Control Earth

To do an experiment — any experiment — you need a control. A baseline state, a normal, an unmanipulated variable. A what-things-would-be-like-if-I-weren’t-changing-them kind of thing. In other words, a natural thing. Then you do your experiment and see what happens, ceteris paribus (i.e., all other things being equal). You change a condition or two, you manipulate, you alter, and you see how your treatment cases differ from that baseline state. You determine what the interventions do and how they matter. Whether they create something better, or worse, or just different from your control. Different from that natural thing.

How is the climate different from what it would have been had we not dug up trillions of tons of ancient plants and animals and sent them up in smoke?

We are changing Earth’s climate, in what the oceanographer Roger Revelle famously called humanity’s “great geophysical experiment.” ¹ But if it’s an experiment, what’s the control? Changing from what? How is the climate we’re creating different from what it would have been had we not dug up trillions of tons of ancient plants and animals, compressed over millions of years into soft black rock and energy-rich black goo, and sent them up in smoke, all within a couple of centuries? What would have happened instead?

To know these things, we need a control Earth. A ceteris paribus Earth. An Earth that would have existed had we not shown up with our hypertrophied brains, our energivorous technology, and our insatiable appetites for more and more and more. There are multiple ways to build such control Earths; the one that we’ll look at here is, naturally (so to speak), climate simulation.

Zoom: Climate Simulations as Interscalar Vehicles

A useful adage from the weather business says that climate is what we expect, whereas weather is what we get. Climate expectations are about space and time. Climate is seasons, four or two or five, depending on where you live. It’s cycles of planting and harvest. It’s clothing and comfort expectations tied to geography: when you’ll wear short sleeves and sweat, where you’ll have to bundle up in a goose-down parka and snow boots. Many terms for regions also describe climates, and vice versa: a desert, a rainforest, the tropics, the Arctic. Climates have characteristic ecosystems and iconic flora and fauna: palm trees, polar bears, camels. To a degree — pun intended — climates shape patterns of daily life: siestas on hot afternoons, huddling around fireplaces in cold winters, how and sometimes whether you can get around in rain, mud, snow, or jungle heat. For architects, climates affect buildings’ structures and systems: heating and cooling, the potential for natural light, roof shapes and overhangs, load-bearing capacity, foundations, entryways, appropriate materials.

To know these things, we need a control Earth … one that would have existed had we not shown up with our hypertrophied brains and energivorous technology.

For climate control Earths, scale is the ultimate problem. In human terms, climate means patterns; stable, more or less predictable cycles; baseline states you can count on. If you live in a temperate zone, you might get some cool days next summer, but you’ll be really surprised if it snows. You may travel in search of a particular climate — beach weather, dry and mosquito-free desert camping, hot days and cool evenings in the south of France — or to escape one you don’t like, just as Canadian snowbirds flock to Florida each winter. In human terms, these patterns are places. They’re where you live, or where you spend certain kinds of time. On the scales of ordinary human life, there’s no such thing as a global climate. Rather, there are multiple climates: local and regional patterns and differences.

Yet from a God’s-eye perspective (or that of a scientist) all those patterns are connected in one gigantic system, driven by colossal forces: gravity, the sun, orbital variations, axial tilt. They are shaped by atmospheric chemistry that’s evolved over eons and has been altered dramatically by living things. As James Lovelock once put it, the atmosphere is the circulatory system of the biosphere. ² Every year, deciduous plants suck up immense quantities of carbon in the spring when they grow leaves, and spew it out again in the fall as their fallen leaves decay, oozing carbon back into the atmosphere. Now we’re adding more carbon — a lot more.

Above all, climate is driven by sun and water. Oceans absorb and release heat and carbon dioxide. Their currents, like great conveyor belts, transport heat and salt around the world. Water rises, evaporating from lakes, rivers, soil, and vegetation, only to fall again somewhere else as rain or snow. Clouds reflect solar heat back into space, trap warm air near the surface, or both. Ancient snows lie locked in mountain glaciers and in the vast ice sheets of Greenland and Antarctica. The huge fields of sea ice in the polar oceans are now shrinking, leaving polar bears, walruses, and other creatures with nowhere to rest.

Thus the spatial scales of climate physics run the gamut from the molecular (cloud nuclei, radiation-absorbing gases) to the solar system. Time scales matter just as much. Human time is measured in days, weeks, months, and years. But climate time demands perspectives of 30 years or more: decades, centuries, millennia. It’s time out of mind, beyond experience, beyond human history. It’s ice ages and interglacials, governed largely by solar output, orbital cycles, and the tilt of Earth’s axis, with carbon dioxide as a catalyst. On the really long time scales, the positions of continents matter a lot.

Climate models make good control Earths because they can travel across spatial and time scales. They’re interscalar vehicles.

Climate models make good control Earths because they can travel across some (though not all) of these scales. They’re “interscalar vehicles.” ³ These models are giant bundles of mathematics, expressed as computer code, that try to capture the major forces responsible for climate and examine them in interaction. The most comprehensive ones, called Earth system models, simulate the global circulation of the atmosphere and oceans: snow and ice, with their radiative effects; clouds; land surface characteristics, such as mountains and albedo (reflectance); aerosols, both natural and anthropogenic; and the carbon cycle, including anthropogenic greenhouse gases. Carving up the atmosphere and oceans into grid cells, these simulations calculate transfers of mass, energy, and momentum from one cell to all its neighbors, on a time step of 10 minutes or so, over tens to hundreds of simulated years. Climate models are very similar to weather forecasting models, but their grid cells are bigger because a climate simulation must be run for at least several (simulated) decades. Some runs model time periods of 1,000 years or more. All this requires tremendous computer power.

Carving up the atmosphere and oceans into grid cells, these simulations calculate transfers of mass, energy, and momentum from one cell to all its neighbors.

Climate models can be operated like zoom lenses. For a spatial zoom, they can bring a particular region into sharper focus by embedding higher-resolution regional models into the main model. Or their results can be downscaled using statistical methods (though this technique remains problematic and controversial). For a temporal zoom, their parameters can be set to resemble those of some previous time period. You can recreate the Pangaea supercontinent of 200 million years ago, or conditions at the peak of the last ice age, or any time you like, including the future. That’s where those scary curves come from, some of them showing temperatures soaring by 5°C or 6°C by the end of this century. What happens if we keep on adding carbon? What happens if we stop?

Versions of the Past: historical vs. historicalNat

Consider the Coupled Model Intercomparison Project, now a crucial element of the modeling work that goes into the periodic reports of the Intergovernmental Panel on Climate Change. CMIP helps modelers to evaluate the strengths and weaknesses of their models. It organizes a set of standard experiments in order to make apples-to-apples comparisons of model results. A “standard experiment” means a model run, or an ensemble (i.e., a related set) of runs, using predetermined parameter values for such variables as greenhouse gases, sulfate aerosols, and solar irradiance. Experiments that project future climate change get the most press, but simulations of the past provide a better basis for model comparison, since these simulations can be compared with actual observational data. As I argued in A Vast Machine, CMIP’s experiments play a critical role in the climate knowledge infrastructure. ⁴

Various climate model intercomparison projects have been running since 1989. CMIP5 provided input to the IPCC’s Fifth Assessment Report, released in 2014. Here are some CMIP5 experiments:

• piControl (preindustrial control run): atmospheric composition and land cover fixed at values for the year 1850; it does not include volcanic eruptions
• historical (also known as 20th century): 1850–2005, including all observed changes (both natural and anthropogenic) to atmospheric composition (greenhouse gases, volcanic eruptions, and aerosols), as well as time-evolving land cover
• historicalNat: 1850–2005, but including only natural changes and events such as major volcanic eruptions, introduced in the years they occurred
• past1000: time-evolving conditions over the last 1,000 years, including solar variations and volcanic aerosols
• midHolocene: orbital parameters and greenhouse gases set to their states halfway through the current interglacial period (the Holocene Epoch), i.e., 6,000 years ago
• lgm (last glacial maximum): orbital parameters, solar output, ice sheets, and greenhouse gases set to their conditions at the height of the most recent ice age, 21,000 years ago ⁵

All climate modeling groups wishing to contribute to the Fifth Assessment Report were required to submit a core set of experiments that included historical and piControl. The historical run was intended to determine the models’ skill in reproducing climate trends since 1850, the period for which reasonably reliable instrument observations are available. Since climate exhibits natural as well as forced variability, the goal of historical is not to reproduce exactly the year-by-year changes that occurred — indeed, if a model did so, it would immediately be rejected as illegitimate — but to capture the overall trend. Models should, however, reflect the abrupt, short-term global cooling induced by large volcanic eruptions. Meanwhile, those events should not appear in piControl runs, which are designed to produce an idealized steady-state baseline — an Earth frozen in time, without volcanic eruptions, changes in solar output, or other natural causes of climate change. The other runs described above belonged to an optional set of “Tier 1” experiments. midHolocene and lgm were intended to show that models can reproduce global climate from 6,000 and 21,000 years ago, respectively, while past1000 covered the period since 1000 AD using time-evolving parameters.

These temporal zoom-ins help to build confidence in climate models, as well as to diagnose their biases and areas of difficulty. They also exhibit different relationships to the real Earth — or rather to our knowledge about the real Earth, which has important limits.

Experiments that project future climate change get the most press, but simulations of the past provide a better basis for model comparison.

historical runs have their own, familiar kind of control Earth, namely the historical instrument records preserved from ships, weather stations, and satellites. Nothing gets closer to what ‘really’ happened than that. Yet as I showed in A Vast Machine, those data are themselves produced by certain kinds of models. The instrument record consists of many types of information, obtained by instruments with widely varying characteristics, recorded by diverse individuals and institutions under different regimes of standards and political arrangements. To assemble anything resembling a coherent global data set, you have to reconcile these differences, a process that involves considerable investigation and adjustment. I call this “making data global.” So even though the instrument record will always be our most accurate and trustworthy control Earth, it is always already the result of a certain kind of modeling. This modeling accounts for the small but noticeable differences among the historical climate datasets produced by institutions such as the US National Climatic Data Center, the Climatic Research Unit of the University of East Anglia, and the European Centre for Medium Range Weather Forecasts.

What about past1000, midHolocene, and lgm? Here the question of the control Earth takes on a new form. On these time scales, almost all data about the real Earth’s history come not from instrumental measurements but from proxies such as tree rings, ice cores, pollens, corals — things that vary with temperature or other climatic phenomena. Tiny bubbles of prehistoric air, trapped in the glaciers of Greenland and Antarctica, provide one of the few sources of directly measurable data. Moreover, for most proxies, the data come from a relatively small number of sources, unevenly distributed across the globe, and mostly on land. As demonstrated by the large differences among different proxy reconstructions, even the best data still display considerable uncertainty. When placed beside them, past1000, midHolocene, or lgm runs start to look more like control Earths in their own right.

But it is historicalNat that interests me most. These runs represent the Earth without us, or at least without the exploding populations and epic fossil fuel consumption of the 20th century. It’s Earth as it might have been in the times of our great-great-grandparents, time still within historical memory, before the massive ramp-up of technology-intensive societies we built with coal, oil, and natural gas (not to mention our 1.3 billion methane-belching cattle).

historicalNat plays a significant role in “attribution,” scientists’ term for relating the effects of climate change to particular causes. After all, many natural factors such as solar output and volcanic eruptions can change (or “force”) the climate. historicalNat helps to disentangle these natural contributions from those of human beings by giving us control Earths against which to contrast the results of our “great geophysical experiment.”

historicalNat creates a shimmering, slightly out-of-focus vision, an ensemble of could-have-been-Earths rather than a single definitive one.

Note the plural. historicalNat provides not a single control Earth but many related ones. It creates a shimmering, slightly out-of-focus vision, an ensemble of could-have-been-Earths rather than a single definitive one. The same goes for the historical runs: they are not exactly controls, yet still are climate histories-that-might-have-been. In climate science, the question is never “Which is the real control Earth?” but rather “What is the plausible range of climate on these control Earths?” If human activities don’t affect the climate much, we would expect historicalNat to generate at least a few control Earths that look a lot like the historical instrument record.

But … we don’t. The graphs below show CMIP3 (a previous version) along with CMIP5. At both the left and the right, the yellow and gray lines show many runs from those experiments; the thick gray and red lines are the averages of the ensemble of runs, while the black line indicates the observed global temperature trend. The top graph (a) shows historical runs, while the bottom graph (b) shows historicalNat. The inset displays three of the major global temperature records, which agree closely but not exactly, as discussed above.

Graph (a) shows that although individual runs bounce around a lot (like the real climate), the ensemble of historical runs matches up pretty well with the observational record. In other words, as a group the climate models do pretty well at reproducing what actually happened. In general, they capture the short-term cooling caused by major volcanic eruptions and match the long-term warming trend. They’re plausible simulations of history.

Until around 1960, what happened on the real Earth could also have happened on a lot of control Earths. But then the models begin to diverge.

In graph (b), things get really interesting. Here we’re looking at control Earths without human greenhouse gases and aerosols. This graph tells us that until around 1960, what happened on the real Earth could also have happened on a lot of control Earth(s). But starting around 1960 — exactly when fossil fuel consumption began to really spike, rising from less than two gigatons of carbon per year in 1960 to nearly 10 gigatons a year in 2010 — the real Earth and the control Earths begin to diverge. From 1960 to 1980, a few control Earths still just barely match the temperature increase that occurred on the real Earth. After 1980, none of the control Earths even come close to reproducing the warming trend that Earth has actually experienced. In other words, without the human contributions — and not only the warming caused by anthropogenic greenhouse gases, but also the cooling caused by anthropogenic aerosols — the control Earths would not have warmed.

There are other simulated control Earths. Lots of them, in fact. In CMIP5, for example, there’s also historicalGHG, a control Earth including anthropogenic greenhouse gases, but without anthropogenic aerosols. historicalGHG control Earths warm more than the real Earth has, thus showing how aerosols’ cooling effect masks the underlying greenhouse trend. Among other things, simulated control Earths generate predictions of things to look for on the real Earth. Scientists call these the “fingerprints” of particular causes: if the sun is the cause, for example, days and nights should warm equally (they don’t). ⁶

Conclusion

Control Earths matter to scientists because there’s no other way to tell what would have happened without us, or to disentangle the contributions of various natural and human factors. The control Earths that matter to the rest of us are the ones that involve the future. They’re the “representative concentration pathways” and emissions scenarios of the IPCC reports. These are worlds that might be, destinies that we might endure or celebrate, depending on the choices we make — or fail to make — from now on. Whatever path we actually choose will be our “grand geophysical experiment.” Simulations give us one way, perhaps the only way, to know how much we have mattered and will matter to the planet’s fate.