The weather turned upside-down?

Abrupt Climate Change: evidence, mechanisms and implications

A report for the Royal Society
and the Association of British Science Writers
by Mike Holderness, March 2003

At lunchtime on 4 February 2003, bemused tourists climbing the Duke of York's Steps in central London found over 300 climate scientists huddling in the snow. The fact that it was snowing was more a topic of conversation than the false alarm that had driven them out of the Royal Society - for it hasn't snowed much in London in the past decade. But over the past century the normal London climate in February has included snow. And, for those scientists who research ancient climates and who think of the past million years as "recent", the normal London climate is at least as cold as the Arctic is now, with brief warm spells every 100,000 years or so.

That scene, then, sums up several of the questions that drew a record attendance for a Royal Society meeting. There is no doubt that the world's climate is changing. There is a strong consensus that this is connected with the "greenhouse effect" through which gases in the atmosphere - notably carbon dioxide - trap the energy from the sun, and that human emissions of these gases are important.

But to answer important questions like how climate changes will differ from place to place and how severe they will be requires detailed knowledge of what "normal" climate has been. The recent realisation that climate may change abruptly - and that global warming could mean a sharp cooling of North-West Europe - adds urgency.

The effort to discover how likely this is depends on drawing together everything we know about the planet, and filling many gaps in our knowledge. It involves understanding myriad feedback loops: how for example people, plants, soils, ice sheets and oceans respond to, and in turn affect, changing conditions. So it brings together scientists of many kinds, from physicists to geologists to botanists. They have had to learn very quickly how to talk to each other - and to politicians and the public, because what they are dealing with is no less than the fate of life on Earth in the near future.

What is ‘abrupt climate change’?

Until very recently scientists have discussed climate change almost entirely as a gradual process in their terms - which means an extremely slow process in political terms. There has been intense debate, for example, over what may happen by the end of this century.

But since the mid-1990s scientists have been asking whether the world's climate might change abruptly, as well as gradually.

Professor Jochem Marotzke, one of the meeting's organisers, opened it by asking what we mean by "abrupt". He worked with Professor Richard Alley on the 2002 US National Academy of Science study Abrupt Climate Change: Inevitable Surprises. After much thought they defined what they were talking about - from the point of view of the climate - as:

• the climate system "flips": it crosses some threshold, for example, from a state with a Gulf Stream like today's to one where it stops short further south; and
• after is has crossed the threshold, further change happens at a speed governed by the climate system itself, not the slow change in whatever pushed the system.

From the point of view of ecosystems, the more important features of these abrupt changes are that:

• once the threshold is crossed, the changes persist;
• the changes are on continental or global scales; and
• because they are abrupt, these are changes for which ecosystems are unprepared or to which they are incapable of adapting.

Marotzke summed up the purpose of the meeting as summarising what scientists know about abrupt climate change. And, "always more interesting for scientists", what we do not know. Listing areas of ignorance is essential to deciding what research to do next.

What has happened? Why did it happen? What might happen in the near future? What will the consequences be?

Gradual climate change poses enough of a challenge. The resulting rise in sea level could inundate entire countries, and staple food crops may no longer thrive in the areas where they are currently grown, for example. Given the will, it may be possible both to minimise the changes - by reducing carbon dioxide emissions - and to plan to reduce the impact of the changes that do happen - by working out what crops will grow, for example.

Abrupt climate change clearly makes such planning difficult. The possibilities include the UK climate switching within a couple of decades - or even a few years - to be more like that of Iceland. Other parts of the world would undergo different changes: the rains could fail across South Asia, for example. And with our present knowledge, it is extremely difficult to provide any early warning, or even to be sure whether one extremely cold winter or one drought is a random fluctuation in the weather, or a sign of a permanent change in the pattern of weather which is "climate".

Abrupt Climate Change has happened already

Professor Richard Alley is a palæoclimatologist - that is, he studies ancient climate - at Pennsylvania State University. Such work is essential to predicting the future.

He started by showing "the picture that most people and most policy-makers have of climate change" - the Intergovernmental Panel on Climate Change (IPCC) graph of predicted temperature changes for different scenarios. "There are uncertainties, depending both on what the world does and on what people do," he noted: the change in global temperature this century may be 1°C or 6°C, as shown in the standard IPCC graphic:

A gradual one-degree increase in global average temperature would have noticeable effects, for example changing the growing seasons for crops and animal fodder in given places, particularly marginal places. Professor Alley reminded the audience that the records of past temperatures that he and colleagues have extracted from Greenland ice cores show "a little blip" in temperature around 1350 that coincides with the extinction of the Viking colonies in Greenland. From recent archaeological work it seems that "they brought their animals into the house for the winter, ate their dogs and disappeared."

This was a change of one or two degrees. A three-degree change - in the middle of the range of uncertainty - would have quite significant effects.

But, Professor Alley asked: "is the IPCC being over-optimistic?"

To answer the question, we have to digress from the state-of-the-art presentations at the meeting, to go back and look at the basics of how scientists make such predictions.

About modelling: consider a cubic Earth...

Climate change predictions come from multiple runs of several super-computer models, each of which aims to simulate the entire planet's climate system.

Briefly, the models divide the atmosphere into cells that are roughly cubes; calculate the physics of what goes on inside one cell in an interval of 20 minutes; communicate flows of air, heat and moisture to neighbouring cells; and repeat over all the cells, over a time period of several thousand years. Cells at the outer edge of the modelled atmosphere are "fed" with solar radiation, and those at the surface with heat, moisture and other gases from virtual land and oceans.

A complete Earth model would do the same for the ocean, rather than treating it as an array of point sources of heat and gases. There are ocean models that model currents in the same way that atmospheric models do winds, and work on integrating them with the atmospheric models is going on. Modelling what goes on on the land surface, from soil bacteria through plants to flatulent cows and car manufacturers and stock exchanges, is quite another matter.

The smaller the cells and the shorter the time-steps, the better the model can be. These "granularities" are determined by the amount of computer power available. Current models are too "coarse" to include, for example, the uplands of Wales as a single feature, let alone mid-afternoon weather effects in the valleys.

If you want to model the climate over hundreds of thousands or millions of years, you have to use a still coarser granularity. For the foreseeable future, climate scientists will be begging for more computer power.

This note on scientific modelling from a report of a Royal Society meeting on high-performance computing may be helpful.

Testing the models...

One important way to test a climate model is to see how well it "predicts" the past. Set it off with reasonable starting conditions for 800,000 years ago, for example, and see whether it "predicts" the nine major ice ages that we actually find in the geological record without being told about them.

To do this checking, we need records of temperatures as far back as we can get them. The longest series of daily weather reports in the world covers central England back to 1772; general Northern Hemisphere coverage was achieved in the 19^th century. Tree rings take us back a few thousand years. A lot of Professor Alley's work has been on ice cores, which take us back many hundreds of thousands of years.

Ice is H₂O; the great majority of the oxygen is the "isotope" oxygen-16, abbreviated ¹⁶O. A small proportion is the isotope ¹⁸O. Water containing ¹⁸O condenses from the air at a slightly higher temperature than that containing ¹⁶O; so snow falling from colder air contains less ¹⁸O, and the ¹⁸O:¹⁶O ratio provides a record of temperature. Similar methods can be used with other isotope ratios, such as deuterium to plain hydrogen (²H:¹H). The ice also contains trapped air bubbles, which include gases of interest such as carbon dioxide and methane, and calcium levels indicate atmospheric dustiness.

The age of ice can be determined by counting annual deposition layers back to about 40,000 years ago. Before that a variety of methods, including physical modelling of the layers squashing under pressure and radioactive decay dating, are used.

So drilling through an ice-cap, making thin slices and measuring the ratio of ¹⁶O to ¹⁸O in each gives an indirect but accurate record of temperature. A core from "Dome C" in Antarctica, for example, is expected to produce a record stretching back 800,000 years.

The big picture of climate

The most obvious feature that researchers find in the temperature record is that for most of the past 800,000 years the Earth has been much colder than it is now, with warm spells every 100,000 years or so. For example:

Closer examination shows less extreme periodicities of 41,000 years, 23,000 and 19,000 years. These correlate with changes in the Earth's orbit: it changes from more circular to more eccentric every 100,000 years, the tilt of the line through the Poles relative to the plane of orbit changes over 41,000 years, and the direction of that line wobbles with the two shorter periods. They are known as the "Milankovitch Cycles", after the Serbian astrophysicist who thoroughly described them in 1920.

These changes mostly alter the distribution of sunshine between the Northern and Southern hemispheres and between the Poles and the equator, rather than the total amount of energy. As a questioner pointed out, the mechanism by which they affect climate is far from clear.

But it is clear that when Greenland has least sun, it's coldest - but Antarctica is also coldest at the same times, when it has most sun. So the global climate follows the amount of sunshine in the north. And though the orbit changes are gradual, the record shows some very rapid temperature changes coming in and out of ice ages. At the very end of the Younger Dryas, a global cooling event of particular interest between 13,000 and 11,500 years ago, average temperatures increased by about 6°C within a decade in some places. (The event is named after the increase in pollen and fossils of Arctic dryas plant species found in sediments.)

The real world looks more sensitive than the models

To return to Alley's presentation: what does correlate with temperature everywhere is the amount of carbon dioxide in the atmosphere. This, as he says, "is one of the reasons we believe human effects on the atmosphere are going to show up in the climate".

So how well do the climate models that underlie the IPCC predictions do at "predicting" this effect? They're much better than people often give them credit for, Alley says. Climate changes show up in the right places at the right times But "if there's an error it's that they don't produce such large changes [in climate] as the real world does."

Clear patterns emerge in other data from the ice cores. When it's cold and dry in Greenland, it's cold and windy in Central Asia, as shown by hundred-fold increases in dust in the ice, and also in wind-blown sea salt. Methane levels indicate the amount of wetland globally: when Greenland is cold it dries up. Independent records from sediments in Venezuela show the same effect.

The temperature records correlate very well across the globe: the big, gradual changes affect the entire planet. But there's one exception: when Greenland is getting cold very quickly, the South Atlantic (north of the circum-Antarctic current) warms.

At the end of each ice age, we see a warming followed by a slip back into deep freeze. Alley described what happened toward the end of the Younger Dryas: as the climate started to warm up again, a huge quantity of melt-water from North America "poured into the North Atlantic in a big hurry - and it got cold". Average temperatures fell by 6°C within a century.

The global ocean current

So how does dumping a lot of fresh water into the North Atlantic make much of the world a lot colder, but the South Atlantic warmer? The answer is the circulation of the water between the world's oceans.

This worldwide current is known as the "Thermohaline Circulation" (abbreviated "THC") because it is driven both by heat and by the saltiness of the water (Greek ’αλος = halos = salt). From the point of view of North-West Europe it's rather important that the THC transports heat from the South Atlantic to us.

When fresh water is dumped south of Greenland, sea-ice cover increases because it's easier to freeze. The models reproduce this effect rather well. They also successfully show other real world effects - though they underestimate them. These include a dramatic drying in "lots of places where a whole lot of people are living and depending on rainfall," as Alley put it: the grain belts of North America, the Saharan region and parts of the South Asian monsoon.

Alley concludes that abrupt climate change is real: it's happened several times in the past, and major changes have happened within a decade or as little as one year. It does not only happen within ice ages. A rapid cooling event about 8200 years ago started from a climate a little warmer than we've had recently.

His big question is: what are policy-makers seeing? Mostly the graph above - in which the scenarios are all shown as smooth lines. None represent the possibility of abrupt climate change.

The most important uncertainty within each scenario, to summarise, is not whether the global average temperature will rise smoothly by 2.9°C or 3.8°C - but what increase in carbon dioxide and other greenhouse gases will be enough to tip the climate from its present state into one where North-West Europe is suddenly many degrees colder and other places are significantly warmer. The mid-range CO₂ forecast levels could be enough to do this. We don't know.

"If there's one thing we're absolutely sure of," Alley observes, "it's that the changes won't look like that. Nothing is ever smooth."

Perhaps we could try something like the following to show what the models are actually predicting:

So is the IPCC over-optimistic? Alley hopes not. But the historical changes - both the slow ones and the fast ones - have been been bigger than what the models are calculating.

Would the 10 per cent reduction in CO₂ emission foreseen in the Kyoto process make any difference? Possibly: some models suggest that adding CO₂ to the atmosphere more slowly may avoid triggering abrupt change. But the system is chaotic...

Atmospheric processes and observations

The focus of the rest of the meeting on climate change in the deep past - and hence in the slow processes of the oceans and ice - led Professor Brian Hoskins FRS to joke with participants that he would try to convince them that the atmosphere was relevant to climate. Hoskins has worked on many aspects of climate modelling in the Department of Meteorology at the University of Reading.

Everything that happens in the atmosphere is "instantaneous", compared to ocean currents or cycles of glaciation. This means that the atmosphere is important in communicating changes - in temperature, moisture content or whatever - from one part of the planet to others on a timescale of weeks. The major force driving the weather is the transport of heat from the tropics toward the poles. Near the equator, ocean currents play a major part in this; further away, the atmosphere is the major conveyer of heat.

The behaviour of the atmosphere is chaotic, in the strict mathematical sense. This means that while the physics of a small region of air over a short time may be relatively simple, the behaviour of the entire system is complex and impossible to predict in any detail for more than a few days ahead: a one-month weather forecast is simply not possible.

But we can make general, probabilistic forecasts - that is, of climate rather than weather. We can give a probability that the weather will be in the vicinity of an "attractor" - a region in the space of all possible weather.

Improbable things will, of course, happen from time to time. One hot summer does not global warming make. But a whole series gives cause to ask whether the entire system has been forced to move to a different attractor.

Consider for example the "North Atlantic Oscillation," which is still well described in the original observation by the missionary Hans Egede Saabye, based on his weather observations in 1770-1778:

"In Greenland, all winters are severe, yet they are not alike. The Danes have noticed that when the winter in Denmark was severe, as we perceive it, the winter in Greenland in its manner was mild, and conversely."

The size of these swings has been increasing for decades (see below). Whether this change is a coincidence of improbable events, or evidence that the climate has moved to a different attractor, is a subtle question.

Folk wisdom has it that the warmth of North-Western Europe compared to Labrador - which is at the same distance from the Equator on the other side of the Atlantic - is due to the "Gulf Stream" current, the THC. But Hoskins points out that the track taken by storm systems swinging across the Atlantic from the Caribbean is at least as important. If the NAO "flipped" to a different attractor and they usually went elsewhere, Scotland for example would be 9°C colder on average.

Though possible changes in the THC, for example, are also important, only work on modelling the atmosphere will allow predictions of the effects on populations and their food sources - which, after all, live in the atmosphere. And though politicians want to know, for example, whether last year's floods in Central Europe were part of systematic global climate change, only global modelling can address these regional questions. Hoskins' group has preliminary evidence from climate models suggesting that these floods were related to the failure of the Asian monsoon rains.

Recent changes in the North Atlantic

The North Atlantic, lying between Western Europe and North America, is both ideally placed between large concentrations of scientists and of direct interest to major funders. But it also happens to be the most interesting part of the world for abrupt climate change.

It is almost certain that changes in less-well-funded parts of the climate, for example the South Asian monsoon, affect the weather in North-West Europe. Some researchers are suggesting, too, that abrupt changes in the THC may affect the climate worldwide, including the monsoon.

Dr Robert Dickson is a hydrographer at the Centre for Environment, Fisheries and Aquaculture Science in Lowestoft, Suffolk and he is studying changes in the North Atlantic over the past 40 years. Within this period, all 10 of the warmest years since records began happened between 1990 and 2002. It also includes extremes in the "North Atlantic Oscillation,". The NAO has been increasing - the weather has been swinging East and West more strongly. In the 1990s it was more extreme than at any time in the past 600 years.

This adds up, Dickson says, to "one of the most unusual climatic episodes in the history of the North Atlantic". He expects that it has the potential to lead not only to a slowdown in the global THC ocean current that other speakers addressed, but to an acceleration in the turnover of water between the oceans and the atmosphere (an acceleration of the global water cycle).

So what are we observing? There are few measurements of the flux of freshwater from the Arctic: instrument arrays both side of Greenland are only now being installed. Measurements from the Labrador Sea do indicate a 20 per cent increase in the near-surface flux of freshwater from the subarctic since the 1960s. Coupled with a freshening of the overflow system from Nordic Seas, the effect has been that of adding an extra 6 m of freshwater to the water column of the Labrador Sea by 1992. But this hasn't produced an "Even Younger Dryas", because it's distributed through the entire depth of the ocean. If it had all been dumped on the surface, Dickson is fairly certain we'd have seen massive changes in the THC. Expenditure on complete measurements of flows in the North Atlantic is certainly justified.

The evidence for an acceleration of water turnover is an increase in saltiness of water further south, through to the equatorial South Atlantic, which can only be explained by increased evaporation. The Pacific and Indian Oceans also show increased salinity in the tropics and freshening at both Poleward margins between the 1950s and 1990s. It is "tantalising to speculate" that this "represents a trend in the climate system that could be introduced by anthropogenic sources," as the researchers who wrote up the effect in the Pacific carefully noted. Water vapour is itself a greenhouse gas, introducing the likelihood of positive feedback leading to rapid change.

The biggest science questions, Dickson said, lie in understanding the detail of the processes by which the ocean responds to climate change. Important clues will be obtained by observing whether the climate "signal" propagates southwards through the west Atlantic in years or in months; this will be tested during the NERC-RAPID programme. We are seeing abrupt change in the oceans; the question is what the effects will be, and where and when, in terms for example of rainfall and drought.

Return of the Varves

In order to test any model, as noted above, we need as much data on the history of the climate as we can get.

Professor Alan Kemp works on gathering evidence of past climate from sedimentary rocks, in the School of Ocean and Earth Science at Southampton Oceanography Centre. This field of research has a curious history. It was founded in 1912 by Gerhard De Geer, but then fell deeply out of fashion - before being resuscitated in the 1970s. New technological developments in taking cores of soft mud from the sea floor have brought this archive of past climate change to the forefront of research.

De Geer, a Swedish geologist, naturally peered into the extensive excavations for the expansion of Stockholm at the turn of the last century. He saw clear layers in the clay, centimetres thick. So he measured the thicknesses of the layers and plotted them as a graph to produce what Kemp called "some of the original wiggly lines in the climate change library". De Geer discovered that the layers, which he called "varves" matched from site to site, and concluded that they were annual layers deposited by meltwater from an ice sheet.

In 1912 de Geer published a "geochronology of the past 12,000 years". In 1921 he published a paper correlating the Swedish varves with laminated sediments in the USA. The scientific community was not convinced that individual varves could possibly correlate across the ocean. As late as 1960 another researcher in the field, Roger Anderson, had a funding application rejected with the reviewer's curt comment "I don't believe in varves."

Laminated sediments are not just the result of glacial melting: many now studied result from increased algal production in the summer resulting in organic matter settling to the lake or sea floor. When there's more oxygen in the bottom water, however, burrowing creatures disrupt the sediment, the matter decays and varves are not formed - itself a useful marker of wider changes such as ocean circulation or the total carbon turnover in the environment.

Drilled cores from Saanich Inlet, a fjord in British Columbia, give a high-precision 6000-year record: 2100 years ago their thickness abruptly doubles to 1 centimetre. That indicates a change from a cold and dry to a warm and wet climate.

Cores from the low-oxygen waters of the Santa Barbara basin off California show another mechanism of varve formation: the remains of individual massive "blooms" of algae. During El Niño events these are smaller and the layers of silt and clay between them are thicker due to storms and greater rainfall causing floods and runoff from land.

Researchers have also done carbon-14 dating on the microscopic skeletons of surface- and bottom-dwelling foraminifera plankton in the Santa Barbara sediments. This shows that at the times when the layers were better preserved the low-oxygen bottom water was older - that is, it came from deeper in the Pacific and had been away from surface sources of fresh ¹⁴C for longer. These changes correlate with changes in North Atlantic ocean circulation, further demonstrating the global nature of these climate system changes.

Further samples from the Cariaco basin off Venezuela show that the Younger Dryas cooling episodes in the North Atlantic correlate "practically instantly" with increased plankton productivity. This is more evidence for the abruptness of these changes and for them having occurred simultaneously around the planet. The same events show up in cores off the coast of Pakistan, as a weakening of the monsoon winds.

Varves produced from annual algal blooms have also been found in the deep Pacific, promising records from up to five million years ago that can distinguish individual years' climate.

Abrupt climate change in the distant past - the methane connection

To get a general understanding of the climate, it is also important to look even further back than the few thousand years covered by varves and few hundred thousand by ice cores.

Dr Hugh Jenkyns, a geologist from the Department of Earth Sciences at the University of Oxford, started by describing the "Palæocene-Eocene Thermal Maximum" (PETM), a warm period some 55 million years ago. Records of this in sedimentary rocks from the Maud Rise near Antarctica show a 5°-6° rise in global average temperature over a few thousand years - which is abrupt change for geologists like Jenkyns.

This rise in temperature went hand in hand with a major disturbance of the "carbon cycle" - the exchange of the element between rocks, water, air and life-forms. The evidence for this is that the sediments from warmer periods have lower levels of the carbon isotope ¹³C.

His explanation invokes a mechanism for climate change that has only been widely studied in the past few years: sudden releases of methane from the ocean floor. Large amounts of methane are known to be locked up there in methane hydrates, in which each molecule of methane (CH₄) is bound up in a "cage" of water molecules. This substance is not stable except at the enormous pressures of the deeps and at low temperatures.

The carbon in the methane hydrates has a much smaller proportion of the rare stable isotope ¹³C mixed in with the common form ¹²C than does the carbon now found at the Earth's surface (in which about 1 per cent is ¹³C). The difference is due to differential removal of the two isotopes by bacteria over millions of years. So the best explanation of the ¹³C levels in the sediments is the release of methane from hydrates - which would explain the rapid global warming, for methane is a greenhouse gas in its own right, and is oxidised very rapidly to CO₂.

Small increases in water temperature - or perhaps earthquakes acting below the seafloor and disrupting the sediments - cause the methane hydrates to break down into gas which bubbles up to the surface to cause global warming, releasing yet more methane and producing rapid climate change (in geologists' terms).

Sediments found beneath the Antarctic Ocean also tell us that at the time of the PETM there was an increase in weathering of the continental crust. For example, kaolininte is found in this region - a tropical weathering mineral currently common in the Niger delta.

In the Early Jurassic period, researchers similarly find evidence for methane release together with tropical pollen in sediments from Siberia. Fossils of belemnites, similar to modern squid, provide a source of shell material for further study; the ¹⁸O:¹⁶O ratio in them confirms another warming 180 million years ago.

This episode also shows evidence of increased weathering. And it coincides with the depositition of black shale rocks, suggesting a burgeoning of planktonic life in the oceans.

Similar shales suggest further warming episodes, for example 120 million years ago in the Early Cretaceous period. This event also correlates with low ¹³C in fossil wood and in sediments, suggesting another methane release, and ¹⁸O:¹⁶O ratios showing raised temperatures. In fact, now people are looking for potential methane release events "an embarrassing number" are showing up. There is also preliminary evidence of cooling events in the distant past.

The ice amplifier - the key to modelling?

Climate modelling began, naturally, enough, with the atmosphere and with the oceans. But, as noted above, meltwater has played a large part in abrupt climate change in the past. So it has recently become clear that modelling the detailed behaviour of ice is an essential part of global climate modelling - and this is what Dr Andrey Ganopolski works on at the Potsdam Institute for Climate Impact Research in Germany.

The thing about the past is that it is different from now: we must not feed our knowledge of current phenomena like the thermohaline circulations into models, but must model past climates from first principles "if we want to get it not only right, but for the right reason," as Ganopolski puts it.

Between ice ages, for example, the record shows "Dansgaard-Oeschger oscillations", involving temperature changes of up to half the difference between ice ages and warm periods, over a decade or less. Atmospheric models that do not include details of ice cover on the sea and land cannot reproduce these.

But, Ganopolski points out, they can be modelled as "hysteresis" effects - see graphic.

One example is the probability that the THC is bistable. But tipping it from one state into another would take a massive amount of fresh water - the output of one or two Amazon rivers. Ganopolski proposes that sea ice is the missing ingredient. In the current climate state, the Gulf Stream reaches into the far North Atlantic. In a cold-period state, it does not turn off entirely but stops short around 50° North - the latitude of the tip of Cornwall. It would take only a much smaller quantity of fresh water to tip the ocean circulation from one state to the other.

This mechanism produces a possible answer to the unanswered question, mentioned above, of how regular changes in the Earth's orbit might lead to periodic ice ages. Random climate fluctuations (weather, in other words) superimposed on these small changes in solar energy distribution could produce "spikes" of fresh water that tipped the ice state, that in turn encouraged the ocean current state to tip.

Electrical engineers have long been familiar with noise strengthening the effect of a weak periodic signal: they call the phenomenon "stochastic resonance", where "stochastic" is essentially a polite term for "random".

Ganopolski and his colleagues in Potsdam has also recently succeeded in simulating another effect that probably produces regularity in climate oscillations: "slip-stick" motion of ice sheets over land.

Sea ice as the climate switch

The 100,000 year cycles of ice ages described above only go back 800,000 years. During this time, ice cover on land repeatedly grew for 90,000 years and then receded for 5000 to 10,000 years. Before that, the records show a 41,000-year cycle. What happened? Professor Eli Tziperman models sea ice at the Weizmann Institute in Rehovot, Israel and described his attempts, with Dr Hezi Gildor, to explain this.

No single theory yet deals with all the questions. Their proposed mechanism is based on two major feedbacks:

• the increased reflectivity or "albedo" of ice reflects solar energy back into space and leads to further cooling; and
• as temperature increases the evaporation of water from the oceans and the precipitation of snow over ice sheets is faster - leading to Tziperman predicting that ice cover should expand more in relatively warm periods; too much warming melts ice, of course.

It it not feasible to run general climate models over millions of years. So Gildor and Tziperman use a simplified "box model" of the coupled climate system in an abstract chunk of "planet" having an atmosphere, an ocean, some land - and ice cover on both ocean and land.

The mechanism based on Tziperman's model predicts that land-ice should have started growing in interglacial (warm) periods; some critical amount of ice on land lowers the global temperature enough for the sea to start freezing. This increases the planet's albedo, causes yet more cooling and more freezing - so sea ice cover in the North Atlantic expands as far south as Spain in a few decades.

In the deep cold that follows, the atmosphere and hence snowfall essentially dry up - but the ice sheets on land still evaporate slowly. So later, when the glaciers have retreated enough to allow temperatures to rise, sea-ice melts rapidly, heralding another warm interglacial period.

A novel - and controversial - feature of Tziperman's model, then, is that land ice grows when sea ice is at a minimum and vice versa. Other models have assumed that they vary together.

The model produces roughly 100,000 year ice cycles whether the effects of the 100,000-year Milankovitch variation in the Earth's orbit are included or ignored. But this small effect is enough to lock the "phase" of the ice cycles so that they do indeed coincide with the solar cycle as observed.

In Gildor and Tziperman's models the ice cycle carries on even when the thermohaline circulation is artificially held constant: "We claim that the THC is not an essential part of the 100,000-year cycles," though they say it does have a rôle in setting atmospheric CO₂ levels. Thus they propose, also controversially, that CO₂ levels have not driven the 100,000-year cycle in the past, but that they have been a result of changes in the southern oceans driven by an ice cycle in the north. The model does not, however, say anything about what today's unprecedented CO₂ levels would do to the ice cycle itself.

They furthermore propose that the change from 41,000-year to 100,000-year cycles 800,000 years ago happened because before that the world was too warm for much sea ice to form - so the rapid switching couldn't happen.

A lively debate followed this presentation. Such simplified models cannot make actual predictions, as Tziperman stressed in response to a question: what they are for is to generate new ideas that can be tested in global climate models. Research to deduce sea ice cover in the past is also indicated. The fate of Tziperman's model depends on whether evidence is found that land ice did in fact start growing in warm periods.

The effect of increasing carbon dioxide on the ocean

What happens if the thermohaline circulation (THC) turns off? Dr Richard Wood does climate modelling at the Hadley Centre for Climate Prediction and Research in the UK Meteorological Office. The one question to which an absolutely definite answer can be given is "is the THC important to climate" and the answer is "yes". What the effect will be leads to a group of interesting questions.

The Hadley Centre runs what the Chair, Jochem Marotzke, describes as "probably the best climate model in the world." And if you tell that model to kill the THC, it predicts that the Nordic Sea cools by up to 12°C; northern Scotland by 8°C and the UK as a whole by up to 5°C in the first 10 years. Then, according to the model, in the third decade the THC starts to recover slightly, the South Atlantic starts to warm and the UK warms by a couple of degrees. For comparison, the coldest year on record is 1740, in what's called the "Little Ice Age", at 2.5°C colder than average. If the average fell 3°C, you'd expect several years much colder than that.

How could the THC collapse? In principle, random variations could tip it into the "off" state, but this is "unlikely".

All but one of nine different climate models predict that increasing CO₂ in the atmosphere weakens the THC. In one model, a small increase in the sensitivity of the rate of exchange of water between air and ocean to temperature causes the THC to turn off and stay off. Another, given CO₂ at four times pre-industrial levels, has the THC turning off and then re-starting 1000 years later. The pre-industrial level was 180 parts per million; it has already more than doubled, to 370ppm.

The conclusion from the present models, then, is that failure of the THC has low probability. The enormous impact that it would have if it did happen justifies intensive work to refine the models, and politically raises the difficult question of applying the "precautionary principle".

The first step is to understand why the models give different answers. That should indicate what climate processes are more important and need further study in the real world. Probably, however, there will be "some processes that we're not clever enough and not rich enough to pin down," as Wood puts it. We're going to have to live with probabilistic predictions - refining the odds on different things happening.

And if we did start to see signs that the THC might be failing, it would probably be several years at least before scientists could give a useful estimate of the chances that it was actually happening, and we weren't just seeing a coincidental run of bad weather.

The future of the carbon cycle

Working out what happens as more carbon dioxide is added to the atmosphere entails knowing where it goes. The first place that comes to the minds of most people who think about this is certainly "into plants". But Professor Harry Elderfield FRS, of the Department of Earth Sciences at the University of Cambridge, is more interested in carbon in rocks. Chalk, limestone and marble are, of course, calcium carbonates (CaCO₃), and they represent a large proportion of the planet's carbon reservoir.

Once carbon dioxide has dissolved in seawater, it is present in three forms: dissolved CO₂ gas; carbonate ions (CO₃^--); and bicarbonate ions (HCO₃^-). Rivers add 10⁹ tonnes of dissolved carbonate to the oceans each year: about that much is deposited on the ocean floor. Before industry started adding CO₂ there were about 600 &mult; 10⁹ tonnes of carbon in the atmosphere, and 38,000 &mult; 10⁹ dissolved in the oceans. But the important thing to climate change is the flows between the three zones. Research to detail the ¹³C:¹²C ratio of carbon in minerals - which indicates how much of it originated in the biosphere and how much from minerals - is urgently needed.

For every 10 molecules of carbonate deposited as CaCO₃, 6 molecules of CO₂ are released back into solution. Adding CO₂ to the ocean shifts this ratio: it is expected to reach 10:8 by the year 2100.

On the face of it, this leads to yet another positive feedback loop: increased CO₂ dissolving from the atmosphere leads not only to deposition in sediments but to CO₂ levels increasing at an increasing rate.

But, Elderfield, pointed out, it's not so simple. Adding CO₂ also makes the water more acidic, which decreases creation of carbonate minerals. Experiments in the artificial coral reef within Biosphere 2 in Arizona confirm this, as do measurements of plankton skeletons in the real world. In sediments from the end of the last ice age, these carbonate shells were almost twice as heavy as they are now.

To add another level of complexity, there is the question of how much of the carbon taken up by plankton near the surface actually reaches the sea bed. The "ballast hypothesis", published late last year, suggests that the more minerals, including carbonate, are present, the more organic carbon makes it all the way to the sea floor. Organic carbon not attached to mineral "ballast" is more likely to be reabsorbed.

So we have another positive feedback: the more CO₂ there is at the ocean surface, the less carbonate is deposited and the less carbon is removed from the ocean in sediments. The amount thus "exported" is expected to halve by 2100. But the additional carbon is retained as carbonate in solution in surface waters.

More research is, as they say, required to pin down these complex interactions - and to incorporate them into global climate models.

Surprises in plant responses to CO₂

No climate model can be complete, of course, without taking account of the impact of carbon dioxide levels on plants on land and vice versa. Here, too, recent research has produced some surprises, challenging the assumptions that seemed like good starting points and continuing the trend that wherever we look we find more complexity.

Carbon dioxide is not just a greenhouse gas, Professor Christian Körner reminded the meeting, but also the "food" for all plants, and hence for all life (outside the deep ocean).

If we could deposit all the carbon in the atmosphere as soot on the earth's surface, it'd be just 1.4mm thick. This relatively tiny amount drives the activity of all the plants and the soil they grow in - which together contain 17 times as much carbon as does the air.

Before the age of industry, that layer of atmospheric carbon would have been only 0.7mm thick. All plant life evolved to cope with a CO₂ level of 180 parts per million: species that couldn't cope with that level are extinct.

So in evolutionary terms ecosystems have already experienced abrupt change in their diet. The trees in the park outside the meeting room started life with a CO₂ concentration at least 30 per cent less than what they're coping with now.

Many people have contested that increased CO₂ levels will necessarily lead to plants taking up more carbon and growing faster. But actual long-term research, growing plants in artificially CO₂-enriched atmospheres, shows that reality is by no means so simple.

Firstly: if you grow seedlings in a tidy greenhouse under increased CO₂, they indeed shoot up like beanpoles as expected. But adult plants may behave very differently.

Secondly, how plants respond to CO₂, depends on whether the soil is acid or calcareous (chalky or limey), whether they have an excess of nitrogen available or not, and what they are. Körner's group at the Institute of Botany of the University of Basel grew trees for three years under various conditions. On acid soils, the European beech Fagus sylvatica grows much less in doubled CO₂, unless you add nitrogenous fertiliser; it grows larger on limey soil whatever you do. Spruce, Picea abies, always grows bigger when CO₂ is doubled. This is the only such comparison done so far; all experiments on plants to date need to be repeated on many different species and all soil types.

Thirdly, the response of a plant community depends on its diversity. An experiment with an artificial species mix will miss something vital.

Fourthly, plants fed extra CO₂ exude carbohydrate into the soil, which affects soil bacteria to make less nitrate available, which restricts plant growth - unless fertiliser is added. Similar effects apply for other plant nutrients.

Fifthly, CO₂ response both depends on and affects water availability. The group did an experiment involving releasing 2 tonnes of cleaned industrial CO₂ per day through a kilometre of piping into a natural forest of 35-metre trees.

So: "have the trees read the textbooks?" They should waste less water when CO₂ goes up and they can close their leaf pores. The hornbeam Carpinus betulus behaves as instructed. But the European beech doesn't respond in this way at all.

Sixthly: there's the laughing gas effect. In Swiss calcareous grassland, doubling CO₂ with earthworms present doubles nitrous oxide emissions - and NO is a far stronger greenhouse gas than CO₂. Who could predict the chain of causality: CO₂ lets plants close their pores; they evaporate less water; more water stays in the soil; earthworms are happy; laughing gas comes out. And, as Körner says, "this is not stuff you could write in a proposal to a science funding agency".

Seventhly: CO₂ affects how nutritious plants are to animals (including us). High CO₂ boosts carbohydrate production, making proteins and all other animal nutrients proportionately scarcer. Lymantrius dispar larvae feeding on oak trees grow 30 per cent less under raised CO₂ - but 37 per cent more on hornbeam. This will affect the relative survival of the trees in a high-CO₂ world...

Eighthly, CO₂ can make the difference between growth and no growth at all for plants in shade. If you test the response of ivy to CO₂ in a clean and well-lit lab, you find no response. In the gloom of the forest, CO₂ doubles its growth. The same goes for tropical lianas, but more so: rainforests could be overrun. The current models assume that tropical forests will stock more carbon; the liana effect contradicts this and implies higher atmospheric CO₂ levels as trees die and decay earlier.

And ninthly, Körner repeats the warning implied in all of the above: do not assume simple first-principles effects. Everything is complicated and non-linear.

In conclusion, "Kyoto is driven by Rio." Policy-making on climate change in the "Kyoto process" meetings is dealing with carbon and the greenhouse effect; but what, Körner asks, "if biodiversity - the centrepiece of the Rio environment conference - can reverse the predictions? What if the tropical forests become a source rather than a sink of carbon?"

Those who talk about forests as "carbon sinks" are, Körner pointed out in response to a question, "mistaking turnover for capital". Trees grow; they may grow faster under increased CO₂; then they die and rot, probably faster under increased CO₂ too.

Can society cope?

If the presentations give the impression that a lot more work needs to be done to make global climate models more complete, think for a moment of the state of global economic modelling. And global social modelling isn't even on the horizon. So: what can we say about how society would cope with abrupt climate change?

The "mantra" in discussing this question is "low probability, high impact," as Professor Mike Hulme, of the Tyndall Centre for Climate Change Research and the School of Environmental Science, University of East Anglia, recalled. There hasn't been a lot of social, economic or ecological science done on what those impacts might be. How likely is abrupt climate change and would it in fact catastrophic?

Defining what is "abrupt" is a difficulty: something that happens over the next 50 years is practically instantaneous to a geologist but in the remote future to an economist. So the Academy of Sciences report Abrupt Climate Change: Inevitable Surprises contains another definition, additional to those mentioned above: it is "significant change in climate relative to the background or accustomed climate experienced by the economic or ecological system... having sufficient impacts to make adaptation difficult".

The key point here is the expectation of what the climate will be. Crudely, a society that is expecting the climate to warm gradually by 5°C may be able to adapt to it; but if that society instead found mean temperatures dropping quickly by 2°C adaptation options would be limited.

It may well be that the reversal of direction of change is more significant in discussing adaptation than is the rate of change. (The frequency of extreme weather events is another factor to consider.)

Can we plan society so that it is more resilient to both (or all three) types of change? If so, how?

The largest single example of a reversal in climate change is the Sahel region south of the Sahara. From the late 19th century (the time when Northern colonial powers started taking an interest and records, as it happens) to the 1950s the climate was getting wetter. Then between 1952 and 1971 rainfall fell every year, by 40 per cent in total.

So there was no momentum, interest or investment in coping with drought: and this lack contributed to the significant human disasters of the famines. But the drought did not, as predicted, entirely destroy the communities of the Sahel. The survivors still live there, rebuilding sustainable societies.

Are there such things as "no regrets" adaptation policies, or policies that increase adaptability to all kinds of change? Funding equivalent to one per cent of the RAPID programme is going to examining such questions. Some idea of what defines "success" in adaptation would be a good start.

We're not even sure what the effects of THC collapse would be, apart from the temperature in North-West Europe being lower than it would otherwise be. It's tempting to assume that it reverses all the changes expected with warming - for example that it makes winters drier and summers wetter. Can the climate system really be that symmetrical? Almost certainly not.

Scientists also need to get a lot better at explaining the odds on a climate event - sustained cooling over five years, for example - being a either sign of a systematic change, or of mere natural variability. Policymakers who understood the implications of probabilities for policymaking would be a good thing, too.

Economics of Abrupt Climate Change

How does the possibility of climate change being abrupt (rather than gradual) affect the way that policymakers make decisions about it? Professor Charles Perrings, an economist from the University of York, introduced the fundamental distinction between policies of "mitigation" and of "adaptation". The former set out to reduce the likelihood of a damaging event - which assumes that there is some action that can affect this. Policies for adaptation aim to reduce the cost of the event.

A "war on poverty" would be mitigation. "Wars" on drugs, terrorism and disease would be adaptation, in that they aimed to deal with the consequences of poverty.

Equally, reducing carbon dioxide emissions mitigates climate change. Relocating populations away from drought-stricken or flooded areas would be adaptation. The main response to climate change in the past has been migration.

Economists' calculations of the cost and benefit of mitigation depend critically on how far into the future their model looks, and on the "discount rate" that weights future effects. The latter is "what makes the economist nobody's friend," Perrings observed: "if we discount the future at three per cent or five per cent a year, if something happens 30 years into the future and it costs a great deal, why worry? It'll be pretty much fully discounted by the time you bring [the cost] back to the present."

The climate system is a public good, and effort and expenditure put into mitigation benefit the whole world. By contrast, no-one gets a free ride on adaptation. Unsurprisingly, then, economic models favour adaptation.

A major complication with abrupt change is that the way people deal with low-probability, high-dread events is irrational from an economist's point of view: compare attitudes to car accidents and nuclear accidents. This leads to the "precautionary principle" - that where costs are uncertain, but both high and irreversible, society should take action before the uncertainty is resolved.

Abruptness, then, favours mitigation. The economic models used in assessing climate change need to be changed to take account of this. They also need to deal with the problem of equity: that abrupt climate change is likely to hurt poor people more than the rich.

But it also encourages rich countries to believe that they will win out, in a competitive sense, through waiting and betting on their capacity to adapt. Their (poor) competitors would lose out twice: absolutely, from the direct costs of climate change, and relatively from rich countries' advantage in adapting better. What that "wait" policy ignores, however, are the costs of the resulting increasing inequality: wars on terrorism, disease and migrants.

Scientific challenges & new research

For scientists, the purpose of such meetings is to identify areas of uncertainty and research that can address them. Dr Meric Srokosz, who works at the James Rennell Division for Ocean Circulation and Climate at the Southampton Oceanography Centre, concluded this one by outlining research under the UK Natural Environment Research Council (NERC) Rapid Climate Change programme (RAPID).

The programme involves scientific collaboration with Norway, the USA and other countries. It has £20 million of funding over six years. Its main focus is the THC; research projects include present-day observations, collecting data about its past state, and modelling it. Understanding of the potential for rapid climate change will depend on linking the results of these.

The objectives (scientific challenges) are:

• To set up a prototype system to observe the strength and structure of the THC in the North Atlantic; this may lead to an "early warning" system.
• Other observations of the oceans, sea-ice and ice sheets.
• Gathering data on past states and refining error estimates in these.
• Developing high-resolution models to synthesise these data.
• Developing models of varying complexity to understand and predict changes.
• Examining responses to changes in the THC through, for example, the pattern of storm tracks changing atmospheric energy and moisture transport.
• Testing models against the data.
• Attempting to quantify the probability and size of future climate change, and to calculate the uncertainty in these estimates.

To achieve the last objective will require bringing together the results of all the others and, crucially, integrating consideration of processes that take place on different time and space scales.

For details of the projects funded under the RAPID programme, and links to programmes with which it is co-ordinating, see http://rapid.nerc.ac.uk