Global Warming: The Balance of Evidence and Its Policy Implications

A Review of the Current State-of-the-Controversy

Charles F. Keller

Former Director of the Los Alamos Branch of the University of California's Institute of Geophysics and Planetary Physics and currently Visiting Scientist at Los Alamos National Laboratory

Received October 19, 2001; Revised March 3, 2002; Accepted March 5, 2002; Published May 5,2003

Global warming and attendant climate change have been controversial for at least a decade. This is largely because of its societal implications. With the recent publication of the Third Assessment Report of the United Nations' Intergovernmental Panel on Climate Change there has been renewed interest and controversy about how certain the scientific community is of its conclusions: that humans are influencing the climate and that global temperatures will continue to rise rapidly in this century. This review attempts to update what is known and in particular what advances have been made in the past 5 years or so. It does not attempt to be comprehensive. Rather it focuses on the most controversial issues, which are actually few in number. They are:

· Is the surface temperature record accurate or is it biased by heat from cities, etc.?
· Is that record significantly different from past warmings such as the Medieval Warming Period?
· Is not the sun's increasing activity the cause of most of the warming?
· Can we model climate and predict its future, or is it just too complex and chaotic?
·Are there any other changes in climate other than warming, and can they be attributed to the warming?

Despite continued uncertainties, the review finds affirmative answers to these questions. Of particular interest are advances that seem to explain why satellites do not see as much warming as surface instruments, how we are getting a good idea of recent paleoclimates, and why the 20^th century temperature record was so complex. It makes the point that in each area new information could come to light that would change our thinking on the quantitative magnitude and timing of anthropogenic warming, but it is unlikely to alter the basic conclusions.

Finally, there is a very brief discussion of the societal policy response to the scientific message, and the author comments on his 2-year email discussions with many of the world's most outspoken critics of the anthropogenic warming hypothesis.

KEYWORDS: climate, climate change, global warming, climate modeling, atmosphere, ocean, greenhouse gases, carbon dioxide, solar activity, environment, ecosystems

DOMAINS: plant sciences, soil systems, global systems, atmospheric systems, marine systems, ecosystems and communities, environmental chemistry, water science and technology, environmental technology, environmental management and policy, modeling, environmental modeling, environmental monitoring, isotopes in the environment

"In the light of new evidence, and taking into account the remaining uncertainties, most of the observed warming over the last 50 years is likely to have been due to the increase in greenhouse gas concentrations." Policy Makers' Summary, IPCC Third Assessment Report, 2001

In 1957 noted climatologist Roger Revelle wrote: "Thus human beings are now carrying out a large-scale geophysical experiment of a kind that could not have happened in the past nor be reproduced in the future. Within a few centuries we are returning to the atmosphere and oceans the concentrated organic carbon stored in sedimentary rocks over hundreds of millions of years..."[1].

Revelle also encouraged David Keeling to make measurements of carbon dioxide (CO₂) levels in the atmosphere. The resulting "Keeling Curve" shows dramatically the rise of CO₂ above pre-industrial levels. Fig. 1 shows this rise from a combination of ice-core (pre-industrial), and atmospheric measurements as well as estimates of fossil fuel burning in recent times, which parallel the rise in CO₂.

Perhaps, more than anything else, this documented rise in atmospheric concentration of a powerful greenhouse gas (GHG) has served to bring the possibility of human warming of the climate to our attention. In fact, this rise makes some warming a certainty since the greenhouse effect - i.e., surface warming due to the presence of atmospheric GHGs - is real and its physics is well understood. The only questions are how much warming and how soon. The well-documented rise in global surface temperature over the past two decades has served to emphasize these questions and to focus society's interest on them.

To get the earliest possible answer to these questions the United Nations with the World Meteorological Organization formed the Intergovernmental Panel on Climate Change (IPCC) and tasked its Working Group I with:
.
1. Getting the best answer to the two questions above, and
2. Communicating it to policy people in terms that they can understand and use.

Most of the ensuing controversy over the answers the IPCC has been publishing probably results from the urgency associated with CO₂ increases in the atmosphere. Because CO₂ has a residence time in the atmosphere of over a century, it is important to get the earliest estimate of how much and how soon any warming might be if we are to do anything to slow projected warming before it occurs. Thus, climate scientists find themselves trying to see an initially small signal in a noisy, chaotic climate. The IPCC's 1996-very-guarded statement that we are beginning to identify this signal was an attempt to give the earliest possible warning commensurate with our understanding. But, because it is such a low confidence warning, there is plenty of room to take exception for those who think we still have not proven a human influence. This was exactly the motivation for establishment of the IPCC with its ability to provide an authoritative scientific review and to reflect a scientific consensus without undue influence from extreme positions. It is this review's reading that the IPCC has given us a fairly comprehensive assessment of the current understanding of these two questions. Summaries of the IPCC's Third Assessment Report can be found at: http://www.ipcc.ch/.

This review of the current status of climate science applied to the issue of possible effects of increasing anthropogenic greenhouse gases (AGHG) on the climate will not attempt to be comprehensive. The subject is far too broad - witness the IPCC's Third Assessment Report, merely the first volume of which is nearly 1000 pages in length. Instead, this review will concentrate on a focused line of reasoning leading to the increasingly certain conclusion that humans are causing the Earth to warm. It will admittedly be guided in part by the objections of critics of the human warming hypothesis. These objections are along the following lines:

1. The observational record of warming at the Earth's surface is flawed and not reliable; there has been little warming since 1945 as attested to by satellite observations.
2. Climate has always varied naturally, sometimes much more than during the 20^th century, and so the warming we are experiencing is probably nothing new.
3. What climate change there is can be attributed to variations in solar activity and other natural factors.
4. The AGHG villain, increases in CO₂, cannot reliably be attributed to human emissions given the huge reservoirs of it in the biosphere and oceans.
5. The climate is far too complex and chaotic to be modeled even by present supercomputers, thus these huge global climate models (GCMs) cannot be relied on either to attribute observed warming to AGHGs or to predict the timing or extent of future warming.

These objections will be discussed in succeeding sections. Other aspects of climate change will be summarized with references to review-type articles. The final section will touch very briefly on policy implications, given the state of our understanding.

Of all the objections to the assertion that humans are warming the climate mainly through emissions of CO₂ and other anthropogenic greenhouse gases (AGHGs), perhaps the most used and discussed is the fact that temperature recording instruments aboard a continuing succession of NASA/NOAA satellites measure only a fraction of the warming observed at the surface since 1979. Some say that the disagreement between warming trends deduced from these two sources shows the surface record to be in error, i.e., the warming in the second half of the 20^th century is not as large as that measured at the surface. A U.S. National Research Council panel reviewed this disagreement and concluded that very likely both records are correct[2]. We agree and will assume that both records are accurate in the following discussions. Thus the discussion will center around explaining how the surface could be warming faster than the middle troposphere measured by the satellites. Fig. 2 shows surface temperature anomalies for the past hundred years or so. The warming, however, is not smooth like the rise in AGHGs (Fig. 1). Instead there is a marked initial warming between 1920 and 1945 interrupted by a 30-year plateau whereupon warming resumes. The reasons for this will be discussed later in the modeling sections; however here we will concentrate on temperature behavior in the last 40 to 50 years. Fig. 3 shows that much of the structure in the surface temperature record is caused by El Niño/La Niña variations and by volcanoes whose eruptions are large enough to loft light-scattering aerosols into the stratosphere. It plots both global and tropics-only temperatures showing the strong influence El Niños and La Niñas have on climate.

The surface record is made from thousands of temperature-sensing installations both on land and at sea. Enormous work has gone into eliminating sources of error from this record[3]. To avoid zero point differences and similar problems unique to the site, each observing station determines its own average temperature for a prescribed number of years. Then each record is differenced from that average, generating a set of anomalies that are then combined to get average records (global, hemispheric, etc.).

Perhaps the most worrisome source of error is contamination from urban areas, which usually warms more than the surrounding countryside and thus might make the warming look larger than it is. This urban heat island effect (UHI) would be expected to become increasingly important as urbanization grows. It would mimic a secular increasing temperature, particularly in the past 25 years. To avoid this problem special precautions are taken for urban data. Also several studies have been done[4,5] that compare the global temperature record between rural and urban sites. The results always come out that, for the stations used, there is not much difference between rural, suburban and urban sites - perhaps a 10% effect. Here it is interesting that the area where one might expect this effect most, Northeast United States, is actually not warming all that much. Also, surface and satellite anomalies agree fairly well over northern hemisphere (NH) land while disagreeing most over the tropical oceans (hardly an UHI effect). More recently Folland et al.[6,7] have quantified the major sources of uncertainty in the surface record. They find that between 1861 and 2000 the Earth's surface has warmed 0.61 ± 0.16°C with most of the uncertainty coming from the 1800s.

The record over the global oceans has different problems (for example, sea surface temperatures, SSTs, are used in place of marine air temperatures [MAT] - Fig. 4), and, while studies show recent disagreement between the two[8], they are not estimated to affect the general accuracy of the record[6] (see also Folland's summary of this in EOS, p. 453, Oct. 2, 2001). Finally measurements of temperatures in the deep oceans are corroborating the surface trends[9].

Satellite temperature observations are made with instruments that detect microwave radiation from oxygen in the atmospheric column (thus the name microwave sounding units - [MSUs]). These are tuned to regions of the electromagnetic spectrum that change measurably with temperature and that sample specific vertical regions of the atmosphere. Two in particular are of interest: MSU4, which measures radiation from the stratosphere, and MSU2LT from middle altitudes in the troposphere[10]. These records are fairly well substantiated by balloon observations[2,11]. From this source we now have a 21-year worldwide record of temperatures in both the stratosphere and the middle troposphere.

One way to gain insight into how accurately the surface records represent global temperatures is to compare them with the satellite (MRU2LT) record. Even though the 2LT record does not measure surface temperatures, both regions of the atmosphere are expected to warm similarly. When this comparison is done, the agreement/disagreement is in the eye of the beholder. To first order the two records agree rather well (Fig. 5a). However, if a simple temperature trend is determined for each data set, they differ significantly - 0.180°C/decade for surface and 0.056°C for satellite (1979-1999). It is this disagreement that has caused skeptics to contend that, what global warming there is must be much smaller than that predicted by greenhouse gas theory since the satellite record covers the entire globe densely and is clearly without sources of surface contamination such as UHI and the SST/MAT approximation. Only part of the trend discrepancy is explained thus far[2,12,13].

Actually the two records agree fairly well over NH land areas, where both show similar large warming. It is over the tropical oceans that the two records disagree most strikingly. Finally, it has been pointed out that, even here, the disagreement is a rather recent one, largely beginning after the 1991 eruption of Mt. Pinatubo. Also a 40-year comparison of sonde (balloon) mid-troposphere temperatures with those at the surface shows both records yielding similar warming trends (Fig. 5b). Another point to keep in mind is that trends appear to vary monotonically from warming to cooling as one proceeds from the surface into the stratosphere. Thus, surface warming is about 0.18°C/decade, low/mid troposphere (MSU) is 0.06°C/decade, essentially zero in the troposphere above that, and strongly negative, -0.5°C/decade in the lower stratosphere. This effect points to a possible explanation for the discrepancy.

The satellite record begins at the end of a dramatic rise in global temperatures measured both at the surface and from balloons (Fig. 5b). At the outset (1979-1981) the satellite record (in agreement with the sondes) showed relatively more warming than the surface record, as might be expected from Clausius-Clapeyearon effect.¹From 1980 to 1991 satellite and surface records were were in essential agreement through several swings of ENSO and the cooling caused by a major eruption of the volcano, El Chichon. Immediately following the massive eruption of Mt. Pinatubo (June 1991) the Earth cooled dramatically, then warmed over the next few years as sun-blocking aerosols, injected into the stratosphere by the volcano, gradually settled back into the troposphere where they were rained out. The process took about 5 years. As expected, the satellite showed more cooling than at the surface, but during the subsequent recovery period (1992-1996) it did not show its customary warming response to three successive El Niños (should have been more than at the surface).¹Thus a difference plot of satellite minus surface temperature anomalies shows:

1979-1981 Satellite - warmer
1980-1991 Agreement
1992-1996 Satellite - cooler

This disagreement - high early and low late - is the major reason that the satellite record shows a lower warming trend than the surface[3].

Beginning with the La Niña of 1997, the satellite once again resumed its customary response to ENSO cycles:

1997, La Niña cooler than surface
1998, El Niño warmer than surface
1999-2000, La Niña cooler than surface

Thus the period 1992-1996 seems anomalous when compared with behavior since 1958. It is thought that Pinatubo's sulfate aerosols may have caused this by adding to the secular reduction in stratospheric ozone (following a brief warming due to absorption of sunlight by these same aerosols, Fig. 6). This warming-followed-by-cooling behavior happened after both eruptions, but the cooling was greater after Pinatubo than El Chichon (0.3°C and 0.5°C, respectively). The cooling was very likely due to dramatic reductions in ozone caused by heterogeneous chemistry on the volcanic aerosols, adding to continued secular ozone decreases. This cooling persisted long after the aerosols settled out because ozone recovers slowly if at all. The dramatically cooled lower stratosphere in turn had a cooling effect on the upper and middle troposphere where the satellite was measuring temperature. This would explain both the lower warming trend and also the failure of the mid-troposphere to respond to El Niño warming. In fact the lower stratosphere continues very cold up to the present, which may be causing a cool bias in the troposphere.

There is further confirmation of this rationale. The lapse rate (the rate at which the troposphere cools with altitude) has been observed to be changing in response to some forcing mechanism. Two recent papers[14,15] have looked at lapse rates in the tropics. (In addition, I recommend the "Perspectives" article by David Parker in the same Science issue as the Gaffen paper. It is a good statement of where we are with this problem.) In the first paper the authors use radiosondes - balloon-borne instrument packages - to measure both the surface and tropospheric warming. In the time period 1960-1997, they find that, in relative agreement with computer codes, the middle and upper troposphere warm at a rate of 0.19°C/decade, more than the measured warming at the surface, 0.12°C/decade, but during the time period of satellite observations (1979-1997) the trends are reversed with mid-upper troposphere cooling at the rate of -0.10°C/decade while the surface trend remains the same. The authors conclude "temperature and lapse-rate trends during the MSU (satellite) period are qualitatively different from the preceding two decades." The second paper concurs and concludes: "This lapse rate variability could account for a significant portion of the difference between the observed surface and the lower tropospheric temperature trends, as measured by MSU and radio sondes, since 1979." Thus, both papers agree that in the topics the differences between warming at the surface and at MSU effective heights are real. They further agree that this is largely caused by marked changes in lapse rates. And finally, when a longer period than that of the MSU record is looked at, the warming trends, surface and troposphere, are about the same.²

Finally, and this might be significant, both papers agree that computer simulations are unable to reproduce this lapse rate variability. This is probably due to some shortcoming in the models. Apparently some other mechanism or forcing is operating that is not included in the models. Several come to mind - some effect of solar activity that acts only high in the troposphere[16], or, in the case of the discrepancy after 1992 - a cool upper boundary condition due to marked tropospheric cooling following Mt. Pinatubo-induced ozone destruction[17,18], or some lower boundary condition such as too simple an ocean model. These possibilities will be discussed in the modeling section below.

Another approach to compare the surface and satellite records is to remove the large interannual variability due to ENSO and volcanic eruptions. Several attempts have been made[19,20,21]. Two find little warming after removal, but the third finds very large warming. Perhaps the most comprehensive effort is by a large team[22]. This last effort shows a range of warming from both surface and satellite (0.210 to 0.250°C/decade and 0.056 to 0.158° C/decade, respectively). Also the authors suggest that this means the effects of large ENSO and volcanoes cool the troposphere more than the surface as suggested above: "Had Pinatubo and El Chichon not occurred, it is likely that the lower troposphere would have experienced more pronounced warming." They perform the same separations on several climate modeling results and show that at least part of the discrepancy in warming trends is being addressed in the models (more recent models in the sections, "Computer Simulation" and "Attribution and What Might We Expect in the Future" further close this gap).

The evidence supports the NRC report's conclusion that both the surface and satellite (tropospheric) records are substantially correct. At issue is the accuracy of the surface record, i.e., where humans live. The reason for the difference in temperature trends between the two remains only partially explained; however, possible errors in the surface record from urban heat island and sea surface temperature approximation of marine air temperatures do not appear large enough to explain the discrepancy in recent warming trends. In attempting to understand the difference between the two, we are realizing the importance of correctly simulating the lower stratosphere and perhaps the oceans for correct modeling of tropospheric temperatures.

If the 20^th century warming is real, it may still be argued that it is not "special", i.e., there have been other warming periods as great or perhaps greater. To answer this question we consider what is known about the Earth's temperature before the instrumental record began in the middle of the 19^th century. To be sure, climate has varied considerably over the age of the Earth; however, for most of that time conditions were so different as to make those climates not good analogues for today's situation. Over the past million years or so the climate has alternated between glacial and interglacial periods. These alternations are thought to result from three cyclic changes in the Earth's orbit and orientation toward the sun. The record from the Antarctic Vostok ice core shows at least five ~135,000-year-long glacial/interglacial cycles (Fig. 7). The question of what caused these oscillations between cold and warm periods was answered in large measure by V. Milankovitch, who noticed that changes in the Earth's orbit and aspect toward the sun could explain the broad features. There are three such changes:

the shape of the Earth's orbit around the sun: eccentricity
the tilt of the Earth's rotation axis: obliquity
the "wobble" of the Earth's axis: precession

These changes cycle completely in about 100,000, 44,000 and 23,000 years, respectively, and the interplay of the three causes a complicated pattern of the amount of sunlight reaching the NH. If the land and oceans were divided symmetrically between NH and SH, these variations would also occur symmetrically and would have little effect. But most of the land mass is in the NH, and, since it responds with a larger temperature amplitude to changes in sunlight, these cycles can have marked climate change implications. When one combines variations in sunlight due to these three effects and compares their combined effect with the Vostok core's record of amount of ice covering the land, the two look very much alike, suggesting that Milankovitch cycles initiate major climate changes over at least the past half million years. However, the cycles must be amplified by some additional feedbacks since the amount of sunlight variation they cause is not nearly large enough to force the observed changes in global temperature. Recent modeling shows that a combination of variations in CO₂ and ice cover can provide the necessary amplification. While researchers still differ on exact timing of CO₂ forcing relative to paleo-temperature response to Milankovitch cycles, recent work by Shackleton[25] and others suggests that CO₂ changes preceded temperature change. However, even if the changes were much closer in time, the picture still makes sense with CO₂ and ice acting as positive feedbacks.

Global temperatures reached a maximum sometime in the early Holocene (about 10,000 years ago) and have been declining slowly ever since wi₂changes in the precessional cycle. Fig. 8[26] shows this effect via three independent proxies of temperature:

ice melt in core from Ellsmere Is.
elevation of treeline (Sweden)
oxygen isotope temperatures derived from stalagmites (Norway)

These agree on a secular decline in atmospheric temperatures, and it is against this backdrop that we look for additional, shorter-scale climate variation. What we find in the last 1,500 years is an alternation between warming and cooling characterized by the two most significant events, the Medieval Warming Period (MWP) and the Little Ice Age LIA).

The MWP occurred between 1000 and 1300 when Europe and the North Atlantic were relatively warm. The LIA followed between 1500 and 1900, when they were abnormally cold. Indeed, if we can believe historic records from Europe, which told of warm weather crops' being grown far north of where they grow now, the MWP for Europe must have been as warm or warmer than at present. But was this warming and subsequent cooling global or even hemispheric at such large amplitude? The best answer with the sparse data we have is maybe, but probably not.

While the LIA seems to have been global in extent, the MWP appears not to have been. The MWP from different records seems spread out over two or three centuries, occurring at different times in different places, but not generally if at all in the SH[27]. One multiproxy reconstruction[28] shows that the peak warming seems to have occurred in the 12^th century. Other records suggest it occurred earlier. The LIA also seems to have been a multi-episodic event generally of the late 17^th and mid 19^th centuries. It is the regional timing of these events that makes them hard to pin down. For the same reason, when the best proxy records from tree rings, ice cores, historical records, sediments, etc. are averaged, many features cancel each other out and the resulting curve shows a much calmer, low amplitude record on average. On the other hand these records all agree on one thing, the rapid rise in temperature in the 20^th century. Also, there seems to have been less spatial and temporal variability in climate change in the last century than in earlier times. And it is the homogeneity of the 20^th century record (as well as its rapid rise) that separates it from these other events, not just its amplitude. But how are these variations determined and how accurately?

Temperature Proxies

Temperature proxies - that is, measurements that provide indirect indications of temperature-exist for many localities and regions back to the 16^th century, but high quality/resolution records back 1,000 years are few. Nevertheless several groups have attempted to determine hemispheric temperatures back that far from an ensemble of regional records, selected to be representative.

Two excellent summaries of this work to date are found in Jones et al.[29,30]. They discuss the problems involved and consider the multiproxy compilations published since 1998[28,31,32,33,34]. Finally I recommend an excellent short summary by Bradley[35].
The temperature proxies most in use are

tree rings
ice cores
ocean and lake sediments
corals
stalagmites
historical records
altitude of treeline

A good introduction to how these are used is given by Erik Stokstad[36]. A more detailed treatment is Bradley's book[37]. Circumstantial evidence comes from archeology and other sources, but the above can be quantified more reliably. Even so, each has its problems largely because temperature is not the only cause of the variations in these records. Old age in trees, growth factors in corals, deposition in sediments, rates of snowfall in ice cores, etc. all contribute to uncertainty in temperatures. Considerable effort thus goes into accounting for these problems[38,39].

There is a plethora of individual records[40,41,42,43,44,45a,45b]. Of particular interest is the temperature amplitude between the peak of the MWP and the LIA minima. Depending where and how you look, this amplitude varies from very small to rather large, mostly in the range 0.3° to 2.0°C. But to compare past climates with the last century, one needs some global or at least hemispheric average of these records. Several such compilations have been done involving a mix of the above proxies. Fig. 9 shows these. Note that the total amplitude indicated is <0.6°C. However, in contrast (Fig. 10), some individual records show much larger changes such as from an ocean sediment record[41], borehole records in the Greenland ice cap[44] and West African sea temperatures[42].

The situation is further complicated by introduction of a nonproxy actual temperature record - thermometry from inside boreholes. This method determines paleo-temperatures by inverting observed temperatures measured within boreholes, either in ice or ground. This is quite an independent method and has much promise. In general it gives similar results to those of other proxies, but in detail it differs in important ways. When records from hundreds of boreholes around the world are compared with results from other proxy data, two things are noticed. First, the borehole results are very smooth and show only long-term, low frequency trends. Second, the borehole results usually show a larger, long-term total temperature amplitude than proxy compilations[46]. This difference in temperature amplitude between borehole and proxy data has been cause for some concern as to which is correct. Given the difficulty of some of the proxy data in showing low frequency variability one is tempted to rely on the borehole results. But they, too, can be biased because of surface changes at the borehole such as land use (clearing, etc.), which would make earlier temperatures look cooler than they actually were.

Indeed Jones and Briffa[29] might be teasing out just this effect. They show an extended instrumental record from England going back to the late 1600s and another for Central Europe from 1750. It agrees with the borehole record for that region. Curiously, while the global borehole record indicates a total temperature variation amplitude of about 1.1°C between 1600 and 1945, the England and Central Europe borehole records suggest an amplitude of ~0.4°C, in agreement with both the instrumental record and that determined from proxy reconstructions calibrated to instrumental records (Fig. 9). Since in England and Europe land use has not changed much in centuries, might the records from those areas not be more accurate than the global one, which has experienced large, recent surface changes? Countering this attractive rationale are two borehole thermometry records from the Greenland icecap[44], which show a long and large amplitude MWP and a relatively deep LIA (which correlates rather well in timing and amplitude with a Sargasso Sea sediment d¹⁸O record[41] and Fig. 10). To further the confusion, temperatures inferred from oxygen isotopes taken from the ice core at the GISP Greenland site show no substantial MWP and only a modest LIA. Thus, on the same ice sheet, proxy and thermometry methods disagree. Given these problems, some are tempted to rely only on boreholes and altitude of treeline for multicentury temperature reconstructions[47]. It will require more study to unravel these conundra. For the present, perhaps the best approach is to continue to compare borehole
thermometry with other types of records and to assume that the real answer is probably somewhere between the two.

One point, however, is becoming increasingly clear. Past climate seems to have been spatially more heterogeneous than recently. Intercomparison of proxy records shows that they disagree as often as agree. The picture they present is a hemispheric/global climate that is a composite of very differing regional temperature histories. This, of course, could simply be defining the level of accuracy (or inaccuracy) of the methods; but if it is real, it suggests that earlier climate, forced only by natural events, was more regionally variable than in the last century or so. This could mean two things:

1. Determining a truly hemispheric or global paleo-temperature record will require a very carefully selected set of records.
2. Climate variations before the 20^th century were not as regionally uniform as recently.

This in itself could be a defining characteristic of GHG warming. As one noted paleoclimate researcher remarked:

"The 20^th century warming, and 19^th century cooling are the most extreme and coherent trends common to the northern sites we have studied. These recent, more coherent trends suggest the possibility of stronger common forcings, becoming more dominant over regional variations, in particular the increasing trace gases in the 20^th century. Less coherent fluctuations among the records back in time may signify more regional effects, and also the need for additional coverage for a more accurate perspective of climatic change"[43].

Paleoclimate prior to the last thousand years or so has been influenced by factors not relevant since. Attempts to determine recent paleo-temperatures are complicated by large uncertainties. With these cautions in mind we can still say that temperatures in the past 1,000 years or so are becoming determined well enough to be compared with the recent warming. The picture is that of a slow, secular cooling throughout the Holocene related to precession. Superimposed on this are solar activity variations that produce significant variation in the signal. Most recently is a slow warming, MWP (at least in the NH), beginning about 1,500 years ago and continuing sporadically, varying from place to place and time to time till about 1200. Conversely, the 16th century saw cooling that deepened and lasted through the 17th century. The relatively warm 18th century saw a reversal in temperatures that lasted into the 19^th century when the earlier levels of cooling returned and lasted until the early 20^th century. Thus the so-called LIA was really two or more episodes with warmings in between. Hemispheric and global warming in the 20^th century appears to have been larger, more rapid and more uniform than any time in nearly 2,000 years (although certain years or even several year periods were quite warm and in some regions that warming probably equaled that of the 20^th century.

In terms of the debate, determination of temperatures in the past 1,000 years reduces to essentially one disagreement. Paleoclimatologists for the most part reconstruct past temperatures by averaging proxy records. Because warmings and coolings are often nonsynchronous, this procedure generates relatively low amplitude variations (0.4-1.1°C). Critics are impressed instead by the amplitudes of temperature variations in individual records and thus they average, not the records, but the amplitudes. This results in significantly larger amplitudes (1.0-2.0°C). Regardless, if the averaged amplitude for the last 1,000 years is close to the upper limit (greater than 1°C), natural forcing will have to be reassessed since solar alone would then be big enough to account for almost all 20^th century warming. If, on the other hand, and this appears to be the case, the amplitudes turn out to be less than 0.7°C and other forcings such as volcanoes provide a serendipitous negative forcing[48,49], then solar forcing becomes smaller, accounting for less than half of 20^th century forcing. Finally, here it could be objected that other sources of natural variation might have been involved in a significant way in determining climate over this time period. It is hard to answer this objection because such unidentified variation might take the climate in either direction. In the absence of indication that additional substantial variations may be present, the community has made the working assumption that if present, they are minor.

A common alternative explanation to anthropogenic greenhouse gas climate forcing in the 20th century is that our Sun is responsible for observed temperature increases. Have not there been significant temperature variations during the Holocene (last 10,000 years) when there were no fossil fuels being burned?³ If the Sun's variability caused those, why could not it be causing the present warming? Perhaps a more focused question would be: "How much of the observed warming in the 20th century is caused by changes in solar activity?"

Much work has gone into looking for correlations between northern hemisphere (NH) temperature variations and proxy determinations of solar variability (because land surface, which predominates in the NH, responds to solar variation more than sea surface).3 There seems no other way to explain such historical events as warming in medieval times and cooling in both the 17^th and 19^th Centuries excepting by solar variations and resulting feedbacks (with the complication of assorted large volcanic eruptions)[49].

The Sun is a variable star. In fact other stars like our Sun are seen to vary also[50]. In a cycle that varies in length from 10 to 12 years, sunspots come and go, the solar wind strengthens and weakens, and other associated phenomena are observed changing. Also the longer-term level of solar activity changes with time - now getting stronger with more sunspots, now getting weaker with fewer (see sunspot numbers plotted in Fig. 12). In fact in the 1600s there were no sunspots at all! This has been called the Maunder Minimum and occurred at roughly the same time as the so-called "Little Ice Age" when temperatures around the planet were cooler than normal.

If one looks at the record of sunspot numbers vs. time, the 10- to 12-year cycle is obvious, but also obvious is that the maximum number of sunspots in each cycle varies. These appear loosely to be modulated by an 80 year envelope, called the Gleissberg Cycle.⁴ Comparison of the ups and downs at the maximum of each activity cycle with NH temperature estimates over the past few hundred years shows a correlation, but it is not strong largely because at times it lags the temperature changes by up to two decades. However, another correlation was found that was much better. Instead of correlating with activity amplitude, Friss-Christensen and Lassen[51a,b] correlated NH temperature change with the length of the solar cycles. Surprisingly, this gave a very good correlation, particularly because its ups and downs in the 20th century were some 20 years on average ahead of a similar pattern found in the solar activity amplitude record, allowing it to coincide with temperature changes. Extension of this proxy back to 1550 seemed to give equally good results (Fig. 11a) - but another attempt to repeat this work arrived at less good correlations (Fig. 11b).

To give these correlations a more physical basis, all that was needed was a quantitative measure of irradiance change with solar activity, because, unless solar irradiance was varying with solar activity, there was no accepted mechanism for how these changes could affect Earth's climate. Satellites began measuring the solar flux in 1979, and so by the early 1990s we had those numbers.⁵ Armed with this 20-year quantitative measure, several groups attempted to calibrate the changes in solar irradiance in the past few hundred years. Most quoted are two works: (1) Hoyt and Schatten[52] reasoned that changes in convective activity were the root cause of changes in the length of the solar cycle. They related irradiance changes to this and developed a proxy record based on length of the cycles that could be followed back to 1700. Lean et al.[53] developed a similar reconstruction but based on activity amplitude of the solar cycles that could be traced back to 1610 (Fig. 12).

To assess how much of the proxy temperature variations over this time interval could be attributed to solar activity variation, energy balance models using these proxy irradiance reconstructions and assuming feedback mechanisms similar to those found for GHG forcing were employed. These showed that solar variations could account for most of the NH temperature changes until midway in the 20th century. They suggested that increases in solar irradiance could account for 30 to 50% of the 20th century warming (Fig. 13). Some attempts to reproduce these results with better models have tended to side with the lower part of this range[54], and recent studies[48,55] and suggest only about 25%. (These included the cooling effects of volcanic eruptions and used the most recent temperature reconstructions [Fig. 9] which show less temperature variation than earlier work.) However, other recent detailed studies using both H&S and Lean et al. forcing[56,57,58] show that, in order to explain the 20^th century record well, one needs 30 to 45% solar and the rest a combination of AGHG positive forcing and aerosol negative forcing.

So the answer to our original question is that, yes, some of the warming in the 20^th century was caused by increases in solar activity-related irradiance, but not all, apparently not even half. See the recent excellent summary article on this by Lean and Rind[59].

Largely because direct (irradiance) solar forcing is unable to explain the 20^th century NH temperature surface record, and because many think temperature amplitude in the last 1,000 years was fairly high, it has been suggested that there is additional forcing from solar activity variations due to some indirect, solar-related amplification mechanism. Several have been proposed[60,61], but most center around two mechanisms:

1. Changes in the strength of the solar wind
2. Changes in the Sun's ultraviolet (UV) radiation

In the first of these, the proposed mechanism is the known modulation of cosmic rays by the solar wind. During high/low solar activity, fewer/more cosmic rays get to the Earth due to stronger/weaker solar wind intercepting them. It is suggested that cosmic rays provide cloud condensation nuclei for condensing water vapor. Thus, during high solar activity, fewer cosmic rays would mean less cloudiness/more sunlight and a warming would occur. Conversely, during low solar activity the reverse would cause cooling. If this were so, the reasoning goes, global cloudiness should correlate well with changes in the solar wind strength. Svensmark and colleagues have published papers showing this correlation using satellite-determined cloudiness[62,63]. However others[64,65] do not concur. When they look at the record, they do not get a very good correlation between cosmic rays and clouds. There has arisen a back and forth exchange on what is the best data set, what is the best way to sample it, what are the problems with it, etc. There are two points to keep in mind here:

1. Cloudiness seems to be in the eye of the beholder.
2. Cloudiness is not climate change. There must be a concomitant global temperature change.

There are basically two records. One is from ground-based observations of cloudiness over the past 50 years or more. These records see low clouds, an important point since these are perhaps the most important in cooling the planet. But this means that often upper level clouds cannot be seen. Also it is clear that ground-based observers sample but a small part of global cloudiness. This record shows no correlation with solar activity. Satellite records are more comprehensive in coverage, but these observations are a mere 20 years old (~2 solar activity cycles). Further, Norris[66] points out that there seems to be an observational selection factor determined by what areas the stationary satellite observes most. A third curious observation adds to this discussion. By observing Earthshine reflected from the New Moon one can watch the Earth's albedo change with time. Early results seem to show that the albedo changes along with the solar cycle as the cloudiness mechanism[67] suggests. This would seem to support the solar indirect forcing hypothesis, but it says nothing about point #2 above, and the solar cloudiness forcing mechanism has been questioned on the grounds that, if it were large enough to account for all the 20th century warming, its effects would be seen as a marked temperature variation over the Sun's 11-year activity cycle (point # 2). Arguments that the ocean's heat capacity would induce a lag in the climate system ignore observations that the Earth's temperature has indeed been observed to vary over the solar cycle albeit with very low amplitude commensurate with variations in direct solar forcing. Indeed, White et al.[68] showed that the very global ocean, which is supposed to cause a lag in temperature, actually is observed to have heat variations in its upper layers in sync with the solar cycle. However, these variations are small and can be largely explained by solar direct forcing, which is too weak to cause the observed secular warming. Still, some solar indirect forcing may be present. (Recently they have suggested that a small solar indirect forcing may be ordering natural modes of variability of the Pacific Ocean[69].) Others come to much the same conclusion[70,71] of very small indirect effect. However, even if some indirect (cloudiness) forcing exists, solar activity peaked around 1960 and so it is unlikely to add much to warming since then.

The second proposed mechanism for solar indirect forcing is changes in stratospheric ozone due to large ultraviolet solar radiation changes over a solar cycle[72,73] . While the Sun's total irradiance change is small over a solar activity cycle (~0.15%), its UV radiation changes by some 10%, strongly affecting ozone abundance in the stratosphere. Since ozone is a greenhouse gas, this should affect global temperatures. Labitze and Van Loon[72] observed changes on alternating cycles of the upper atmosphere's quasi-biennial oscillation, and Haigh[74] and Shindell et al.[75] have attempted to model the resulting affect. It appears to be present but again small. Finally, Udelhofen and Cess[76] found a positive correlation (0.71) of U.S. cloud cover with solar activity, but not with the variations of cosmic galactic rays as Svensmark has suggested, which should be anticorrelated. They also found that cloud cover simulated in a coupled climate model[109] was correlated with solar activity. However, since only solar direct forcing was employed, they concluded that the modeled variations were due to the modulation of UV radiation as suggested above. Other modeling studies are able to reproduce multidecadal global temperature observations without recourse to the Svensmark mechanism[58,108]. There remains the possibility that improved computer models will show that some solar indirect forcing is necessary and why the 11-year solar activity cycle is not more evident in the temperature data.

The magnitude of solar forcing varies with the amount of solar activity. In addition to the 10- to 12-year variations, solar maxima vary over multidecadal periods. Of particular interest were the Maunder Minimum, when solar activity was minimal for several decades, and the sharp multidecadal increase in solar maxima during the first half of the 20^th century. Satellite observations of solar irradiance for the past two decades have allowed calibration of variations in solar activity. A number of attempts at reconstructing past climate variations via solar forcing have shown that, prior to 1940, direct forcing by changes in solar irradiance can explain most of these changes, but these have been unable to account for warming since 1975. Indirect solar forcing effects seem to be small, but merit continued study.

GCMs integrate physical processes that regulate the Earth's climate by coupling models of the atmosphere, land, oceans and sea ice. The climate system is so complex that without these models it is difficult to make sense of the myriad of data on the various aspects of how the Earth handles its heat budget from incoming sunlight to outgoing longwave radiation (OLR).⁶ Consequently, these models are very useful in guiding the intuition of researchers and have taken their place as indispensable in the ongoing attempt to understand how our world works. This was an extremely useful role when climate was viewed as something totally natural to be studied. But with the advent of the possibility that humans are changing the climate, this role has expanded as policymakers seek advice on the magnitude of changes and how to deal with them. GCMs now carry this heavier burden because their performance can affect entire changes in societal actions and have large economic and social consequences. The question is, just how well can one expect to simulate the chaotic, complex processes that go into making up climate. Is it even possible? What can we reasonably expect the models to tell us? What will they never be able to simulate? Can they be relied upon to predict climate?

The first step in answering these questions is to compare model results with observations of the present climate to see how well the GCMs reproduce the real world. These comparisons can be divided into a few categories that address the questions: Can the GCMs reproduce climate:

· At differing latitudes?
· At different seasons?
· In various regions?

In making these comparisons, the climate community has settled on several quantities that best approximate the system. These are spatial and temporal patterns of precipitation, cloudiness, temperature and OLR. Of course in validating individual processes a much larger number of observed quantities is compared, but these four are most often looked at.

Two of the components of the GCMs - atmosphere and land - are evaluated through intercomparison efforts. The Atmospheric Model Intercomparison Project (AMIP) and the Program for Climate Model Diagnosis and Intercomparison (PCMDI) intercompare some 20 different atmospheric models from around the world[85]. Modelers are asked to present results of runs forced by a standard set of observations. The results are compared against observed quantities such as precipitation, temperatures, etc. Figs. 14 and 15 show some typical comparisons of temperature and precipitation, respectively.

Land models are also intercompared in this way, but there are fewer of them. Ocean and sea ice models are not yet formally intercompared, but a considerable effort goes into validating them against data, and intercomparison projects are beginning.

Ultimately, if the climate models are to be useful for predicting future climates with anthropogenic effects, they must be able to simulate larger integrated observables.

Hartmut Grassl[86] has suggested that the models need to simulate four types of events:

1. The present climate
2. Climate variability since the start of the instrumental record (involving a given set of forcings, and reproducing typical interannual and decadal variability)
3. A different climate from the past such as glacial periods or times of much warmer climate
4. An abrupt climate change event such as the Younger Dryas abrupt cooling during the last transition from glacial to interglacial temperatures

Most GCMs pass this test. First, we recognize that there is a large natural greenhouse effect in operation keeping the average temperature of the Earth above freezing. The average temperature of an airless planetary body at this distance from the Sun is 255°K or -18°C, well below freezing. Water vapor in the Earth's atmosphere is the major GHG and has increased its surface temperature by some 33°C. Computer models simulate this greenhouse effect; however, they often exhibit a slight warm or cool bias, running stably at an average temperature a degree or so different from the observed climate.

Also, GCMs are able to reproduce the seasonal cycle, and they no longer need to be adjusted to keep the ocean from rendering them too cool. Finer vertical resolution in the oceans and better advection methods have largely solved the problem of climate drift. (Some models continue adjust SST and salinity modestly to keep them from gradually cooling.)

The natural variability test is more complicated since there are roughly three kinds of natural variability:

1. Those forced by changes in solar activity and by large volcanic eruptions whose aerosols are lofted into the stratosphere where they scatter sunlight away from the Earth
2. Aemiperiodic variations associated with large ocean areas such as ENSO, PDO, NAO, etc.
3. "Intrinsic variability" - chaotic variability of the interactions within the system

The first of these (1) is complicated since there is not general agreement on just how much or in what manner solar variability affects the climate (as discussed above in the previous section).

Since some aspects of solar forcing have yet to be adequately understood and quantified, the ability of GCMs to simulate solar variations must be considered to be adequate but unfinished.

Simulation of cooling due to massive volcanic eruptions, however, has been well done. Following the eruption of Mt. Pinatubo several modelers decided to attempt not only to simulate its effects, but to predict them. Using the rates of decay of volcanic aerosols from the stratosphere based on El Chichon but the estimates of the quantity of aerosols from Pinatubo, they drove their GCMs with the resulting changes in sunlight reaching the lower atmosphere and surface. The codes predicted a rapid global cooling over half a year followed by a more gradual 3- to 4-year warming back to normal climate. The modelers waited to see if their predictions would be correct in both magnitude of temperature change and in timing. The climate performed essentially identically to the models' predictions (Fig. 16). This is a fairly significant result, because it showed that ocean inertia to heating changes was accurately simulated, and, further, it gave a good indication of the magnitude of that inertia in terms of lags between forcing and response (~6 months).

Test (2) is a more stringent test since it requires the GCM's interacting components (in particular the atmosphere and ocean) to produce semiperiodic behavior internally without outside forcing. The most dramatic of these is ENSO, in which El Niños occur every 3 to 7 years. This forcing is indeed chaotic, making precise reproduction of observed ENSO events unreproducible; however, the GCMs are expected to be able to produce ENSO-like events of similar frequency and amplitude. Until recently, GCMs did poorly on this test, but the most recent simulations with higher spatial resolution now seem to be reproducing all the major decadal and interdecadal semi-periodic events - ENSO, PDO, NAO, etc.[87,88,101] and Fig. 17).

On the down side, the models are still not able to predict ENSO events when started from actual climate conditions. They have some skill and do a fair job 6 months ahead of time, but have not been able to extend that to a year or 18 months. Since ENSOs are initiated partly by changes in ocean upwelling, it may be that the model initiation is deficient in capturing deep-water behavior.

Intrinsic variability (3) is of lower amplitude and occurs at all frequencies. It is due primarily to turbulence in the system. Again, the better models are now reproducing fairly well the observed frequency/power curve (Fig. 18).

Test 2 also requires the GCMs to reproduce the observed temperature record over the past 100 to 150 years. To do this, one must drive the climate with both natural and anthropogenic forcings. In addition to reproducing the general rise in 20^th century temperatures, the models must deal with several specific problems, the main ones being:

1. Rapid rise in temperature between 1920 and 1945, when AGHGs were at low concentrations; this rapid rise is thought to be due to a marked increase in solar activity, in addition to modest increases in AGHGs (and possibly increases in the North Atlantic thermohaline circulation)
2. Abrupt end to the rapid warming after 1945 followed by actual cooling in the NH until about 1975 with rapid warming thereafter
3. Greater warming at night and during winter
4. Departure of mid-troposphere warming from agreement with continued surface warming after 1992
5. Changes in heating of the deep ocean as observed by Levitus and colleagues[89]

Many model runs have been made, studying the effects of these forcings. If forced only by increases in AGHGs, all models fail criterion by not warming rapidly enough before 1945, and criterion 2 in that they predict smoothly rising temperatures to levels significantly higher than observed. However, when forcing from increases in solar activity plus aerosols from heavy air pollution following WWII are included, the entire record can be reproduced fairly closely (see "Attribution and What Might We Expect in the Future"). Thus GCMs satisfy criteria 1, 2 and 3. This conclusion has to be tempered, however, in view of the large uncertainties involved in radiative forcing (Fig. 19), particularly that of aerosols and from amplification of warming from feedback.

Aerosols alternately scatter, absorb, and reradiate radiation in the atmosphere[90]. Additionally they affect cloudiness and cloud properties. Fig. 19 shows the many different forcings that need to be incorporated into GCMs, few of which include them all. Given the uncertainties of their amount, type and radiative properties, one can see that aerosols might be sufficiently cooling as to completely counter AGHG warming. This review cannot give a detailed discussion of aerosol studies, but will note that new work[91] indicates soot to be more of a warming forcing than previously thought although Ramanathan et al. show it sometimes simply changes the vertical distribution of the heating[92]. The ability of aerosols to either warm or cool the climate complicates any attempt at prediction of future warming, because society may decide to make large reductions in them or they may continue to increase as developing countries advance.

Feedbacks cause the climate to warm or cool more than it would simply from increases in AGHGs or solar activity. Positive feedbacks are due to increased water vapor, itself a GHG, melting sea ice (changing the Earth's albedo), and possible changes in cloudiness. How much feedbacks dominate is called the model's "climate sensitivity". A standard measure of this is the model's equilibrium response in global temperature rise due to a doubling of CO₂ in the atmosphere. Doubling AGHGs produces a temperature increase of about 1.3°C. But feedback amplification increases this warming to between 2.0° and 4.0°C depending on the model used (see Fig. 22 later). This is very important for studies of greenhouse warming since high sensitivity models predict larger temperature rises in the 21^st century than the lower sensitivity ones. And so it is important to find ways to determine which sensitivity is correct. This involves validating the feedbacks in the model, which is very difficult to do. Studies of paleoclimates both considerably warmer and cooler than present give a similar range in feedback uncertainty, lending credence to the models but not reducing the uncertainty[93], and so considerable work is being done to improve treatment of water vapor and clouds in the models.

Water vapor feedback is especially difficult to determine because it depends on the vertical distribution water vapor[94]. In the boundary layer close to the Earth's surface, evaporation is a fairly well-known function of temperature and wind speed. On the other hand, in the free troposphere above, where vertical convection interacts with the large scale flow, our understanding is considerably less (models tend to be too moist high in the troposphere). This is disquieting since it is here that increases in water vapor produce the large responses to increased greenhouse gases. Improvements are complicated because observations are either lacking or inadequate.

Cloud feedback is closely related to water vapor, which complicates uncertainties in formation, type, altitude, duration, extent and perhaps most of all interaction with radiation. Clouds mostly scatter visible, shortwave radiation (SWR) while they absorb and re-radiate infrared, longwave radiation (LWR). However recent observations suggest that they absorb about 20% more SWR than previously thought, which may be due to soot particles in the droplets. The behavior of both SWR and LWR in clouds depends on many factors: thickness, droplet size, total moisture content, aerosol content and effects. High clouds warm by re-radiating LWR downward, and low clouds cool by upward reflection of SWR in addition to radiating LWR. Satellite observations show that net radiative cloud forcing is negative[95], causing cooling of the climate system. However, their feedback response to warming is poorly known - even to the sign of the forcing! In addition clouds produce precipitation. Not only is this a key quantity important to humans, the attendant phase changes during the precipitation and re-evaporation process are important to the total heat transfer within the atmosphere as well as to the vertical distribution of moisture. Thus, studies of the microphysical processes involved are an important part of climate research. In addition they are necessarily subgrid scale in size and so must be parameterized.

Until very recently all models failed criterion 4 in that they predicted warming aloft would be as much or slightly greater than at the surface. This prediction was largely true prior to 1992, but, following the cooling caused by Mt. Pinatubo's eruption, warming in the mid- and upper-troposphere slowed markedly nearly to zero while observed surface temperatures continued to rise. Models are only now beginning to explain how this is happening. (For a discussion of this conundrum, see "Surface vs. Satellite Temperatures".) Many think the problem is inadequate treatment of energy exchange between the stratosphere and troposphere. For most models vertical resolution is too coarse and/or ozone depletion is not precisely included, and thus they do not adequately treat the observed dramatic cooling in the stratosphere since 1992. The stratosphere is currently cooler than any time since observations began 50 years ago. This could cause the upper troposphere also to be cooler, thus changing the temperature lapse rate so as to reduce warming aloft below what it is at the surface. Spatially one-dimensional models show this effect[96].

It remains for complete GCMs to attempt this difficult calculation comprehensively. It has been attempted in at least two recent simulations, those done at the Max-Plank-Institute (MPI)[97], and more recently at NASA's Goddard Institute for Space Studies[18]. The MPI runs d