Can computer models predict future climate?

General circulation computer models (GCMs) are deterministic, which is to say that they specify the climate system from the first principles of physics. For many parts of the climate system, such as the behaviour of turbulent fluids or the processes that occur within clouds, our knowledge of the physics is incomplete, which requires the extensive use of parameterisation (read “educated guesses”) in the computer models. The modellers themselves acknowledge that they are unable to predict future climate, preferring the term “projection” to describe the output of their experiments which the IPCC then incorporates into socioeconomic “scenarios” (e.g. IPCC, 2001). This terminology highlights the fact that GCMs are unvalidated and do not provide skilled predictions of future climate out to 2100. Also, it transpires, first, that none of the models was able to forecast the path of the global average temperature statistic as it elapsed between 1990 and 2006. And, second, GCMs persistently predict that greenhouse warming trends should increase with altitude, especially in the tropics, with the highest trends at around 10 km height; in contrast, actual observations show the opposite, with either flat or decreasing trends with increasing height in the troposphere (CCSP, 2006).

Individual GCMs differ widely in their output under an imposed regime of doubled carbon dioxide. The IPCC (2001, Fig. 5d) cites a range of 1.8 to 5.6⁰ C warming by 2100 for the model outputs that they favour, but this range can be varied further to even include negative outputs (i.e. cooling) by minor adjustment of some of the model parameters (Essex and McKitrick, 2002). When climate modelling experiments produce such cooling, the output is discarded as “obviously wrong” (Stainforth et al., 2005).

At the time of writing this paper, it was becoming widely conceded, even amongst the IPCC community, that GCMs are not capable of delivering useful climate predictions. Thus IPCC lead author for the Working Group 1 science report, Kevin Trenberth, writes on Nature’s Climate Feedback blog that “The state of the oceans, sea ice, and soil moisture has no relationship to the observed state at any recent time in any of the IPCC models. There is neither an El Nino sequence nor any Pacific Decadal Oscillation that replicates the recent past; yet these are critical modes of variability that affect Pacific rim countries and beyond”. Accordingly “... there are no (climate) predictions by IPCC at all. And there never have been”, but instead only ““what if” projections of future climate that correspond to certain emissions scenarios”. This view is reinforced by another WG1 lead author, Jim Renwick, Renwick, responding to an audit which showed that the climate forecasts issued by New Zealand’s National Institute of Water and Atmosphere (NIWA) were accurate only 48% of time. In other words, one can do just as well by tossing a coin. Renwick’s comment was that “Climate prediction is hard, half of the variability in the climate system is not predictable, so we don’t expect to do terrifically well”.

A third blow to the credibility of IPCC GCM forecasts was delivered by Armstrong and Green (2007) in an audit of Chapter 8 in the latest IPCC report. They find that “in apparent contradiction to claims by some climate experts that the IPCC provides “projections” and not “forecasts”, the word “forecast” and its derivatives occurred 37 times, and “predict” and its derivatives occur 90 times in the body of Chapter 8”. Having analyzed the IPCC’s approach in detail, Armstrong and Kesten’s conclusion is that “because the forecasting processes ..... overlook scientific evidence on forecasting, the IPCC forecasts of climate change are not scientific”.

These various criticisms of GCM climate modelling can be summed up in the following brief statement - there is no predictive value in the current generation of computer GCMs and therefore the alarmist IPCC statements about human-caused global warming are unjustified.

A second use of GCM modelling is in climate attribution studies, whereby the known 20^th century meteorological record is retrodicted using models fed with known or presumed forcings, such as increasing carbon dioxide, volcanic eruptions and other aerosols (e.g. Stott et al., 2001; Hulme et al., 2002, Fig. 4). After many years of trials, the IPCC (2001, Fig. 12.7) reported simulations that mimicked the historic temperature record if and only if human emissions were included in the forcings. These results have been widely misrepresented as evidence for human-caused global warming. They are, of course, evidence only that a curve matching exercise involving many degrees of freedom has plausibly mimicked the 20^th century temperature curve. They are exercises in virtual reality, and not evidence of any type.

As an alternative to the deterministic GCM approach, there exist several other types of computer model of empirical nature. Such models use analysis of a portion of the climate record to establish the pattern of past temperature change and then project this pattern into the future. Papers include Kotov (2001; application of chaos theory to last 50,000 years of Greenland ice core data), Klyashtorin and Lyubushin (2003; analysis of last 150 year record of the global average temperature statistic), Loehle (2004; analysis of last 5,000 years of temperature record contained in a Caribean deep sea core and a South African speleothem) and Zhen-Shan and Xian (2007; analysis of Chinese temperature record from 1881-2002). These papers yield almost unanimous estimates of 21^st century cooling rather than warming (e.g. Fig. 5). Unlike GCM scenarios, the results are consistent with the observation that global average temperature peaked in the El Nino year of 1998 and has remained static or slightly declined since (cf. Fig. 6). In parallel studies rooted in solar physics, projection of the cyclic historic pattern of sunspot activity suggests that a forthcoming 21^st century cooling will be driven by falling solar activity, perhaps even to the level of the cold Maunder minimum in the 17^th century (Bashkirtsev and Mashnich, 2003; Abdussamatov, 2006).

To summarise, empirical computer projections of 21^st century cooling are more consistent with the available data than the greenhouse warming projected by GCMs. Though deterministic GCMs are a valuable heuristic tool, they all rest upon the Kelvin fallacy, i.e. the assumption that the physics of the system is fully known. In essence, GCMs do not produce accurate climate predictions and they are therefore unsuitable for direct use in policy making (Khandekar, 2004).

Is there a consensus?

Argument based on consensus is not usual in science, for reasons that have been summarised by writer Michael Crichton. “Let’s be clear: the work of science has nothing whatever to do with consensus.  Consensus is the business of politics. Science, on the contrary, requires only one investigator who happens to be right, which means that he or she has results that are verifiable by reference to the real world. In science consensus is irrelevant. What is relevant is reproducible results. The greatest scientists in history are great precisely because they broke with the consensus…”.

It would be hard to write a more accurate statement about the way that science works than Crichton’s pithy summary. It can be noted in support that we do not usually say that “there is a consensus that the sun will rise tomorrow”. Instead, the confident statement that “the sun will rise tomorrow” rests on repeated empirical testing and the understanding conferred by Copernican and Newtonian theory. Therefore, statements such as “there is a consensus that dangerous global warming will occur” convey sociological rather than scientific information. Individuals, organisations and governments that espouse such views signal mainly that they have a political agenda.

It needs to be stressed that the claimed “consensus” advice to policy makers provided by the IPCC is political, rather than exclusively scientific as portrayed in the press. Complaints from climate rationalists that the first, second and third IPCC reports were subject to political manipulation centred on the over-egging of the case for dangerous human-caused warming. Significantly, the recent release of the Fourth Assessment Report (4AR; IPCC, 2007) was greeted by strong criticisms also from supporters of the dangerous warming case; they allege that bureaucrats involved in the preparation of 4AR removed statements by scientists that highlighted climate risks, and that 4AR therefore understates the risk of catastrophic warming. Thus David Wasdell (2007), an IPCC reviewer, writes that he was “astounded at the alterations (to the final scientific draft of the full 4AR report) that were imposed by government agents during the final stage of review”. It obviously matters not whether bureaucratic interference results in exaggerating the climate change risks or minimizing them; in either case, and as is now agreed by both main sides to the global warming dispute, the “consensus” advice tendered to governments by the IPCC is political and not scientific.

Can we measure average surface global temperature meaningfully?

Essex et al. (2007) have recently argued that climate change cannot be summarized adequately by a simple statistical average of temperatures from around the globe. They assert that no average global temperature calculated for the Earth (and many different averages are possible) can have any physical meaning in the context of climate change, any more than an average telephone number has any meaning for using the telephone system. A temperature can be defined only for a homogeneous system, which climate most definitely does not represent. The processes which control climate, such as ocean currents and atmospheric circulation, are driven by local and regional temperature differences, not by a “global average temperature” statistic.

Let us ignore these arguments for a moment. Like the IPCC – which has widely promulgated a global average temperature curve based on surface thermometer readings since 1860 (IPCC, 2001, Fig. 2) (Fig. 6) – let us assume that the concept of an average global surface temperature is meaningful and pose the additional question: “is it possible to establish an accurate estimate of its magnitude?” As Dr Vincent Gray (pers. comm.) and other climate rationalists have pointed out, even if we accept the IPCC curve as a starting point for discussion, its use as an accurate measure of climate change faces five insuperable difficulties.

First, the temperature measurement sites are located non-randomly, more than 90% being onland despite about 70% of the earth’s surface being represented by ocean. Second, over time many of the measurement sites have experienced changes in their surroundings that impact on local temperature (e.g. new buildings, trees cut down or planted, ageing paint on sun enclosures), introducing a warming bias into the measurements; studies suggest that both urban heat island and rural land-clearing effects have a material influence (Christy et al., 2006; Pielke et al., 2006; Ren et al., 2007). As one example, 1881-2004 temperature data from Europe reveal a warming rate of 0.67⁰/century for urban meteorological stations as opposed to 0.37⁰/century for rural stations (Janssens, 2007). Third, the number of measurement sites used varies dramatically through time, starting in 1850 at 200 sites, building to c. 8000 in 1980, and then declining to c. 2500 today. Fourth, the temperature at each site is constructed using the statistically doubtful historic method of averaging the maximum and minimum temperatures measured once each day at the site. Fifth, and finally, the data used to construct the version of the global surface temperature used by the IPCC is not released to the public; the curve is therefore unreproduceable in the sense that it cannot be checked independently (e.g. McIntyre, 2007).

The issue of the non-release of data that underpin papers used by the IPCC, resulting in there not being able to be independently audited, is a non-trivial issue. For example, Canadian Steven McIntyre is one of several persons who have requested temperature data from the Climate Research Unit at the University of East Anglia, and in particular the data used in a classic study of the urban heat island effect by Jones et al. (1990) This request, like other similar requests, was initially refused. After appealing the decision, McIntyre recently received a letter from the Information Policy Officer at the University of East Anglia stating that the data and metadata for this study that can still be identified was to be posted for public access on the University’s website no later than April 13^th (see ). Noting the success of McIntyre’s statistical challenges to the validity of the Mann et al. hockey stick curve, his re-analysis of the Jones et al. urban heat island dataset will be of great public interest.

One is forced to the overall conclusion that – despite their pre-eminence in the public debate, and despite the laborious statistical analysis involved in compiling them - the historic temperature records reconstructed from ground thermometer data are of little value. Changes of less than 1⁰ C/century displayed on such curves may not exceed the true error bars of the average temperature estimates. Therefore, the climate records that are of most value for estimating 20^th century climate change in true context are those from high-quality proxies such as sediment cores, ice cores or tree rings. Many such proxies show no untoward warming at the end of the 20^th century, and that they usually represent local or regional rather than global climate change is no reason to discount them.

Is global average temperature rising or falling?

There is no simple answer to this question. Despite the uncertainties just discussed, the global surface thermometer dataset and various high-quality geological proxy temperature datasets are widely used as a basis for climate trend estimates. For any such dataset, the answer to the apparently innocent question posed in the heading depends entirely on the chosen end-points of the data being considered. For instance, using the Greenland ice core oxygen isotope data (proxy for local temperature), warming has taken place since 16,000 years ago, and also since 100 years ago (Davis and Bohling, 2001) (Fig. 7). Over intermediate time periods, however, cooling has occurred since 10,000 and 2000 years ago, and temperature stasis characterizes both the last 700 years and (globally, from meteorological records) the last 8 years. Considering these facts, is the temperature in Greenland warming or cooling?

Both the 8 and 100 year-long intervals of temperature change are too short to carry statistical significance regarding long-term climate change. However, though the last 100 years of  temperature record has limited climatic significance (for instance, representing only 3 climate normal datapoints), it is nonetheless important because it corresponds to the span of instrumental meteorological records from the earth’s surface. Accepting the 1860-2006 temperature record used by the IPCC (2007; Climate Research Unit, University of East Anglia) as a best measure, we find that there has been no significant increase in surface global temperature since the peak El Nino year of 1998 (Fig. 8). This result is confirmed by the two most reliable records of average tropospheric temperature, drawn from weather balloon radiosondes (since 1958) and satellite-mounted microwave sounding units (MSU; since 1979).  Of all these datasets, the MSU record is accepted to be the most accurate and globally representative. Once the effects of El Nino warmings and volcanic coolings are allowed for, this record shows no significant warming since its inception in 1979 (Gray, 2006) (Fig. 9). This conclusion is robust. Though several other global temperature datasets exist, and though the MSU record has been subject to repeated corrections in interpretation, none of the available datasets document significant recent greenhouse warming.

The global temperature stasis between 1998 and 2006 occurred despite continuing rises in atmospheric carbon dioxide over that period. Consistent with this, Karner (2002) showed from an analysis of global temperature series that “… antipersistence in the lower tropospheric temperature increments does not support the science of global warming developed by IPCC. Negative long-range correlation of increments during the last 22 years means that negative feedback has been dominating in the Earth climate system during the period”. These facts, and the lack of a discernable human greenhouse effect in late 20^th century temperature records, are consistent with Khilyuk and Chilingar’s (2006) estimate that the human greenhouse forcing is 4-5 orders of magnitude less than the major natural forcing agents.

In summary, the slope and magnitude of temperature trends inferred from time-series data depend upon the choice of data end points. Drawing trend lines through highly variable, cyclic temperature data or proxy data is therefore a dubious exercise. Accurate direct  measurements of tropospheric global average temperature have only been available since 1979, and they show no evidence for greenhouse warming. Surface thermometer data, though flawed, also show temperature temperature stasis since 1998. This pattern is not what is portrayed in the daily news media.

Are temperatures changing at a dangerous rate, or have they reached a dangerous level?

Fitting short-term trend lines through temperature or proxy temperature data with no regard to underlying climate cycles is meaningless, but this is widely ignored in the climate change literature. Prestigious science academies and ad hoc expert committees deliver reports that say (or imply), first, that meaningful trends can be identified, and, second, that the rates and magnitudes of temperature increase observed are ipso facto unusual or dangerous.

For example, the US Climate Change Science Program (CCSP, 2006), using all available instrumental data, reported late 20^th century rates of temperature increase of 1-2⁰ C/century, and IPCC (2007) estimated the overall magnitude of the temperature increase over the last 100 years to be 0.74⁰ C. However, both of these estimates ignore the presence in all climate data of decadal-centennial cyclicity (incidentally, they also ignore the presence in this particular dataset of clear El Nino and volcanic eruption signals of non-greenhouse origin). Meaningful comparative judgements about climate change cannot be made on the basis of the trivially-short, 150-yr-long thermometer surface temperature record, much less on the 27-year-long satellite tropospheric record.

To compare late 20^th century warming with earlier geological warm events requires the use of local proxy data, because no global temperature statistics are available prior to the 20^th century. One of the best such datasets that extends over an adequate period of time is the oxygen isotope record from the Greenland ice core already referred to (Grootes et al., 1993; Davis & Bohling, 2001). These data show a ~1500 year warming-cooling cycle of 1-2⁰ C magnitude. This cyclicity is probably  of solar origin (Bond et al., 2001; Singer and Avery, 2006), and the late 20^th century warming period represents a peak within it (Fig. 10). Consistent with this, Solanki et al. (2004) have shown that the activity of the sun has been building since the end of the Little Ice Age in the late 19th century, and that over the last 60 years it has been at its strongest since the early Holocene, c. 8000 ybp. In turn, Svensmark (2007, and other papers) has identified a possible mechanism whereby solar activity affects cosmic ray influx which in turn controls the cloud formation that acts as one of the Earth’s main thermostats.

The Greenland ice core data also reveal typical rates of temperature change of up to 2.5⁰ C/century for periods of cooling and warming of decadal to centennial time span (Fig. 11).

In Greenland, then, the late 20^th century warming proceeded at unalarming rates to reach a peak that was cooler than were the preceding Mediaeval and Roman warm periods. And at the other pole, in Antarctica, similar ice core evidence shows that late 20^th century temperature was up to 5⁰ C cooler than temperature highs associated with geologically recent interglacial periods (Watanabe et al., 2003). Therefore, the magnitude of the late 20^th century warming, and its rate of change, both fall well within known natural limits. In addition, the late 20^th century warming that is widely attributed to human greenhouse emissions is of similar rate and magnitude to an earlier natural warming between 1905 and 1940; in relationship to which, it has been shown that the warmest decade of the last 1250 years in the European Alps was the 1940s rather than the 1990s (Buntgen et al., 2006).

The IPCC’s (2001, p. 97) prescient diagnosis therefore remains true today:

"The fact that the global mean temperature has increased since the late 19th century and that other trends have been observed does not necessarily mean that an anthropogenic effect on the climate system has been identified. Climate has always varied on all time-scales, so the observed change may be natural. A more detailed analysis is required to provide evidence of a human impact."