined,varies on all longer timescales.

Knowledge of past variations in the Earth’s climatehas been acquired from a wide variety of both directmeasurements and other indirect or proxyinformation. This reconstruction of the climaterecord shows that climate, for example, annual ordecadal averages, has varied significantly on a widerange of timescales. Information about the variation

of temperature in Antarctica is available from theVostok ice core records over the past 400,000 years;see Figure 1. This is a period that covers only about0.01% of the lifetime of the Earth’s atmosphere.During these 400,000 years there are variations inAntarctica from about -10ºC to +4ºC relative topresent day values. The periods with the lowesttemperatures are the ice ages whilst warmer epochs,such as at present are interglacials. There has been arelatively regular pattern of four ice ages and fiveinterglacials over this period. The transition from anice age to the warmest temperatures in the followinginterglacial is relatively rapid (~10,000 years or less)followed by a much more gradual cooling,interrupted by significant fluctuations, over~100,000 years or more towards the next ice age.

Whilst important for understanding the causes ofclimate change, such timescales are of less interestto the development of human societies.Considerable effort has been devoted to using proxyinformation, such as from tree rings, to estimatechange over the past 1000 years or so and theclimate appears to have been noticeably lacking insignificant variations. The so-called instrumentalrecord period, when there have been acceptedmeteorological instruments that can be utilised tomeasure climate, has been available since about1860, which is a very small fraction of the lifetimeof the Earth’s atmosphere. There is an acceptedglobal change over the last 100 years of nearly0.8ºC (with an uncertainty of ±0.2ºC, 95%confidence interval) in the global-average nearsurface temperature, with this rise – a.k.a. globalwarming – focused into the periods 1910 to 1945

Figure 1: The variation of air temperature (red), carbon dioxide(blue) and methane (green) content over the last 400,000 years

(present day on left of the axis) as deduced from the VostokAntarctica ice core information.

5Instituteof Physics 2005

and 1980 to present; see Figure 2. Regionally in2004 this translates into a maximum warmingrelative to the end of the 19th century over, forexample, parts of the Arctic by about 5ºC andoverall the land has warmed at about twice the rateof the oceans. These variations are noteworthy notleast as they have occurred over a short time periodand society is increasingly vulnerable to climatechange both because of the huge increase in the totalpopulation and because of the way it organises itself.

The physics of climate

Weather and climate exist because of the distributionof energy from incoming short-wave solar radiationthat drives the circulation of the atmosphere relativeto the underlying surface. For a steady state, climateproperties such as temperature can be supposed toresult from a long-term equilibrium betweenreceived energy from the Sun and outgoing energyemitted by the warm planet. The existence of anatmosphere that is capable of absorbing and re-transmitting certain wavelengths in the electro-magnetic spectrum means that there must be a so-called greenhouse effect whereby the atmospheretraps outgoing infrared radiation, thereby increasingthe atmospheric temperature (see box insert andAndrews 2000). This was first postulated by Jean-Baptiste Fourier in 1827 and further elaborated uponby John Tyndall in 1860 and Svante Arrhenius in1896. It was Arrhenius who first noted that, say, adoubling of the carbon dioxide concentration in theatmosphere could lead to an increase in surfacetemperature of some 5 to 6ºC.

Atmospheric absorption in the infrared is caused bythe presence in the atmosphere of gases (so-calledgreenhouse gases) such as water vapour, carbon

dioxide and ozone. These gases occur in the present-day atmosphere in small concentrations and so theyare referred to as minor constituents. Pre-industrialconcentrations of carbon dioxide were about 280parts per million (ppm) whilst the current day valueis around 370ppm – the difference being attributableto human emissions from, for example, the burningof fossil fuels; see Figure 3.

To simplify significantly, the equilibriumtemperature at the surface depends on three factors:the concentration and vertical distribution of theminor constituents that determine the magnitude ofthe greenhouse effect, via τ ; the Sun’s output ofradiation, via S; the reflectivity of the Earth to thatincoming solar radiation determined by surface,aerosol and cloud properties, via a. These threeparameters tell us a lot about the climate changeproblem. Solar output leading to S is not a constantbut note that it is independent of the atmosphere.The concentration of minor constituents is beingchanged by human activities such as fossil fuelburning but also by changes in the flora and faunaand volcanic out-gassing etc. The planetaryreflectivity, or albedo, depends on the internaldynamics and physics of the atmosphere and inparticular the cloud content, as well as land use,which is controlled by human activities. The humaninput of aerosols to the atmosphere reflects backincoming solar radiation and may make clouds morereflective – it is thought this has acted to partiallyoffset the amount of global warming (sometimescalled global dimming). The fact that the amount ofcloud is altered, in principle, by temperature showsthat there is the possibility of feedbacks in theclimate system. Other feedbacks include: (i) themelting of sea-ice leading to reduced albedo and

Figure 3: The carbon dioxide content of the atmosphere asmeasured at Mauna Loa and from ice cores from 1750 to the end of

the 20th century. Figure 2: The global average near-surface temperatures from 1861

to present relative to the value at the end of the 19th century.

6 Instituteof Physics 2005

further warming, and (ii) higher temperaturesleading to more atmospheric water vapour and anenhanced greenhouse effect. It is the ability ofhumans to alter the greenhouse effect that shows thatthe term “anthropogenic climate change” is ameaningful concept.

The concentration of greenhouse gases also variesnaturally over the geological history of the Earth.Over the period of the Vostok ice core record, backto 400,000 years ago, levels of carbon dioxide are

thought to have varied between about 180 and280ppm with this variation mirroring that of the iceage-interglacial temperature cycles. It is believedthat it is only prior to about 20 million years ago thatcarbon dioxide levels exceeded current day valueswith epochs hundreds of million years ago whenconcentrations were in excess of 5000ppm.

If there is a human-induced climate modificationthen present-day climate variations are a mixture ofnatural and anthropogenic contributions. The

SFC

τ consistent with current concentrations of greenhouse gases might be 0.2 and so equation (5) gives a crudeestimate of the globally averaged surface temperature, T , in the presence of the greenhouse effect, of287K, i.e., a greenhouse effect that warms the surface by about 32ºC. If the atmosphere were entirely opaquein the infrared we would have, = 0. Then if there was no change in albedo in this simplified model,T ~ 303K.

Using radiative transfer theory we can relate the intensity of outgoing infrared radiation at a givenfrequency at the top of the atmosphere, ITOA, to surface and atmospheric properties:

(1)

where ISFC is the infrared radiation emitted by the Earth’s surface at temperature TSFC; τ is the atmospherictransmittance at a given frequency; B(T) is the Planck function; T(z) is the height (z) dependentatmospheric temperature; W(z) is the height dependent weighting function governed by the verticaldistribution of constituents of the atmosphere responsible for absorption and emission of infrared radiationby the atmosphere itself. The fact that τ is less than unity (and W ≠ 0), because of the presence in the

If we simplify equation (1) by integrating over frequency, assume that the atmosphere has a characteristic(infrared brightness) temperature, TA, and use Stefan-Boltzmann’s law, then for radiative equilibrium:

444 )1( ASFCe TTT στστ −+= (2)

where τ is an average transmittance and the emission temperature Te is given by:

4

4

)1(

σSa

Te

−= (3)

where S is the total solar irradiance at the mean distance of the Earth from the Sun (~ 1367Wm-2), σ isStefan-Boltzmann’s constant and a is the planetary reflectivity, or albedo, to incoming solar radiation(~ 0.3). This albedo arises from reflection of solar radiation from bright surfaces such as snow and clouds;note that the atmosphere is relatively transparent to incoming solar radiation. Thus T ~ 255K and this would be the chilly temperature at the Earth’s surface if there wasn’t an atmospheric greenhouse effect (i.e.,if = 1). In fact the atmosphere absorbs up-welling infrared radiation and radiates infrared both upwardsand downwards. Hence:

τ

44 )1()1(2 SFCA TT στστ −=− (4)

Substituting equations (4) into (2) gives the following estimate for the temperature of the air near thesurface of the Earth:

(5)

Note that this simplified description (wrongly) ignores other heat transfer mechanisms. A value of

atmosphere of absorbing greenhouse gases, means that the atmosphere exerts a significant influence on theglobal temperature – the greenhouse effect.

τ

I = I + ∫ B(T) W(z) dz TOA SFC

ZTOA

0

σ

44

)1(2

)1(

)1(

2

σττ +−=

+= Sa

TT eSFC

Greenhouse Effect

SFC

e

SFC

τ

7Instituteof Physics 2005

detection of climate change relies on measurementsof recent past climate variations and the attributionof climate variations to anthropogenic sourcesattempts to find the contribution to observed orpredicted change from these sources. Given that wehave only one climate system to measure, it isextremely difficult to be definitive about attributionalthough this can be done using statistical andmodelling approaches. The problem with attributionbeing that a natural trend can exist over certain timeperiods, even without human modifications, as partof a longer-term natural oscillation. It is particularlypopular to ask questions such as “is the recent severeweather event caused by global warming?” andextremely difficult to answer them definitively.

How is weather and climate forecast?

Whilst the debate about why climate change hasbeen happening is often heated, that concerningpredictions of future climate change can be vitriolicin the extreme. Being able to predict the outcome ofan experiment is the touchstone for whether we haveunderstood the underlying physics. So how do wepredict climate change?

Weather forecasting is the best place to start becausethe forecasts are more familiar and the methodologyis very similar in many important ways to climateprediction. A weather forecast involves numericallyintegrating forward in time equations that describethe evolution of the atmosphere starting from a set ofinitial conditions. The equations used are theclassical laws of (fluid) mechanics andthermodynamics that are known to apply well to theatmosphere. The numerical solutions require theatmosphere to be divided up into a large three-dimensional lattice of grid points at which theatmospheric variables are held in the model and onwhich the equations are solved using finitenumerical approximations. The initial conditionsarise from global measurements of the state of theatmosphere interpreted using a prior short-rangeforecast of that state using the model forecastsystem. The measurements have an uncertaintybecause they are: (a) insufficient in number toinitialise every variable at every grid point, (b)usually not located at grid points, and (c) have adegree of measurement error.

From the viewpoint of knowing whether the physicsin the model is correct, the fact that weather

forecasting has been done routinely every day forabout the last thirty years is significant. Many, manyweather states have been forecast over that periodand so experimental repetitions certainly exist –although one could argue that few states have beensampled compared to the lifetime of the atmosphere!On the whole, weather forecasts are both incrediblysuccessful and useful to society, which is why theycontinue to be produced even though they involveconsiderable expense. Numerical weatherpredictions have an element of uncertainty andmeteorological science has devoted considerableeffort to understand why. The answer has led to amajor shift in the way classical physics isunderstood.

The numerical model introduces uncertaintiesbecause of the finite approximation to thecontinuous equations. This approximation has tworelated aspects – one that there is a truncation errorbecause of the numerical method and the otherbecause the effects of scales of motion smaller thanthe grid resolution (the distance betweenneighbouring grid points) on the resolved scale flowmust be included. The representation of these subgrid-scale effects is called parametrisation.Secondly the measurements of the atmosphereintroduce an uncertainty because of their insufficientnumber and inherent inaccuracies. Uncertainty ininitial conditions means that two forecasts startedfrom almost identical initial conditions will divergeslowly at first but then radically such that the twopredicted states become as different from oneanother as two observed states picked at random.This sensitivity to initial conditions and to modelformulation, are prime examples of chaos, aboutwhich much has been written following on from itsintroduction into modern physics by themeteorologist Ed Lorenz; see Figure 4. Therealisation of the pervasive relevance of chaos tophysics has been a reminder that whilst physicistsmight like to imagine that classical physics isunderstood in principle, there are fundamentalaspects still left to be uncovered in its practice.

Knowledge of uncertainty has been turned to ouradvantage. By calculating a set of multiple forecasts– an ensemble – the members of which differ onlyslightly in their initial conditions and in theirparametrisations, forecasts can now not only predictthe most likely future weather but also the risk thatnature will deviate from this most likely predicted

8 Instituteof Physics 2005

state. So we are now in the era of predicting both themost likely weather to come but also the chance thatthe forecast is wrong; see Figure 5. This ability topredict in a probabilistic way represents a profoundadvance in the science that is often overlooked ormisinterpreted.

On both empirical and theoretical grounds it isthought that skilful weather forecasts are possibleperhaps up to about 14 days ahead. At first sight theprospect for climate prediction, which aims topredict the average weather over timescales ofhundreds of years into the future, if not more doesnot look good! However the key is that climatepredictions only require the average and statistics ofthe weather states to be described correctly and nottheir particular sequencing. It turns out that the waythe average weather can be constrained on regional-to-global scales is to use a climate model that has atits core a weather forecast model. This is becauseclimate is constrained by factors such as theincoming solar radiation, the atmosphericcomposition and the reflective and other propertiesof the atmosphere and the underlying surface. Someof these factors are external whilst others aredetermined by the climate itself and also by humanactivities. But the overall radiative budget is apowerful constraint on the climate possibilities. Sowhilst a climate forecast model could fail to describethe detailed sequence of the weather in any place, itsclimate is constrained by these factors if they areaccurately represented in the model. Physicists areused to such ideas – the kinetic theory of gasesallows average properties of the system to beaccurately quantified even though the location andmomentum of each molecule need not, and cannot,be predicted accurately.

The analogy with the kinetic theory of gases tells usthat it is probably possible to describe certain grossaspects of climate (for example, global averagenear-surface air temperature) without recourse todetailed numerical models and the radiative transfercalculation given earlier is an example of this. But ifwe want to predict local properties of the climatesystem and their evolution in time, we need to use anumerical climate model.

How does a climate model differ from a weathermodel? Currently computational limitations restrictclimate models to be run with grid pointssignificantly further apart than weather models.

Figure 5: Examples of an ensemble of ten-day weather forecasts(each red line) for a period in June in two successive years for

London using the ECMWF ensemble prediction system. In one year(upper panel) the atmosphere is relatively predictable and all the

members of the ensemble give similar predictions that mirror whatactually happened (dark blue line) whereas for the other year (lower

panel) there are a wide range of predictions within the ensembleshowing that the risk of significant departures from mean conditions

is high, indicating a more unpredictable regime.

Figure 4: The Lorenz strange attractor indicating the evolution of adynamical system with two “attractors” located at the centre of

each of the “butterfly” wings – these could represent a cold and awarm period of weather. Each yellow dot represents the state of the

atmosphere at that particular time. The superimposed evolvingellipse indicates the spread of forecasts if the atmosphere resided atthat particular location on the attractor at the start of the forecast.These forecasts show that the flow can be more or less predictabledepending on the particular initial conditions for the forecast. Forexample the bottom right panel produces forecasts that predict the

weather to be equally likely to be cold or warm.

9Instituteof Physics 2005

Also a climate model requires the interactionsbetween the atmosphere, the oceans and theland/ice surface to be included. The atmosphericpart is a global weather model, extended to be ableto include the temporal evolution of keyatmospheric constituents, whilst the oceancomponent consists of a similarly structured fluid-dynamical model of ocean properties such ascurrents, heat and composition. The land/ice surfaceproperties are included particularly as theydetermine the reflectivity and other key aspects ofthe climate system.

Putting these components together to make a climatemodel is a complex task. Critical for belief in thescience of climate change predictions is thedemonstration that such climate models are capableof producing accurate predictions for the rightreasons. Physicists will want to describe a givensystem/experiment with theories/models with arange of complexity and it is the consistency ofunderstanding amongst these theories/models that ispowerful in allowing us to believe that we have acredible physically consistent explanation. The factthat we have simplified descriptions, such as thatbased purely on radiative considerations, for thegreenhouse effect is crucial and enhances thecredibility of the predictions from detailed climatemodels.

There are, of course, significant uncertainties in theparametrised physical, chemical and biologicalprocesses in climate models. Physical processesrepresent the difficult parts of the physics – we aresure that Newton’s laws of motion apply but theprecise details of how to represent the multiplicity offorces that exist and how they depend onatmospheric, and other, properties is extremelychallenging. One example, of many, concerns thedrag coefficient that determines the degree offrictional retardation of the flow as it moves acrossthe rough Earth’s surface. This is known to within atolerance of say, ± 10%. The method to explore theway in which such uncertainties in the model effectthe predictions is to create a climate ensembleprediction in much the same way as is done forweather forecasting. The climate ensemble uses a setof similar but slightly different initial conditions andmodel formulations to span this uncertainty space.Then the risk of possible climate change can beevaluated explicitly. The uncertainties of the initialconditions and model formulations are a measure of

the level of knowledge we have about the physics ofthe climate system. However quantifying the effectof these uncertainties on the climate predictionsthemselves is a vital aspect of the scientific method.Many commentators imagine that this implies wecannot believe the output of climate models. Butthey have misunderstood the critical nature ofuncertainty and risk in science, particularly thescience that feeds directly into policy.

How can we evaluate whether these climatepredictions are credible – or to put it another way,how can we estimate the information content of theclimate forecast? The fact that a weather forecastmodel is at the heart of the climate model and thisis independently shown to be valid is an importantfeature. Also climate models can be run for currentday climate conditions assuming no further human-produced increase in greenhouse gases. Thisshould reproduce the statistics of current climateand not drift because of imperfections in theoverall modelled radiation balance. In addition themodel can be run including the known varyingconcentration of greenhouse gases over the period1850 to present and other external forcing factorssuch as volcanic eruptions and variability in solaroutput (also used to simulate past climates in thegeological record). The “hindcasts” made in theseways reproduce many aspects of the (relativelywell) observed climate over that period. Indeed it ispossible to show, by including each forcingseparately, that the solar variations that haveoccurred cannot explain the recent few decades ofwarming and that human input of greenhouse gasesare most likely responsible for this warming; seeFigure 6. A more complete ensemble may show,however, that there are a number of different wayssuch climate models can balance processes to passthis test.

It is sometimes imagined that such tests of themodel are rigged in some way because, as has beendiscussed, there are various empirical and otherparameters in the parametrisation components ofthe model that are to a degree uncertain and soperhaps these could be tuned to get the “right”answer. It is wildly over-simplistic to suppose thata few parameter values can be adjusted toreproduce the many diverse attributes thatconstitute the complex behaviour of the climatesystem. If it were possible to do this we wouldindeed have emerged with a climate “theory of

10 Instituteof Physics 2005

everything”! The very complexity of the model(maybe three-quarters of a million lines ofcomputer code) tells us that it is almost certain tobe impossible to cheat in this way.

It is clear, however, that climate forecasting does nothave the luxury of the quantity of multipleindependent forecasts to prove its overall level ofaccuracy that is available to weather forecasting. Inaddition we may need to wait for some time beforewe can test fully the predictions of future climatechange. But this does not mean the predictions arenecessarily either inaccurate or not credible as issometimes implied. There is little doubt that there isstill some way to go to simulate accurately all facetsof the climate system with such models. The era ofensemble climate prediction is only just beginning

now. But we will soon be able to say morequantitatively what level of uncertainty we attach topredictions of various facets of the climate system –the fact that some are more uncertain than othersdoes not mean that all predictions are to be treated asworthless.

What are the future prospects for reducinguncertainty in climate predictions? There are verysubstantial improvements that will be madepossible by increasing the resolution of the modelsutilising next generation supercomputer power. Asreported recently in the press, UK scientists arecollaborating with Japanese colleagues to use theEarth Simulator supercomputer in Yokohama tocarry out ground-breaking climate simulations. Thismeans we will have, for the first time, sufficient

Figure 6: Hindcast of twentieth century global temperature record with the Hadley Centre model. The red line is from the observations and the grey bandis the range of model predictions; (a) includes only natural climate forcings such as solar output and volcanoes, (b) includes only human input of

greenhouse gases and aerosols, (c) includes all forcings.

11Instituteof Physics 2005

information to provide realistic estimates of thechange in the frequency and intensity of the weathersystems under climate change. This is criticallyimportant, as climate impact on society cruciallydepends on these extreme weather events. Inaddition scientific knowledge is accumulatingrapidly on the critical areas of uncertaintyassociated with biogeochemical cycles and oceancirculation and this will be incorporated in themodels as soon as it is available.

Predictions of climate change over the nextcentury and climate change policy

The importance of climate change to society makesthe stakes very high in making predictions withclimate models. This fact, amongst others, led to theestablishment by the World MeteorologicalOrganization and the United Nations EnvironmentProgramme of the Intergovernmental Panel onClimate Change (IPCC), in 1988. IPCC has issued aseries of assessments including the state of thescience of climate change. This has involved aprocess of drawing together all published researchand assessing the degree to which there is aconsensus and to identify the areas of remaininguncertainty. In 2001, IPCC issued its thirdassessment report (TAR) of the science of climatechange, and this is widely used as the authoritativeview of the predictions and of the science. Scientistsand policy-makers use it as a reference; see IPCC2001a and 2001b. The TAR describes the level ofuncertainty with statements such as “it is likely” or“it is very likely that…” where these words have apercentage of likelihood associated with them (66-90% and 90-99% chance respectively). Theseestimates are based on expert judgement but asensemble climate prediction develops we expect tohave more objective criteria. There are facets of thefuture evolution of the climate system that we can bereasonably sure about whereas other facets are muchmore uncertain. For example, there are soundmeteorological reasons why rainfall predictionshave a larger degree of uncertainty than those fortemperature. This is because rainfall is highlyvariable in space and so the relatively coarse spatialresolution of the current generation of climatemodels is not adequate to fully capture thatvariability. But this does not mean we cannotbelieve, for example, the larger scale aspects of thewater cycle in the models. This cycle depends on theaverage effects of rainfall systems, amongst other

things, and climate models are capable of capturingthis with an acceptable level of accuracy. Thisvariation in accuracy depending on which propertyof the system is being considered is perhaps onereason why some commentators are too negativeabout the accuracy of climate predictions.

The spatial resolution of current climate models is,to a large degree, determined by the availability ofsupercomputer resources. The models assessed inthe TAR typically used a horizontal resolution ofabout 250km. The typical horizontal scale of thecyclonic storm systems, that are such a ubiquitousfeature of tropical and extra-tropical weather, isabout 1000-2000km but with significant sub-structure occurring at fronts, for example, on muchsmaller scales. To describe comprehensively thedetails of the interaction of such weather systemsand the large-scale global circulation of theatmosphere, it is thought necessary to have a modelresolution in the order of 100km or better.Consequently there is currently low confidence inthe ability of these models to quantify the change tothe frequency and intensity of such storm systems.As most of the impacts of climate change on societyand the economy arise from the winds and rainfall(such as leads to local flooding) associated withsuch systems, this remains a key research problem tobe addressed. The relentless advance in the power ofcomputers means that we are in sight, over the nextdecade or two, of being able to simulate the globalclimate with a horizontal lattice with grid pointsseparated by as little as a few kilometres, thusremoving the need for many of the parametrisationsused currently. This will help reduce substantiallymajor uncertainties such as those associated with theeffects of clouds.

It is well known that a range of climate models,using a range of representative scenarios for thecontinuing human input of greenhouse gases andaerosols, predict a global-average surfacewarming by 2100, relative to 1990, of betweenabout 1.4 and 5.6ºC; see Figure 7. The modelspredict many other regional properties of thesystem, such as temperature, wind speed anddirection, humidity, rainfall, snowfall, sea level,sea-ice and ocean currents. The warming is notuniform, having a general increase towards thePoles, particularly the North Pole; see Figure 8.This pole-ward increase may be connected to thereduction of sea-ice as warming occurs and the

12 Instituteof Physics 2005

consequent positive feedback as the reflective icesurface is replaced by the essentially non-reflective sea surface. Also the near-surface airover the land warms more than that over theoceans. As temperature rises near the surface theair holds more moisture and the hydrological cyclebecomes more intense.

The ocean circulation is also predicted to changeand sea level is expected to rise at a rate of about1.7mm/year as the ocean expands (and glaciers meltenhancing river flow) as heat is gradually diffuseddownwards in the ocean. Predictions of sea level riseshow that, perhaps surprisingly, it variessubstantially regionally but also the rise willcontinue for several hundred years even if we couldstop emitting greenhouse gases into the atmospherenow. This is because of the commitment to sea levelrise arising from warming of the atmosphere that hasalready occurred. This feature is one of the mostrobust and potentially damaging aspects of thepredicted change to the climate system arising fromhuman activities.

An important aspect of the ocean circulation is thethermohaline circulation (THC) that is driven byspatial variations in the density of sea water, whichis determined by its temperature and salinity. The

North Atlantic Ocean plays a fundamental role in thedevelopment and maintenance of the THC. Warmsalty upper ocean water moves northward andeastwards towards northwest Europe and this isknown as the Gulf Stream. The Gulf Stream and theatmospheric storms moving across the NorthAtlantic, transport heat north and keep the climate ofnorthwest Europe relatively warm compared toother places at similar latitudes. This water is thencooled as heat is given up to the atmosphere. Thewater then sinks rapidly in the Greenland and theLabrador Seas before returning south at depth. Thisis the Atlantic component of a global THCcirculation; see Figure 9. Comprehensive climatemodels predict that the intensity of the THC mayweaken by as much as 30% as climate warms but notactually shut down. This is important for the climateof northwest Europe as a complete shutdown wouldimply a local reduction in the predicted warmingand in fact may actually lead to a cooler climate.Despite the 30% reduction, current models suggestthat there will be an overall warming of the climatein this region. There have been periods in thegeological past when climate has cooled relativelyrapidly, for example, the Younger Dryas period, and

Figure 7: Projections for the Earth’s surface temperature from theIPCC TAR over a wide range of scenarios of greenhouse gasemissions and climate models. The graph also includes the

temperature inferred from measurements over the last 1000 yearsshowing the relatively slow variations over that period. Figure 8: Northern-hemispheric regional variations of surface

temperature (the global mean warming is about 3.5K), averagedover a twenty-year period, in a simulation of climate change caused

by a doubling of carbon dioxide levels (from Sarah Keeley). Thisshows regions with substantially greater and lesser warming than

the global average.

13Instituteof Physics 2005

this may be, at least in part, explained by changes tothe THC. It is still an open scientific question as tothe circumstances in which the THC might shut offcompletely and how rapidly this might happen.There has been speculation that this could occur asrapidly as a few tens of years but this is highlyuncertain at present.

The human input of carbon dioxide needs to be putinto the context of the natural carbon cycle. Currentclimate models attempt to represent the carbon cyclewith, most recently, the inclusion of a dynamicvegetation component that allows for feedbacksbetween the biosphere and climate change. The effectof land use changes, such as deforestation and thepotential partial amelioration of global warming byreforestation, can be included in the models. Currentknowledge suggests that the carbon cycle is verysensitive to climate change with, for example, a recentHadley Centre calculation showing that dieback of theAmazon rainforest because of a reduction of rainfall aswarming occurs, represents a major potential positivefeedback as the carbon from the forest is released backinto the atmosphere. Such dieback processes representanother potentially rapid (and large) climate change

but the risk of this happening is still very difficult toquantify because of uncertainties in thebiogeochemical feedback processes.

Societal vulnerability to climate change, includingthat caused by human input of greenhouse gases, ispotentially large depending on the characteristicsand organisation of each region. This led to thecreation of the United Nations FrameworkConvention on Climate Change. Arising from this isthe Kyoto Protocol that seeks to cut each country’sgreenhouse gas emissions to a level that in total,within the commitment period 2008-2012, will be5% less than 1990 levels. This international andlegally binding agreement on signatory countries inthe developed world entered into force in February2005. The effect on atmospheric concentrations ofcarbon dioxide is likely to be small. However theKyoto Protocol is regarded as being highlysignificant from a political viewpoint even if theamelioration of global warming is likely to be toosmall to make a real difference. A sustainedreduction of emissions would require major changesto the way in which, for example, energy andtransport are structured.

Figure 9: A schematic of the global conveyor belt thermohaline circulation in the ocean.

14 Instituteof Physics 2005

In conclusion

So why do commentators imagine that top scientistsare deluded about anthropogenic climate change?The stakes are high and rarely are scientists undersuch scrutiny. Scientists are appalled that they couldbe suspected of distorting the evidence to enhancetheir reputations or funding opportunities. Ofcourse scientific hypotheses and analysis can berefuted by later discoveries but this is not the sameas complicity. The fact that everyone experiencesweather and climate may explain why non-scientists feel confident in attempting to refute thescientific evidence. The complexity of the climatesystem and its many interacting and compensatingphysical processes means that simple argumentsthat gloss over this complexity have to beapproached with a significant degree of scepticism.A common method of arguing starts by identifyinga single cause or physical process that either has notbeen included or has been included in an imperfectway, into climate models. But the climate changesbecause of a multiplicity of interacting processesand any one process alone cannot be the wholestory. The search for the one and only cause ofclimate change is doomed to failure. Climatemodellers attempt to include in the models all theprocesses that are even remotely likely to have adetectable effect – any newly discovered processwill quickly find itself incorporated into themodels!

The multitude of non-scientist comments and viewsabout climate change is important and to bewelcomed as they represent society’s engagementand participation in the scientific process. An

exciting initiative in this regard is the availabilityfrom http://climateprediction.net of a comprehensiveclimate model that can be run on anyone’s personalcomputer in the background to predict climatechange using idle time on the computer. The resultsfrom this very large distributed-computing ensemblewill be used to evaluate, in an improved way, theuncertainty in current predictions of climate change;see Figure 10.

Few if any scientific problems have had such a hugedegree of scrutiny by specialists and non-specialistsand, whilst one can never say never, it would seem tobe perverse not to take the risk of human-inducedclimate change very seriously indeed. At the veryleast the possibility of human-induced climatechange is a known unknown but arguably it is closeto making the transition to becoming a knownknown!

Figure 10: The predicted frequency (a measure of risk) of variouslevels of global warming for a doubling of carbon dioxide, usingover 2000 members of a climate prediction ensemble produced by

climateprediction.net; from Stainforth et al 2005.

15Instituteof Physics 2005

Acknowledgements:

I would like to acknowledge the assistance of ChrisBrierley and Sarah Keeley in the preparation of thispaper and of Tajinder Panesor at the Institute ofPhysics for commissioning it. Thanks go to DavidAndrews, Keith Shine and anonymous reviewers forhelpful comments on an earlier draft of the paper.

Sources for figures:

Figure 1: Petit, J.R. et al 1999: “Climate andatmospheric history of the past 420,000 years fromthe Vostok ice core, Antarctica”. Nature, 399 pp 429-36; and PAGES IPO.

Figure 2: The Hadley Centre for Climate Predictionand Research.

Figure 3: Climate Change 2001: The Scientific Basis.Contribution of Working Group I to the ThirdAssessment Report of the Intergovernmental Panelon Climate Change (IPCC). Cambridge UniversityPress, UK.

Figure 4: ECMWF Model Forecasts.

Figure 5: ECMWF Model Forecasts.

Figure 6: Climate Change 2001: The Scientific Basis.Contribution of Working Group I to the ThirdAssessment Report of the Intergovernmental Panelon Climate Change (IPCC). Cambridge UniversityPress, UK.

Figure 7: Climate Change 2001: Synthesis Report. Acontribution of Working Groups I, II and III to theThird Assessment Report of the IntergovernmentalPanel on Climate Change (IPCC). CambridgeUniversity Press, UK.

Figure 8: Department of Meteorology, University ofReading.

Figure 9: Climate Change 1995. The SecondAssessment Report of the Intergovernmental Panelon Climate Change (IPCC). Cambridge UniversityPress, UK.

Figure 10: Stainforth, D. A. et al 2005: “Uncertaintyin Predictions of the Climate Response to RisingLevels of Greenhouse Gases”. Nature, 433, pp 403-06.

References:

Andrews, D. G., 2000: An Introduction toAtmospheric Physics. Cambridge University Press,UK.

Climate Change 2001: The Scientific Basis.Contribution of Working Group I to the ThirdAssessment Report of the Intergovernmental Panelon Climate Change (IPCC). Cambridge UniversityPress, UK.

Climate Change 2001: The Scientific Basis.Summary for Policymakers and Technical Summaryof the Working Group I Contribution to the ThirdAssessment Report of the Intergovernmental Panelon Climate Change (IPCC). Cambridge UniversityPress, UK.

Stainforth, D. A. et al 2005: “Uncertainty inPredictions of the Climate Response to Rising Levelsof Greenhouse Gases”. Nature, 433, pp 403-06.

76 Portland PlaceLondon W1B 1NT

Telephone: +44 (0) 20 7470 4800Facsimile: +44 (0) 20 7470 4848

Email: physics@iop.org Website: www.iop.org

Registered Charity No. 293851

Instituteof Physics

The Institute of Physics is a leading international professional body and learned society, with over 37,000 members,which promotes the advancement and dissemination of a knowledge of and education in the science of physics,

pure and applied