World Student Community for Sustainable Development

by Philip Michael Sheehy

Debates on global warming are often framed as uncertain vs. certain climate science. The critics of climate science refer to existing uncertainties as the basis for their rejection of aggressive climate policy. On the other hand, environmentalists consider available scientific knowledge sufficient to call for immediate reductions of greenhouse gas emissions. However, the reasoning of both sides is problematic. Like any scientific endeavor, global warming projections are not free of uncertainties. The issue facing us is not whether predictions are certain or uncertain, rather how to quantify the uncertainties of global warming. The climate science community has begun analyzing the probability distribution of global warming as a measure of risk. I introduce this new mode of thinking, and discuss its strengths and weaknesses.

By Masahiro Sugiyama.

It was a great disappointment for the international community when, in March 2001, President Bush withdrew the United States of America from the Kyoto Protocol, an international accord for combating global warming. The Bush administration argued that, considering existing uncertainties in climate science, it would be wiser to wait for more scientific knowledge to accumulate, with the assumption that more research will reduce the uncertainties. Environmentalists were quick to respond with their own claims of an existing scientific consensus in climate science that pointed to a necessary and immediate reduction of greenhouse gas emissions. Both sides have been quite critical of each other, and the argument is now a familiar one. However, the logic on both sides is problematic—neither side fully appreciates the nature of uncertainty inherent in global warming science.

As a beginning climate scientist, I am forthright in my admission of the large uncertainties associated with global warming projections. Climate scientists have developed a measure of global warming, so-called climate sensitivity, which is defined as a global-mean surface temperature increase in response to a doubling of carbon dioxide concentration in the atmosphere, compared to the pre-industrial level. In spite of recent and rapid advances in the field, such as the study of the El Niño phenomenon and increasingly sophisticated climate simulations, estimates of this parameter have stayed almost in the same range for decades.

Then it is not surprising that the most recent report by the Intergovernmental Panel on Climate Change (IPCC), the most authoritative review in the field, still projects a wide range of warming: 1.4 to 5.8 ºC (2.5ºF to 10.4ºF) in 2100 relative to 1990 (IPCC 2001). (This range reflects uncertainties not only in natural science but also in socio-economic trends, which translate into different emissions scenarios). Tending to emphasize the extreme end of this scale, the media are often responsible for general confusion and misrepresentation of global warming projections. However, the IPCC assigns no likelihood to these numbers. The warming may be as disastrous as 5.8ºC (10.4ºF), or it might be as moderate as 1.4ºC (2.5ºF), or the temperature increase could be outside this range (Reilly et al. 2001).

In the modern era, we benefit from science and technology in our everyday lives—from the Internet to subways, to automobiles, to medicine. We take it for granted that science and technology are firmly established, solid, and unbreakable. True, theories in natural science are tested by experiments, and only the theory, which survives harsh experimental tests, is accepted as scientific knowledge (Lindzen 1990).

The process of building scientific knowledge is a demanding one; the rigors of this process are amplified in the case of an observational science, such as climate science. In atmospheric and oceanic sciences, it is difficult to conduct experiments. There is only the single earth, and you cannot modify the earth just for your experiment (although it is continuously being affected by natural and anthropogenic influences). Like much of the field of environmental science, climate science is classified as an observational science, which can be contrasted with experimental science. Typically testing a theory requires several observations, which implies that we would need several climate change events to fully confirm our theory on global warming. Clearly, the progress of climate change science is strongly constrained by observational limitations.

Yet researchers haven’t lost hope. Advances in atmospheric science and oceanography have demonstrated that basic physical and chemical laws, which can be expressed as mathematical equations, govern the behavior of the climate. The recent surge in computing abilities has allowed climate scientists to perform increasingly sophisticated climate simulations by solving complicated mathematical equations. We enjoy such advancements in our everyday life: Contemporary weather forecasting is done mostly by computer simulations similar to climate simulations, making the weather forecasts more and more reliable.

Even for climate simulations, there exist uncertainties. To perform a climate simulation, a computer needs approximations of the equations, rendering the simulations imperfect (which would be perfect otherwise aside from the uncertainties in how you start the simulations). The way in which a computer approximates the real equations is similar to the way a digital camera recognizes the world. A computer uses a “mesh†to simulate the earth. The finer the “mesh†is, the better the outcome is.

(See a schematic of “mesh†and compare fine and a coarse “mesh†.)

If you have a digital camera or any graphic computer software, you might already be familiar with the trade-off between the resolution and the computer resources. Finer resolution is desirable, but it consumes more computer resources and time. This is also the case with computer climate simulations. The reality is that even with the fastest supercomputer available at present, you cannot make the “mesh†in the computer fine enough for satisfactory simulations of global warming.

To date, simulations have been unable to directly “see†clouds, precipitation in the atmosphere, and so-called small-scale mixing in the oceans, and so forth. Oceanographers and meteorologists have invented semi-theoretical/semi-empirical approaches (so-called parameterizations) to calculate the effect of the unresolved phenomena. Thanks to these devices, a modern supercomputer can reproduce the current climate reasonably well. The trouble is, there is no guarantee that the semi-empirical/theoretical equations would hold later in this century, or under different climate states. The ultimate way of testing the approximations is, of course, to observe the real climate change, waiting until 2100. The dispersion of global warming projections from different research laboratories is primarily due to differing approximations of this kind in their calculations. Critics of global warming focus on this particular aspect of global climate simulations.

If global warming projections are uncertain, should we postpone climate policy decision while we wait for more scientific findings? The answer is, in fact, no. Or to be more accurate, it is not so simple. Conventionally the debate on global warming has been framed as certain vs. uncertain climate science, but this framing is misleading. Existence of uncertainty simply tells us that there is a risk of global warming. The issue is not whether to act now or wait for more research; rather how to quantify the uncertainties, and how we should deal with this risk. (For more discussions on decision-making and risks, see Webster 2002, for example. He discusses the sequential decisions under uncertainties declining with time.)

The climate community has taken its first step toward this mode of thinking. Let’s look at some results from MIT. Webster et al. (2003) calculate the probability distributions of the global-mean temperature increases for hypothetical no-policy and policy cases. The estimates are based on uncertainty estimations of parameters of the earth’s climate (Forest et al. 2002) and emissions of greenhouse gases and other radiatively active agents, so-called aerosols (Webster et al. 2002).

If no restrictions on greenhouse gas emissions are placed, there is a 50% chance that global-mean surface temperature will increase by more than 2.4ºC (4.3ºF) by the year 2100, and a 5% chance that the increase will be outside the range 1.0 to 4.9ºC (1.8 to 8.8ºF). Webster et al. also consider a policy case, where Kyoto Protocol-like reductions are assigned to all countries so that developed countries’ emissions are 35% below 1990 levels in 2100, and that for developing countries is 30% below. In this case, the temperature increase is lowered to a 50% chance of exceeding 1.6ºC (2.9ºF), and a 5% chance of being outside the range 0.8 to 3.2ºC (1.4 to 5.8ºF). Although small, the assigned probabilities for substantial warming are striking. It is important to note that while imposing a stringent reduction on greenhouse gas emissions reduces the likelihood of the unwanted event, it does not eliminate the risk.

This is one example of such calculations, and there are other estimates using different methodologies. Although there are serious debates concerning the technical details, we must also confront a fundamental issue: What does a probability distribution mean for only a single event?

Let’s take a familiar example of weather forecasting. When you come back home and turn on the TV, you see the TV weather forecaster explaining, “there is a 50% chance of rain tomorrow.†The next day, you actually don’t have rain. Does this mean that the forecast is wrong?

In fact, you cannot say that it is either right or wrong. Determining whether your favorite forecaster is doing her/his job right needs multiple forecasts. If she/he has forecast a 50% chance of rain 100 times, and actually it has rained 50 times out of 100, then her/his forecast is accurate. Unless you have a number of forecasts and verifications, you cannot be sure whether the probabilistic forecasts are good or not.

What about probabilistic climate forecasts? In practice, we would have only one actual global warming scenario, and would never be able to determine whether the forecast is correct or not. Then, is this new scientific endeavor useful? Yes, the virtue of probabilistic forecasts is that they at least make it clear that uncertainties are inevitable, and that the climate policy is about how to deal with the risk.

There are more topics regarding probabilistic climate forecasts to be explored. It is hard to communicate this sort of information to the public and taxpayers without degrading the content. The nature of information on uncertainty does not easily fit the traditional style of politics—two competing parties debate on an issue, and voters decide which to choose. We have just begun our first step toward assessing the risk of global warming. Much remains to be learned.

General references:
- Stone, P.H., 1992: Forecast cloudy: The limits of global warming models. Technology Review, February/March 1992, 32—40.
- Center for Climate System Research, University of Tokyo, website. http://www.ccsr.u-tokyo.ac.jp. (Mostly in Japanese. Some English pages are available)

References:
- Forest, C.E., P.H. Stone, A.P. Sokolov, M.R. Allen, and M.D. Webster, 2002: Quantifying uncertainties in climate system properties with the use of recent climate observations. Science, 295, 113—117.
- Lindzen, R. S., 1990: Dynamics in atmospheric physics. Cambridge University Press, 310pp (especially, chapter 9).
- IPCC, 2001: Climate Change 2001: The Scientific Basis. Contribution of Working Group I to the Third Assessment Report of the Intergovernmental Panel on Climate Change [Houghton, J.T.,Y. Ding, D.J. Griggs, M. Noguer, P.J. van der Linden, X. Dai, K. Maskell, and C.A. Johnson (eds.)]. Cambridge University Press, 881pp. (Also available at http://www.ipcc.ch/.)
- Webster, M.D., 2002: The curious role of “learning†in climate policy: Should we wait for more data? The Energy Journal, 23(2), 97—119.
- Webster, M.D., M. Babiker, M. Mayer, J.M.Reilly, J. Harnisch, M.C. Sarofim, and C. Wang, 2002: Uncertainty in emissions projections for climate models. Atmos. Env., 36(22), 3659—3670.
- Webster, M.D., C.E. Forest, J.M. Reilly, M. Babiker, D. Kicklighter, M. Mayer, R.G. Prinn, M.C. Sarofim, A.P. Sokolov, P.H. Stone, and C. Wang, 2002: Uncertainty analysis of climate change and policy response. MIT JPSPGC Report 95 ()