The Future of Everything: The Science of Prediction

by David Orrell

  The greenhouse gases, which include water vapour, carbon dioxide (CO2), and methane, are therefore vital to our survival. However, you can have too much of a good thing. Increasing CO2 from its pre-industrial levels of about 280 ppm to 380 ppm is a substantial relative change. And even if we were to freeze CO2 emissions at current levels, its slow rate of decay means that the total amount will still continue to grow well into the future. Furthermore, because of the slow response of the ocean/atmosphere system, the effects of high CO2 levels will be with us for centuries, and may even be irreversible. A common property of non-linear systems is hysteresis: once a change has been made, it is difficult or impossible to undo.

  In 1979, the meteorologist Jule Charney organized a meeting at Cape Cod to investigate what would happen if CO2 levels were double the pre-industrial level. (This may occur sometime during the present century, with climate effects tending to lag behind as the system adapts; though as discussed below, the exact rate of CO2 production depends on a wide range of social and economic factors.) At that time, there were only two American research groups actively involved in climate modelling: Syukuro Manabe’s at the Geophysical Fluid Dynamics Laboratory (GFDL) at Princeton, and James Hansen’s at NASA’s Goddard Institute. Manabe had been involved with climate models since 1963. (These are usually simplified versions of GCMs, with some of the details stripped out so they can be run over periods of hundreds or even thousands of years.) On the first day of the meeting, he told the assembled group that according to the model, a doubling in CO2 would lead to a rise of 2°C—not too bad. The next day, though, Hansen presented the results of his model: it indicated a much larger rise of 4°C, a factor of two difference. Since the aim was only to get a rough idea of the possible magnitude, Charney chose 0.5°C as the margin of error on both calculations, which left a range of 1.5°C to 4.5°C (Arrhenius had estimated 5°C back in 1896). The lower limit was not out of line with natural variations that have occurred in recent centuries, but the upper was a real icecap melter. In words, the estimate meant that either nothing much could happen or we could have a serious problem.38

  Not everyone at the meeting was happy with the agreed range, and some described it as hand waving.39 However, it did seem to indicate that CO2 could have a considerable effect on the planet, and it helped mobilize scientific interest. The Intergovernmental Panel on Climate Change (IPCC) was established to report on the matter. At its second meeting in 1995, there were thirteen climate models to choose from; by the time of its 2001 meeting, climate modelling had grown into a major research area for groups all over the world, including the Max Planck Institute for Meteorology in Germany, the Hadley Centre in England, and the Lawrence Livermore National Laboratory in California. In a remarkable display of consistency, though, the estimate of potential warming, obtained by consensus among experts, remained little changed from what is sometimes called the “canonical” range of 1.5°C to 4.5°C, with an average of about 3°C.40 (The panel did not assign probabilities to the different outcomes, because there was no sensible way to determine them. One of the scientists observed, “The range is nothing to do with probability—it is not a normal distribution or a skewed distribution. Who knows what it is?”41)

  Some more recent sensitivity studies indicate a rise beyond that canonical range. As the climatologist Stephen Schneider noted, “Despite the relative stability of the 1.5 to 4.5°C climate sensitivity estimate that has appeared in the IPCC’s climate assessments for two decades now, more research has actually increased uncertainties!” 42 This degree of uncertainty reflects the challenge of modelling the climate system. Just as a biological system incorporates complex regulatory loops that make it difficult to model, the climate system is made up of feedback loops that elude simple parameterizations.

  GLOBAL COOLING

  Can the future climate be predicted by looking at past climates? An alternative to using mathematical models is to adopt a data-driven approach, and study how greenhouse gas concentrations and climate have varied together with the ebb and flow of past ice ages—a field known as paleo-climatology. Sources of data include ice cores (which trap gases), petrified tree rings, and geological features such as sedimentary rocks.

  There are two difficulties with this approach. First, the data contains a high degree of uncertainty. Scientists argue over how to interpret satellite pictures, let alone tree rings.Second, there is no exact analog in the past for the current situation—dinosaurs didn’t drive cars. Estimates of climate change based on paleoclimatology are therefore highly uncertain. As might be expected, they have a tendency to fall in the “canonical” range. Some scientists, though, point to extreme events in the past as proof that the climate system is more sensitive than models allow.

  One such extreme event occurred during the Younger Dryas period. This was named after a small, cold-loving plant—Dryas octopetala—which pollen records show suddenly began to flourish across much of Europe about 12,700 years ago. The cause, it seems, was a sudden and precipitous drop in temperature, which plunged the warming earth back into ice-age conditions for over a thousand years. Some believe that it could have been triggered by an abrupt reversal of the ocean circulation pattern that drives the Gulf Stream, and that melting Arctic ice could initiate a similar event in the future. Global warming may mean that some areas, such as the U.K., become much cooler.

  IN THE LOOP

  As mentioned in Chapter 4, one of the key processes involved in determining our weather is the formation and dissipation of clouds. They play an equally important role in the global climate. If the planet warms, more water evaporates into the atmosphere. Since water vapour is a greenhouse gas, this can trap heat and lead to further warming—positive feedback. If the water vapour condenses into clouds, though, these create shadow and lead to cooling— negative feedback. Except at night, when cloud cover prevents heat from escaping—positive feedback.

  Water in the form of snow and ice also affects the planet’s albedo, which is its ability to reflect light. White surfaces have high albedo, and dark surfaces have low albedo (the word comes from the Latin for whiteness). Most of the earth is blue, green, or brown, colours that are somewhere in the middle; but snow, being white, reflects about 80 percent of the sun’s energy. It therefore acts to keep the planet cool. Antarctica is white all year-round, but in the Arctic, snow and ice cover is seasonal and sensitive to changes in climate. If temperatures increase, the cover reduces and the albedo goes down. The ocean, deprived of its ice layer, also releases more water vapour to the atmosphere, which accentuates the greenhouse effect. Together these cause further warming, in a positive feedback loop known as the Arctic amplification. Small differences in ice albedos have a large effect on model results.43

  In fact, climate models must contend with feedback loops that are as complicated as those in any biological system—and usually have a biological component. Another example is the carbon cycle. Carbon is the most important building block for life. Its atomic structure gives it a unique ability to combine with other elements to form complex molecules such as DNA. Like the blood in our bodies or money in the economy, it is constantly being cycled around the biosphere. Plant life uses carbon dioxide for photosynthesis, combining it with water to create sugar and oxygen. The latter is breathed back out to the atmosphere.44 Algae on the ocean surface and other sea organisms similarly absorb huge amounts of carbon for photosynthesis. (Algae are also believed to be cloud-makers— they emit sulphur gases, which oxidize to form minute particles that encourage the formation of clouds.)

  We too are now a major part of the carbon cycle: fossil-fuel emissions and cement manufacture release about 5.5 billion tons of carbon per year, while deforestation and other land use contribute another 1.5 billion tons.45 Furthermore, global warming can affect carbon levels in a positive feedback loop, by increasing the number of forest fires and reducing the amount of CO2 held in ocean water. Large amounts of carbon, as much as 450 billion metric tons, are also stored in the frozen tundra and bore
al forests of the North.46 We have to hope that they stay frozen, since their release could lead to a runaway warming.

  The climate system therefore consists of a nested series of nonlinear feedback loops that are in a kind of dynamic balance. Unlike short-term weather models, climate models do not suffer from sensitivity to initial condition, since the aim is to predict the average weather far in the future (which should be insensitive to the exact starting point). This is especially so if the model is run for a long period under fixed conditions, in which case it settles onto its own attractor. Error in climate models is therefore primarily a result of model error.47 If GCMs were perfect, so that “predictability limitations [were] not an artifact of the numerical model,”48 then climate prediction would be easy. But this is far from being the case. Weather and climate predictions are directly linked, and to believe that models can fail at the former but succeed at the latter is nothing but wishful thinking—especially when the most basic properties of the ocean/atmosphere system prove so difficult to model.49

  WATER WORLD

  One substance that consistently slips through the grasping fingers of weather and climate modellers like water is water. Because water is so ubiquitous, we don’t often reflect on its properties; but it is perhaps the most mysterious, upside-down, shape-shifting, and life-giving substance on earth. If carbon represents the earth’s yang, fixing and organizing other substances into its geometry, then water is its yin. The two are as different as water and oil (which is 85 percent carbon).

  As Antoine Lavoisier demonstrated, a water molecule consists of a single molecule of oxygen (O) and two molecules of hydrogen (H). The frost on the ground, the mist in the air, the water in our bodies—all have this simple chemical structure. The molecule is highly polarized: one side has a positive charge, the other a negative. Since opposites attract, the positively charged side is drawn to the negatively charged side of nearby molecules, in a process known as hydrogen bonding. Collections of water molecules form a complex interacting group, each molecule aligning itself with the others in a vibrant dance, switching partners billions of times per second. This gives the substance a number of unique properties. For example, the solid form is less dense than the liquid form, so ice creates a protective layer that insulates the water below. From the level of the single cell to the entire globe, water is essential for life. About 70 percent of a typical cell is water, about 70 percent of our body weight is water, and about 70 percent of the earth is covered in ocean.

  In the climate system, water appears in far more guises than is suggested by the simple categories of solid, liquid, and gas. In its solid form, it might be ice or snow or different types of crystal in the atmosphere. As a liquid, it can appear as rain, fresh water in rivers and streams, or brine in the sea. Water vapour can evaporate from the oceans or land or be absorbed by plants, and it can condense into every different type of cloud that you see in the sky as a mix of vapour, liquid, and crystals. And water constantly changes its form: a single molecule might start off in the ocean, evaporate into the atmosphere, join a passing cloud, fall to the ground as snow or rain, enter a river, and flow back into the ocean, all within days. As Leonardo da Vinci wrote, “The waters circulate with constant motion from the utmost depths of the sea to the highest summits of the mountains, not obeying the nature of heavy matter; and in this case it acts as does the blood of animals which is always moving from the sea of the heart and flows to the top of their heads. . . . These waters traverse the body of the earth with infinite ramifications.”50 Each ramification has a different effect on the climate. Water is therefore fluid, non-linear, constantly changing its appearance, and generally un-Pythago rean.

  CLIMATE ATTRACTOR

  Because of the importance of water in the global climate system, model predictions depend on exactly how its behaviour is represented. GCMs generally agree that as global warming continues, more moisture will evaporate and form clouds, therefore affecting the planet’s albedo; however, the exact change in albedo depends critically on the type of clouds.51 Climate modellers may in the future be able to hammer out a consensus on these and other difficult points, but there is no guarantee that the consensus will be right, because no model can capture the full scope of the climate system.

  A typical GCM used for climate studies might divide the atmosphere into a three-dimensional grid with cells about the size of a small country like Belgium. For a particular property, such as albedo or cloudiness, they assign a single number or simple distribution to everything within a cell. Like fractals, though, natural surfaces reveal greater amounts of detail the more you zoom in. There is never a point where more resolution doesn’t reveal a finer degree of structure.52 Since GCMs can offer only rough param-eterizations of the physics of clouds—or the growth of trees, or changes in grasslands, or the deformation of ice under pressure, or the exact way water evaporates from exposed ground, or a whole host of other things—they are not true physics-based models. They cannot be demonstrated from fundamental laws. Like models of the economy, they are really a collection of approximate equations that have been combined and balanced to give reasonable results.

  Many climate models, if run forward for thousands of years, would not give a realistic-looking climate at all without the use of so-called flux adjustments. Left to their own devices, models will cheerfully boil away all the water in the oceans or cover the world in ice, even with pre-industrial levels of CO2. (The fact that the earth doesn’t do this says something interesting about its regulatory networks, as we’ll see in the next chapter.) The only way to bring them back to something reasonable is to somehow fudge the heat balance—altering, for example, the transfer of heat between ocean and atmosphere.53

  The models are also strongly affected by changes in parameters. In one recent Oxford University–led experiment, the largest of its kind, several key parameters controlling the representation of clouds and precipitation in a GCM were set to alternative values “considered plausible by experts.” The effect was to explode the range of predictions, pushing it as high as 11.5°C (after omitting some simulations that became unstable, or even showed cooling).54 The climate system consists of an intricate balance of opposing feedback loops—what Heraclitus described as a “harmony of opposite tensions”—and small changes in their representation can have large effects. This sensitivity does not imply that the climate system itself is unstable. All that can be concluded is that the models are sensitive to parameterization.55 The average rise was close to 3.4°C, but this only reflects the fact that perturbations were made around a model that happened to show that amount of warming.

  Even such experiments do not reveal the true uncertainty in the calculation, because the very structure of the GCM equations fails to capture the underlying dynamics. Parameters are properties or constructs of the model, not the system. As one paper put it, “As soon as we begin to consider structural uncertainty, or uncertainty in parameters for which no prior distribution is available . . . tidy formalism breaks down. Unfortunately, the most important sources of model error in weather and climate

  forecasting are of precisely this pathological nature.”56 Since the errors are “pathological” (i.e., not expressible by equations), it is as difficult to estimate the correct parameter range, or forecast uncertainty, as it is to predict the climate itself, and for exactly the same reasons. We don’t have the equations. In an uncomputable system, they don’t exist.

  As the pioneer climate scientist Manabe put it, “Uncertainty keeps increasing with the more research money they put in. . . . It hasn’t gotten any better than when I started forty years ago.”57 Perhaps this is why, despite huge advances in computer speed, earth-observation techniques, and climate research, the IPCC and most climate scientists still quote the canonical band for carbon doubling of 1.5°C to 4.5°C.58 This range gives an image of stability, and it helps, as the Oxford scientist Steve Rayner described it, to “domesticate climate change as a seemingly manageable problem for both science and policy.
” The estimates for global warming seem to say as much about the dynamics of science as they do about the dynamics of climate.59

  REASONS TO BE VALID

  Below are some common arguments for the validity of climate models—and the reasons why they are not valid.

  “The models are derived from basic laws of physics.” There are no laws for the formation and dissipation of clouds or many other processes.

  “The models can reproduce the current climate.” Versions with different parameters can adequately reproduce the current or recent climate, while giving a very different response to doubled carbon dioxide. It is always possible to tune models to fit past data; it’s much harder to predict the future.

  “The models can simulate past climates.” Same problem. Also, there is a great deal of uncertainty in estimating climates in the distant past.

  “The butterfly effect does not apply to long-term climate forecasts.” This isn’t very relevant, since chaos is rarely the most important factor in the modelling of physical systems.