Forty Years Of Numerical Climate Modelling

Preparing to load PDF file. please wait...

0 of 0
100%
Forty Years Of Numerical Climate Modelling

Transcript Of Forty Years Of Numerical Climate Modelling

INTERNATIONAL JOURNAL OF CLIMATOLOGY Int. J. Climatol. 21: 1067–1109 (2001) DOI: 10.1002/joc.632
FORTY YEARS OF NUMERICAL CLIMATE MODELLING
K. MCGUFFIEa,* and A. HENDERSON-SELLERSb a Department of Applied Physics, Uni6ersity of Technology, Sydney, Australia b Australian Nuclear Science and Technology Organization, Sydney, Australia
Recei6ed 14 April 2000 Re6ised 21 August 2000 Accepted 1 No6ember 2000
ABSTRACT
Climate modelling is now a mature discipline approaching its fortieth birthday. The need for valid climate forecasts has been underlined by the recognition that human activities are now modifying the climate. The complex nature of the climate system has resulted in the development of a surprisingly large array of modelling tools. Some are relatively simple, such as the earth systems and energy balance models (EBMs), while others are highly sophisticated models which challenge the fastest speeds of the most powerful supercomputers. Indeed, this discipline of the latter half of the twentieth century is so critically dependent on the availability of a means of undertaking powerful calculations that its evolution has matched that of the digital computer. The multi-faceted nature of the climate system demands high quality, and global observations and innovative parameterizations through which processes which cannot be described or calculated explicitly are captured to the extent deemed necessary. Interestingly, results from extremely simple, as well as highly complex and many intermediate model types are drawn upon today for effective formulation and evaluation of climate policies. This paper discusses some of the important developments during the first 40 years of climate modelling from the first models of the global atmosphere to today’s models, which typically consist of integrated multi-component representations of the full climate system. The pressures of policy-relevant questions more clearly underline the tension between the need for evaluation against quality data and the unending pressure to improve spatial and temporal resolutions of climate models than at any time since the inception of climate modelling. Copyright © 2001 Royal Meteorological Society.
KEY WORDS: Archaean; climate models; climate system; deforestation; earth models of intermediate complexity (EMICs); feedbacks; global climate models (GCMs); greenhouse warming; land-use change; last glacial maximum; mid-Holocene; Milankovitch; models; ozone hole; palaeoclimate
1. THE CLIMATE SYSTEM
Today, the atmosphere of planet Earth is undergoing changes unprecedented in human history and, although changes as large as those we are witnessing now have occurred in the geological past, relatively few have happened with the speed which characterizes today’s climate changes (e.g. Pearman, 1992). Concentrations of greenhouse gases are increasing, stratospheric ozone is being depleted, and the changing chemical composition of the atmosphere is reducing its ability to cleanse itself through oxidation (e.g. Keeling et al., 1976, 1995; WMO, 1994; Houghton et al., 1996; MacKay et al., 1997). These global changes are threatening the balance of climatic conditions under which life evolved and is sustained. Temperatures are rising, ultraviolet radiation is increasing at the surface and air pollutant levels are increasing. Many of these changes can be traced to industrialization, deforestation and other activities of a human population that is itself increasing at a very rapid rate (e.g. Bruce et al., 1996; Giambelluca and Henderson-Sellers, 1996; Watson et al., 1996).
Now, for the first time in the history of our planet, emissions of some trace gases from human activities equal, and for some even exceed, emissions from natural sources. This is important for the climate system
* Correspondence to: Department of Applied Physics, University of Technology, Sydney, PO Box 123, Broadway NSW 2007, Australia; e-mail: [email protected]
Copyright © 2001 Royal Meteorological Society

1068

K. MCGUFFIE AND A. HENDERSON-SELLERS

because the atmosphere’s primary constituents, molecular nitrogen and molecular oxygen, are transparent to infrared radiation, and so the greenhouse gases (mainly water vapour, carbon dioxide (CO2), ozone (O3), methane (CH4), nitrous oxide (N2O) and the chlorofluorocarbons, or CFCs, which have no natural sources) present in much smaller amounts, play a major role in the Earth’s energy budget and climate. Trace gases also govern the production and destruction of ozone, affect many biospheric processes, and play other important roles in the climate system. It is of great importance to determine where and how these constituents enter the atmosphere, how they are distributed and transformed by the complex interactions of sunlight, air, land, sea and living organisms; and how they behave in the climate system (cf. Litfin, 1994). Trace gases are carried from their surface sources to the upper atmosphere and around the world by numerous, interdependent processes, such as mixing in the atmospheric boundary layer, vertical exchanges associated with weather systems, moist convection in the mid latitudes, deep convection in tropical storms, and the mean circulation of the atmosphere (e.g. Houghton, 1984).
Possibly the most important species in the atmosphere and hence in the climate system is the hydroxyl radical (OH). Generated by interactions among ozone, water vapour and ultraviolet radiation, OH is the atmosphere’s primary agent of oxidation and the means by which many compounds are transformed into others more readily removed from the atmosphere. The role of the OH radical in climate change is difficult to quantify because it is short-lived in the atmosphere (less than a second), and concentrations can only be inferred by examining the concentrations of other participants in its reactions. The concentration of OH has probably changed since pre-industrial times, as concentrations of CH4, NOx and O3 have increased, but because of the complex chemistry involved and competing pathways of destruction and generation (e.g. Wang and Jacob, 1998) it is difficult to quantify these changes. Climate changes cannot be fully understood without improved determination of the net effect of these complex interactions on the abundance of this dominant oxidant (e.g. Brasseur and Solomon, 1986; McKeen et al., 1997; Carslaw et al., 1999; Kohlmann et al., 2000).
Global climate system changes resulting from human impact on the atmosphere and surface have been described as ‘creeping climate crises’. Their characteristics seem to be that they are slow to develop and, therefore, may not become apparent until their effects have become dangerously far advanced. The iconic demonstration of this dates back to 1985 when British scientists (Farman et al., 1985) discovered that the mean ozone column abundance for October over the Antarctic station of Halley Bay had been decreasing very rapidly since the late 1970s, forming the so-called Antarctic ozone hole. The phenomenon is now a well established feature of the Antarctic atmosphere (e.g. Solomon, 1999). Stratospheric ozone concentrations at the South Pole in spring are now very much less than half of the values of only 30 years ago. Worldwide, stratospheric ozone has declined noticeably (now by a few percent) and, in the Northern Hemisphere, where the stratospheric circulation is more complicated, springtime depletions similar to those over Antarctica have developed over the last decade (McKenzie et al., 1992; Goutail et al., 1999; Hansen and Chipperfield, 1999).
Climate was once defined, rather simply, as ‘average weather’, but ‘the climate system’ has come to be defined more completely over the last few decades. In 1975, the Global Atmospheric Research Programme (GARP) of the World Meteorological Organization (WMO) stated that the climate system is composed of the atmosphere, hydrosphere, cryosphere, land surface and biosphere (WMO, 1975). The United Nations Framework Convention on Climate Change (FCCC), signed in March 1992, coming into force in March 1994, provided an updated definition of the climate system: the totality of the atmosphere, hydrosphere, biosphere and geosphere and their interactions. While these definitions are similar, the emphasis on interactions can be seen to have grown in the 19 years which separate them. The atmosphere, the land surface, the oceans and surface water (the hydrosphere), those parts of the Earth covered with ice and snow (the cryosphere) and the biosphere (the vegetation and other living systems on the land and in the ocean) are all very strongly coupled. This coupled climate system presents a special challenge for modellers, and this has led to a number of very significant volumes which detail the construction of models of the global climate (e.g. Schlesinger, 1988; Trenberth, 1992; Jacobson, 1998; Mote and O’Neill, 2000).
Concepts about the climate system are also concerned with personal and societal issues of habitability and sustainability. Most people evaluate climate in fairly simple terms such as temperature: is it too hot

Copyright © 2001 Royal Meteorological Society

Int. J. Climatol. 21: 1067 – 1109 (2001)

NUMERICAL CLIMATE MODELLING: A REVIEW

1069

or too cold?; chemistry: is the air breathable?; sustenance: is there enough water for drinking and for growing crops; and ambient environment: does it feel comfortable, i.e. not too humid nor too dry? Trying to predict answers to these questions and to the larger question, can this planet continue to sustain life, is the goal of numerical climate modelling. The science of climate modelling, now just 40 years old, is currently being tested in attempts to understand past climates, relate the present climate to human activities, as well as predicting future climates (e.g. Budyko, 1969; Manabe and Bryan, 1969; McGuffie and Henderson-Sellers, 1997).
To first order, the Earth’s climate is controlled by the amount of incident solar radiation that is absorbed by the planet and by the thermal absorptivity of the gases in the atmosphere which controls the balancing emitted infra-red radiation (e.g. Paltridge and Platt, 1976; Goody and Yung, 1996). Solar radiation is absorbed principally at the surface of the Earth, and over the mean annual cycle, this absorption is balanced by radiation emitted from the Earth (Figure 1). This global radiative balance, which is controlled by the surface and atmospheric characteristics, by the Earth’s orbital geometry (e.g. Berger, 1981, 1988) and by the variability of solar radiation itself over time (Shindell et al., 1999), controls the habitability of the Earth, mean temperatures and the existence of water in its three phase states (e.g. Robinson and Henderson-Sellers, 1998). These characteristics, together with the effects of the rotation of the Earth on its axis, determine the dynamics of the atmosphere and ocean, and the development of snow and ice masses. This combination of a distributed radiative budget and the forces resulting from the planet’s axial rotation characterize any ‘snapshot’ of the Earth’s climate (Figure 1).

Figure 1. (a) Schematic of the latitudinal energy budget of the Earth (modified from A Climate Modelling Primer, by K McGuffie and A Henderson-Sellers, 1997, reproduced by permission of John Wiley & Sons, Ltd); (b) astronomical controls on the climate system include the planetary radiation balance (absorbed solar equals emitted infrared) and the effect on atmospheric and oceanic circulation of the planet’s spin around its axis (after Henderson-Sellers, 1995). The prime characteristics of the global climate are
listed

Copyright © 2001 Royal Meteorological Society

Int. J. Climatol. 21: 1067 – 1109 (2001)

1070

K. MCGUFFIE AND A. HENDERSON-SELLERS

The second and complementary timeframe that needs to be borne in mind in characterizing the climate system is the evolutionary time-scale which controls the very long-term aspects of the climate components and those factors which force it, such as the physics and chemistry of the planet itself and the luminosity of the Sun. Viewed in this timeframe, the Earth’s climate is prey to the forces of astronomical, geological and biological processes which control the persistence of ice caps and glaciers; the biota; rock structures and global geochemical cycling (e.g. Crowley, 1983; Schneider and Boston, 1991). For example, the

Figure 2. The Earth’s climatic history is shown (a) over the planet’s lifetime of 4.6 billion years (reprinted from after Future Climates of the World: A Modelling Perspecti6e, World Survey of Climatology Series, Vol.16, by A Henderson-Sellers (ed.), Chapter 1 – Climates of the Future, Figure 1f, copyright (1995), with permission from Elsevier Science); (b) in terms of the geo and astrophysical events during its 4.5 billion year life (from Schneider and Boston, 1991, reproduced by permission of The MIT Press; (c) geochronology of the last billion years (figure reproduced from A Berger, Milanko6itch theory and climate, copyright © 1988, by the American Geophysical Union, reproduced with permission), data from Baron, 1995; (d) estimated summer sea surface temperature (°C) over the last 130 000 years (reprinted from Future Climates of the World: A Modelling Perspecti6e, World Survey of Climatology Series, Vol. 16, by A Henderson-Sellers (ed.), Chaper 1 – Climates of the Future, Figure 1d, Copyright (1995), with permission from Elsevier Science); (e) mid-latitude air temperatures (°C) since 800 AD (figure reproduced from Climate Change, 26, 1994, pages 109 – 142, Was there a ‘warm period’, and, if so, where and when, by K Hughes and F Diaz, Figure 3, with kind permission of Kluwer Academic Publishers); and (f) global air temperatures (°C) since 1854 AD (reprinted from Future Climates of the World: A Modelling Perspecti6e, World Survey of Climatology Series, Vol. 16, by A Henderson-Sellers (ed.), Chapter 1 – Climates of the Future, Figure 1a, Copyright
(1995), with permission from Elsevier Science
The keyed events are: 1. Thermal maximum of the 1940s. 2. Little ice age. 3. Cold interval (Younger Dryas). 4. Present interglacial. 5. Previous interglacial. 6. Present glacial age. 7. Permo-Carboniferous glacial age. 8. Ordovician glacial age. 9. Late Precambrian
glacial ages. 10. Earth’s origin. GG, global greenhouse

Copyright © 2001 Royal Meteorological Society

Int. J. Climatol. 21: 1067 – 1109 (2001)

NUMERICAL CLIMATE MODELLING: A REVIEW

1071

Archaean, which represents almost half of the Earth’s history from the formation of the Earth around 4.6 billion years ago to the Proterozoic transition 2.5 billion years ago, enjoyed a variety of geospheres before plate tectonics became established (Figure 2(b)). During this period, the planetary surface was first bombarded by cometary and meteoritic material, as the debris from the planetesimals was ‘swept up’. This was followed by a period of surface vulcanism and then by a platelet tectonic regime. Life must have become established around or before 3.5 billion years ago because rocks from that era contain fossil evidence of viable and diverse microbial communities (e.g. Schopf et al., 1983). The oldest known rocks, the Isua supercrustal, are dated to 3.83 billion years, and show evidence of a global climate system not grossly different from that of the present. In particular, they contain sedimentary material that was waterborne. This suggests that the Earth’s climate has been very stable when viewed in this evolutionary timeframe (e.g. Lovelock, 1991).
Figure 2 illustrates two interesting observations about the Earth’s climate system. The first is that, over the lifetime of the planet and despite massive upheavals, the climate has remained remarkably stable (Figure 2(a)). The second is that excursions in temperature, and presumably any other climatic variables, have been large and aperiodic over the whole history of the Earth. The two characteristics of very long stability upon which short and medium term excursions are superposed are themselves a function of another fundamental quality of the climate system: the time required for equilibration, i.e. the time needed to adjust to a new forcing. The equilibration times for different subsystems of the Earth’s climate system differ very markedly (Table I). The longest equilibration times are those for the deep ocean, the glaciers and ice sheets (1010 – 1012 s), while the remaining elements of the climate system have equilibration times nearer 105 – 107 s.
The result of these two time-scales (evolution and equilibration) and of the complex interactions between these components of the climate system is a rich spectrum of climatic variability (Figure 2). The largest peaks in this spectrum relate to astronomical forcings: the Earth’s rotation, its revolution around the Sun, variations in this orbit and the formation of the solar system (Berger, 1981, 1995). Coupling and feedbacks amongst processes within the climate system components, the atmosphere with the oceans, surface water with ice masses and the biosphere are responsible for the myriad of variations in this climate system spectrum. To try to understand, analyse and predict such variations, climate scientists have developed numerical models.
The Intergovernmental Panel on Climate Change (IPCC) reports on climate change have underlined the extent of our dependence on numerical climate models. For example, Gates et al. (1996, p. 233) state:
The most powerful tools available with which to assess future climate are coupled climate models, which include three-dimensional representations of the atmosphere, ocean, cryosphere and land surface . . . [and] . . . More detailed and accurate simulations are expected as models are further developed and improved.

Table I. Representative equilibration times for components of the Earth’s climate system

Climatic domain

s

Equivalent

Atmosphere Free Boundary layer
Hydrosphere Ocean mixed layer Deep ocean Lakes and rivers
Cryosphere Snow and surface ice layer Sea ice Mountain glaciers Ice sheets
Biosphere Soil/vegetation
Lithosphere

106 105
106–107 1010–1011 106
105 106–1010 1010 1012
106–1010 1015

11 days 24 h
Months–years 300–3000 years 11 days
24 h Days–100s of years 300 years 3000 years
11 days–100s of years 30 million years

Copyright © 2001 Royal Meteorological Society

Int. J. Climatol. 21: 1067 – 1109 (2001)

1072

K. MCGUFFIE AND A. HENDERSON-SELLERS

This recognized dependence on the results of today’s numerical climate models must be matched by an appreciation of their history and weaknesses, as well as of their benefits and strengths. In this paper, we offer a review of some aspects of these characteristics of climate models.

2. CLIMATE MODELS

The date of the true origin of ‘climate modelling’ depends, of course, on both the definition of ‘climate’ (e.g. local, regional or global over weeks, months or millennia) and of modelling (e.g. physical construction, correlation with, say, latitude or numerically based). In this paper, we review only numerical and global climate modelling, i.e. representations of the global climate constructed by calculations arising from an equation-based characterization. The father of today’s climate models was Richardson. He published the first description of a method for constructing a weather forecast by means of numerical calculations (Richardson, 1922). It is well known that this method was at least thirty years ahead of even the very modest capability of the earliest computers. Richardson’s farsighted parenthood of climate modelling is further underlined in his recognition of climate components other than the atmosphere. In common with many climate modellers since his time, Richardson knew and acknowledged the importance of currently neglected aspects, in his case the ocean, via sea surface temperatures. He also shared the aspiration of many of today’s climate modellers when he wrote of the concept of developing a numerical model of the ocean similar to that which he had developed for the atmosphere:

It may come to that, but let us hope that something simpler will suffice (Charnock, 1993, p. 32).

Climate models are tools employed to enhance understanding of the climate system and to aid prediction of future climates. Although there have been great advances made in the discipline of climate modelling over its forty year history, even the most sophisticated models remain very much simpler than the full climate system. Indeed, such simplicity is an unavoidable and, for some, also an intended, attribute of climate models (e.g. Washington and Parkinson, 1986; McGuffie and Henderson-Sellers, 1997). Modelling of a system which encompasses such a wide variety of components as the climate system is a formidable task, and it requires co-operation between many disciplines if reliable conclusions are to be drawn. Even the most elementary characteristics of the atmosphere vary considerably between climate models as illustrated in Figure 3, although whenever such a figure is shown, there are immediate explanations of the differences among the illustrated results. Intercomparisons such as these are now an integral part of climate science, and an important means for advancement of understanding of the climate system. Indeed, it is a genuine measure of the maturity of the climate modelling community that such intercomparisons occur.
An essential ingredient for all climate modelling is the speed with which calculations can be made (cf. Richardson, 1922). The rapid increase in computing power over the last 40 + years has meant that climate models have expanded both in terms of complexity, as measured by the total time they can simulate, and in the spatial and temporal resolution they can achieve (e.g. Trenberth, 1992).
Multi-decadal to millennial simulations are now common; full diurnal and seasonal cycles are now standard in climate experiments; and transient changes in, for example, the atmospheric trace gases such as CO2 are becoming commonplace (e.g. Manabe and Stouffer, 1996; Boer et al., 2000). As knowledge increases, more aspects of the climate system are being and will be incorporated into climate models, and the resolution and length of integrations will further increase. At the same time, the goal of developing and evaluating the modelling tool most appropriate to each task will remain (cf. Shackley et al., 1998).
The simplest possible way of constructing a model of the Earth’s climate is to consider the radiative balance of the globe as a whole (cf. Figure 1). This is a zero dimensional model often written in the form of a pair of equations

S(1 − h) = |T e4

(1)

Copyright © 2001 Royal Meteorological Society

Int. J. Climatol. 21: 1067 – 1109 (2001)

NUMERICAL CLIMATE MODELLING: A REVIEW

1073

Figure 3. The zonally-averaged distribution of selected variables simulated by the AMIP models for December– January – February of 1979– 1988, and that given by observations (solid black line) from Gates et al. (1999): (a) is the sea-level pressure, with observed data from the European Centre for Medium-range Weather forecasts (ECMWF) reanalysis; (b) is the surface air temperature, with observed data from Jones (1988) and COADS (Comprehensive Ocean-Atmosphere Data Set; da Silva et al., 1994). Many of the differences among the model simulations are due to differences in spatial amd temporal resolution and in parameterization sophistication. See Gates et al. (1999) for model identification. (This figure is reproduced from PCMDI Report No. 45, Gates et al., 1998. The U.S. Government’s right to retain non-exclusive, royalty-free licence in and to any copyright covering this figure is acknowledged. Credit is given the University of California, Lawrence Livermore National Laboratory, and the Department of
Energy under whose auspices the work was performed.)

plus

Ts = Te + Tgreenhouse

(2)

Here, S (the amount of solar radiation instantaneously incident at the planet per unit area of its (spherical) surface) has a value of about 342 W m−2 and the Earth’s albedo, h, is 0.3. Thus, the effective blackbody radiating temperature of the Earth, Te, is found to be around 255 K. This is lower than the current global mean surface temperature, Ts, of 288 K, the difference, about 33 K, being a result of the greenhouse effect. In Figure 4, the components of the Earth’s globally and annually averaged radiation budget are presented as percentages of the average solar constant (342 W m−2) at the top of the atmosphere. Nearly half the incoming solar radiation penetrates the clouds and greenhouse gases to the Earth’s surface. These gases and clouds re-radiate most (i.e. 88 units) of the absorbed energy back down toward the surface. This is the basis of the mechanism of the greenhouse effect. The magnitude of the greenhouse effect is commonly measured as the difference between the blackbody emission at the surface temperature (a global average of 288 K gives 390 W m−2), and the outgoing infrared radiation at the top of the atmosphere (here 70 units or 239 W m−2), i.e. 151 W m−2.
Within the very long time-scale of the Earth’s history, it is possible to take a ‘snapshot’ view of the climate system (e.g. Figure 1(b)). In this ‘instantaneous’ view, the shortest time-scale processes are most evident. Of these, the most important are the latitudinal distribution of absorbed solar radiation (large at low latitudes and much less near the poles) as compared with the emitted thermal infrared radiation which varies much less with latitude (Figure 1(a)). This latitudinal imbalance of net radiation for the surface-plus-atmosphere

Copyright © 2001 Royal Meteorological Society

Int. J. Climatol. 21: 1067 – 1109 (2001)

1074

K. MCGUFFIE AND A. HENDERSON-SELLERS

Figure 4. Schematic of the Earth’s energy budget (modified from SH Schneider, 1992, Coupled Climate System Modelling, by KE
Trenberth (ed.), reproduced by permission of Cambridge University Press). Units are percentage of the incident solar radiation, 342 W m−2. Major characteristics of the climate system are listed

system as a whole (positive in low latitudes and negative in higher latitudes) is partitioned into energy fluxes at every location (e.g. Figure 4), and combines with the effect of the Earth’s rotation on its axis to produce the dynamical circulation system of the atmosphere (Figure 1(b)) and the oceans.
The latitudinal radiative imbalance tends to warm air which rises in equatorial regions, and would sink in polar regions were it not for the rotation of the Earth. The westerly waves in the upper troposphere in mid-latitudes and the associated high- and low-pressure systems are the product of planetary rotation affecting the thermally-driven atmospheric circulation (Figure 1(b)). The overall circulation pattern comprises thermally direct cells in low latitudes, strong waves in the mid-latitudes and weak direct cells in polar regions (Peixoto and Oort, 1991). This circulation, combined with the vertical distribution of temperature, represents the major aspects of the atmospheric climate system (e.g. Schneider, 1992).
There is, today, a wide range of climate models available for the variety of simulation tasks associated with improving understanding of the climate system and predicting future (and past) climate changes. Currently the most highly developed tools available for climate assessment are the global climate models (GCMs) and the earth models of intermediate complexity (EMICs). These models, based on knowledge of physics, chemistry, biology, as well as economics and social science, portray this understanding in simplified representations, called parameterizations, of the processes they are designed to characterize. In a climate model, an atmospheric component is coupled to a model of the ocean, a representation of the biota and sometimes characterization of technological trends and food and water resources (e.g. Schlesinger, 1988; Wiebe and Weaver, 1999). The term GCM is nowadays taken to mean at least fully three-dimensional models of the atmosphere and oceans coupled together. If only the atmospheric (or oceanic) component is represented, the acronym AGCM (atmospheric GCM) or OGCM (oceanic GCM) is used. The difference in response (or equilibration) times of, for example, the ice masses and the carbon cycle compared to the atmosphere (Table I) means that different components are explicitly incorporated into different climate model types. For long time-scale simulations of future and past climates, the EMICs are used, while for periods of days and decades to a century or two, GCMs are employed. Although a few GCM integrations have extended over 10000 years or more (Broccoli, 2000), the main focus of GCM studies continues to be on the decadal to century scale.
This review is organized in a roughly historical narrative, which is summarized in Table II. This 40-year story of numerical climate modelling does not, however, fit tidily into either an evolutionary structure or

Copyright © 2001 Royal Meteorological Society

Int. J. Climatol. 21: 1067 – 1109 (2001)

NUMERICAL CLIMATE MODELLING: A REVIEW

Table II. Historical evolution of climate models

Decade and landmark papers

Climate model status

51969 Manabe and Mo¨ ller (1961) Manabe and Strickler (1964) Sellers (1969) Budyko (1969)
1969–1981 Manabe and Bryan (1969) Green (1970), Stone (1973) Manabe and Wetherald (1975) CLIMAP (1981)
1981–1989 Hansen et al. (1981) Sellers et al. (1986) Oort and Peixoto (1983) Luther et al. (1988)
1989–1999 Houghton et al. (1990) Semtner and Chervin (1992) Flato and Hibler (1992) Cubasch et al. (1994) Santer et al. (1996)
2000s ???

Numerical weather forecasts extended RC models developed Dynamics and radiation virtually separate EBMs newly described
Multi-layer oceans added to GCMs SD models developed Greenhouse modelling with GCMs Palaeo datasets first employed for ‘validation’
GCMs becoming predominant model type Surge in computational power and capacity Satellites generate global observations Model intercomparisons suggested
Simpler models required by IPCC OAGCMs established but need flux correction Sea-ice and land-surface components evolving First ocean–atmosphere coupled ensemble Validation and attribution first described
EMICs as important as GCMs Past climate simulations re-emerging for testing Observational need driven by evaluation demand Policy needs a major driver of numerical models

1075

allow a neat sectionalization into systems’ and components’ descriptions. This is the result of two, sometimes competing, factors: increased computer power and sparcity of observations. First, the development of numerical climate modelling has always been dependent on the state of development of the numerical platforms, i.e. the computer. This interdependence, for climate modellers needs have also prompted computational developments, is a tangled affair, which has, at some points, seen computation undertaken without clear motives other than to use the power (e.g. Semtner and Chervin, 1988, 1992). At the same time, scientists challenging the ‘received wisdom’ have always disputed the predictions of numerical models and even their underpinning premises. These debates and disputes have, quite naturally, often been tied up with the issues of funding, influence and publicity (see Shackley et al., 1998; Henderson-Sellers and McGuffie, 1999).

3. DEVELOPMENT OF CLIMATE MODELLING

Climate modelling has developed considerably since the first global atmospheric models were applied to climate simulation in the 1960s (cf. Table II). Models have been developed in response to scientific probing of existing model components, and have drawn on existing models. Throughout the 40 or so years of climate modelling, the growth in the complexity and physical realism of models has been facilitated by developments in computer technology (Henderson-Sellers and McGuffie, 1987). Here, we review the historical framework for some of the developments which have taken place over the last 40 years, and look at how the evolution of climate modelling has paced the emergence of each generation of high performance computers.
As climate models are sometimes described in terms of an hierarchy (e.g. McGuffie and Henderson-Sellers, 1997; Henderson-Sellers and McGuffie, 1999), it is often assumed that the simpler models were the first

Copyright © 2001 Royal Meteorological Society

Int. J. Climatol. 21: 1067 – 1109 (2001)

1076

K. MCGUFFIE AND A. HENDERSON-SELLERS

to be developed with the more complex GCMs being developed most recently. This is not the case (Table II and Washington and Parkinson, 1986). The first atmospheric general circulation climate models were being developed in the early 1960s (e.g. Smagorinsky et al., 1965) concurrently with the first radiative convective (RC) models (Manabe and Mo¨ ller, 1961; Manabe and Strickler, 1964). On the other hand, the simplest (energy balance) climate models, as they are currently recognized, were not described in the literature until 1969; the first discussion of two-dimensional statistical dynamical (SD) models was in 1970 (Saltzman, 1978) and the ESMIC are the youngest model type (e.g. Opsteegh et al., 1998; Rahmstorf and Ganopolski, 1999).

3.1. Complex climate models
The first atmospheric general circulation climate models were derived directly from numerical models of the atmosphere designed for short-term weather forecasting. These had been developed during the 1950s (e.g. Charney et al., 1950; Smagorinsky, 1983) and, around 1960, as advances in computer technology allowed more extensive simulations, ideas were being formulated for long enough integrations of these numerical weather prediction schemes that they might be considered as climate models. Indeed, it is rather difficult to identify the timing of the transition from weather forecasting to climate prediction in these early modelling groups. The numerical requirements of weather prediction were extended to hemispheric domains (global calculations were not introduced until later) and the extension to longer integration periods sometimes became simply a matter of availability of computer resources. Indeed, to this day, climate modelling and numerical weather forecasting groups co-exist, especially in national meteorological bureaux. However, the needs and focus of the two disciplines differ: for example GCMs have to conserve mass, energy and moisture, while many forecast simulations are over too short a period for conservation to be an issue.
Many of the early pioneers of climate modelling came from numerical weather prediction. For example, Manabe joined the National Oceanic and Atmospheric Administration’s (NOAA’s) Geophysical Fluid Dynamics Laboratory in the USA in 1959 to collaborate in the development of numerical weather prediction models. He was to go on to become one of the pre-eminent leaders of the climate modelling community (e.g. Manabe and Bryan, 1969; Manabe and Wetherald, 1980; Manabe, 1985; Manabe and Bryan, 1985; Manabe and Stouffer, 1999). Scientists concerned with extending numerical prediction schemes to encompass hemispheric or, later, global domains were also studying the radiative and thermal equilibrium of the Earth – atmosphere system (Table II). It was these studies which prompted the design of the RC models, which were once again spearheaded by Manabe (Manabe and Mo¨ ller, 1961). Other workers also expanded the domain of numerical weather prediction schemes in order to derive GCMs (Adem, 1965). The low-resolution thermodynamic model first described by Adem in 1965 is an interesting climate model type. Although the methodology is simpler in nature than that of an atmospheric GCM, it captures many aspects of a full GCM. Similar in basic composition to the energy balance models (EBM) developed later, Adem’s model includes, in a highly parameterized way, many dynamic, radiative and surface features and feedback effects. It could be argued that Adem’s model is the ancestor of today’s EMICs.

3.2. Simpler climate models
Not all climate models originated from weather forecast models. In 1969, two very similar models were published within months of each other. Budyko and Sellers published descriptions of models which did not depend upon the concepts already established in numerical weather prediction schemes, but attempted to simulate the essentials of the climate system in a simple way (Budyko, 1969; Sellers, 1969). These EBMs drew upon observational data derived from descriptive climatology; for example, the reasons why the major climatic zones are roughly latitudinal. In particular, EBMs are computationally very much faster than GCMs because instead of calculating the dynamical movement of the atmosphere, using the Navier –Stokes equations (as in GCMs), they employ much simpler parameterizations, and typically, much coarser grids. As a consequence of the intrinsically simpler parameterization schemes employed in EBMs, they could be

Copyright © 2001 Royal Meteorological Society

Int. J. Climatol. 21: 1067 – 1109 (2001)
AtmosphereModelsClimate ModellingClimate ModelsRadiation