Sending out an SOS
19 Jun 2011
The protests and uprisings that shook the Arab world this year took political leaders and analysts by surprise — not least in the US government, which has spent more than $100 million on trying to predict such events. Hauled before Congress to explain why US spooks failed to foresee regime-topping protests in Tunisia and Egypt, Director of Intelligence James Clapper said: “We are not clairvoyant.”
Predicting such radical change is not just of interest to politicians. Among the experts now picking through the Internet and Twitter chatter from before the protests for any indications of what was to follow is Marten Scheffer, an ecologist at Wageningen University in the Netherlands. It is a daunting task, but he has hope it could work, because of early successes in looking for warning signs of impending upheavals in an entirely different realm — in biological systems under stress.
Such upheavals in the natural world, called regime shifts, can bring rapid change on a massive scale. Pushed by stresses such as rising temperatures and acidification, many ecosystems do not respond gradually, but instead in abrupt and rapid jumps from one state to another. Tropical forests die off and turn to savannah. Corals bleach and are overrun by seaweed. Clean freshwater lakes, havens for fish and plants, can be swamped by cyanobacteria that crowd out other life. And as the effects of climate change deepen, more and more ecosystems become vulnerable to such regime shifts.
“One of the big challenges is can we somehow predict such radical changes?” asks Scheffer. Some studies suggest that we can. “If a system is gradually becoming more fragile, you can sense that the system is losing resilience.”
Such loss of resilience can bring severe impacts. A classic example is the cod population on Canada’s Atlantic coast, where overfishing has led to a string of catastrophic stock collapses. After rising for hundreds of years, the catch suddenly plummeted in the early 1970s, and then again in 1990. In 1992, Canada put a moratorium on cod fishing — putting tens of thousands out of work — and yet cod populations there still remain low.
When it comes to regime shifts caused by climate change, most attention has focused on critical and large-scale systems that could cross tipping points, whereby a small push can lead to large changes, such as a die-back of the Amazon rainforest or disintegration of Greenland’s ice sheet. Yet by the time we realize that these large- scale systems are close to moving into an alternative state, it will probably be too late to prevent a shift. By that time, the global climate may respond too slowly to actions such as cuts in greenhouse-gas emissions to make a difference.
Regime shifts in smaller ecosystems could be easier to prevent, and new studies — drawing on both computer models and field data — offer encouraging signs that this could be possible. Researchers have identified a number of tell-tale signs that such changes are imminent, and are now looking at whether these clues show up in other systems as well. If they do, then scientists will be one step closer to being able to identify vulnerable systems and to stop regime shifts before they happen.
Against the grain
Monitoring ecosystems for these early warning signs could allow more targeted conservation efforts. “You could see whether certain lakes or corals are closer to critical transitions than others,” says Scheffer. “It’s good to know which ones are almost on the edge so you know where to put your money.” That will be important as climate change is likely to make such shifts occur sooner and more frequently. And once ecosystems enter a new state, it can be tough, or sometimes near-impossible, to flip them back.1
Research on early warnings of regime shifts is still in its early days, and exact predictions are still elusive. “We can say that things are getting worse than they used to be and are heading towards a tipping point,” says ecologist Vasilis Dakos, a colleague of Scheffer’s at Wageningen University. “But how close? In practice, it still seems impossible to say.”
Early research is, however, beginning to disprove the old belief that such events strike by surprise. “One of the key features of these regime shifts has been their unpredictability,” says zoologist Stephen Carpenter at the University of Wisconsin in Madison. In the past, he says, “when talking to ecosystem managers, we always made a big deal about how unpredictable these things are.” Scientists, he explains, wielded this uncertainty as a reason to be careful about pushing systems towards vaguely understood tipping points.
Now, though, researchers are detecting possible early warning signs by analysing the circumstances that surround regime shifts in a controlled environment. One way to do this is to use computer models, such as those of freshwater lakes.
Regime change in such environments can be devastating. Lakes that support healthy fish populations typically have relatively clear water. However, some lakes undergo dramatic and rapid colonization by cyanobacteria, which cloud the water and block out the light needed to support plants, and in turn fish. “The cyanobacteria were usually considered the effect of high nutrient loads [such as nitrogen and phosphorus in run-off from farms],” says Scheffer. But climate change also has an effect.2 “If there’s warming, then what used to be safe limits for nitrogen and phosphorus are no longer safe limits,” he says. In part this is because the higher temperatures make cyanobacteria grow faster.
These lessons have come at a price. “Sadly, people have wrecked a lot of lakes in the world,” says Carpenter, “but some were monitored as they collapsed.” Similarly, overgrazing has ruined much cattle pasture, with grasses replaced by thickets of shrubs. “These are systems where we’ve been able to learn something,” he says. The data gathered during these ecological changes could contain clues needed to identify other systems poised to go the same way.
Research suggests that as ecosystems get closer to a tipping point, they lose resilience and are less able to recover when hit by a shock, such as a severe storm. There’s no way to measure loss of resilience directly, but its effects do show up subtly in the way an ecosystem fluctuates in response to change in the environment around it.
Models have thrown up several signals that could act as red flags and indicate that a system is losing resilience, and so may be about to undergo a regime shift. One of the strongest candidates is the appearance of wild fluctuations in a key aspect of an ecosystem, such as an organism’s population size.
Many of these possible early warning signs trace back to a phenomenon known as critical slowing down. This sees a system lose its ability to recover from even small disturbances as it approaches a tipping point. Although critical slowing down is a well-known effect in physics and mathematics, it is new to the study of living systems.
Critical slowing down in an ecosystem means that small changes in the environment stick. This allows several minor events to combine to push a system variable, such as population size, far from its historical average. Statisticians call this measure of divergence the variance. In a steady system the variance will be roughly constant. In systems that are approaching a tipping point, however, the variance will often gradually increase, acting as a distress signal that the system is more and more likely to flip into a new state.
Scientists have focused on changes in variance as a way of hunting for early warning signs in living systems. Heavily fished species of fish along the California coast, for example, have shown much larger fluctuations in abundance than unharvested species.3 In a computer model of a lake receiving nutrient-rich run-off, the difference is seen in measurements of a key nutrient, phosphorus. Primed for a regime shift from clear water to cloudy, the model showed phosphorous levels in the lake were steady until about a decade before, when the variance of the measured levels began to rise steadily. About a year before the regime shift, the variance leapt higher still.4
Even though such early warning signs show up clearly in models, the models are not yet sophisticated enough to predict when a real ecosystem will cross a tipping point. They can indicate which systems could be affected, but not when the regime shift will occur.
To be useful in real-world situations, researchers agree that possible early warning signs need to be put to the test in living ecosystems. Ecologists John Drake of the University of Georgia in Athens and Blaine Griffen at the University of South Carolina in Columbia have done this with Daphnia magna — a species of zooplankton known in common parlance as the water flea — to understand the conditions that would drive them towards extinction.
The scientists gradually starved groups of water fleas, pushing their populations towards a point where individuals reproduced too slowly to keep up with the rising death rate. It took about nine months for the water flea populations to reach this tipping point. But three months before that, on average, the warning signs began to show up — including rising variance in the population size.5
The water-flea experiment was a very simple system, with a population gradually approaching extinction — and it also had the benefit of rigorous controls in the lab. “Our data were very high quality compared with what you get in the field,” says Griffen. Seeing early warning signs in nature “will be a lot more difficult, simply because things will be noisier.”
It is difficult, but not impossible. In a study published in April in the journal Science, Carpenter and his colleagues tested the idea at a lake in Michigan.6 They added more largemouth bass, doubling the population of this predator, and watched as the lake ecosystem went through a complete re-organization. The bass took over, and the number of plant-eating fish dropped off; pushed out of the open water, they instead hugged the shorelines.
Over several months, the researchers studied the populations of the predator and prey fish. To capture a detailed record of the response of photosynthetic plankton, a key player in the lake, they measured chlorophyll levels in the water every five minutes using robotic samplers.
Sure enough, before the fish were added, the chlorophyll levels bounced up and down from day to day. But after the population of the predator was doubled , these ups and downs began to stray increasingly further from the average. The variance increased, and it did so months before the bass became dominant in the lake — providing a possible window of time to intervene and prevent the regime shift.
“Variance provided the clearest signal,” Carpenter says. Another measure that worked as an early warning signal was the ‘return rate’, which reflects a system’s ability to return to its average state after being perturbed. As the bass took over the lake, the return rate of chlorophyll levels dropped to near zero. “The statistical signals were evident long before there were obvious changes.”
There are other indicators of impending change too. Some researchers — including Dakos — are pursuing a technique in which they search for early warning signs in high-resolution satellite images of a landscape, taken one year after another. Gathering quality data can be challenge, Dakos explains. “But the good part is you could take snapshots from now and a year ago, and don’t need to measure the points in between.” Earlier work with a variety of models — with different ways of representing particular ecosystems, such as arid grasslands, and other kinds of ecosystems, such as lakes — suggest these might manifest as changes in simple measures, such as patterns that appear in where plants grow.
For example, models of regime shifts in freshwater lakes also show changes in plant distribution around the lake. Dakos and colleagues found this spatial correlation — a measure, at any given moment, of how similar the amount of plant cover was in one patch of a lake compared with neighbouring patches — could serve as an early warning sign. They found that as the lake approached a tipping point, the spatial correlation in plant cover began to rise gradually, and then spiked just before the lake flipped into the cloudy state. “Locally, the patches became very similar to their neighbours [in the amount of plant cover],” Dakos says. “It’s as though you don’t have your own will anymore, and start looking like all of the people around you.”
They also compared this spatial correlation with another statistical measure called auto-correlation — a measure of how much overall plant cover changes from day to day. In the simulations, they found that this auto-correlation also worked as an early warning signal — but the spatial signal was stronger, and showed up sooner.7
Studies like this are uncovering a variety of possible early warning signs of regime shifts, some of which may work better than others. Yet not one of these, Dakos cautions, is yet considered infallible. “All of the indicators fail in some cases.” There is no single signal — be it variance, spatial correlation or auto-correlation — that works in all ecosystem types.
In a study that demonstrates these difficulties, Dakos and colleagues considered how arid grassland converts to desert by using three models that each simulated the ecosystem in a slightly different way. Such grassland can be pushed to shift regimes either by declining rainfall or by overgrazing.
In two models, they saw several clear warning signals of impending collapse, including increasing day-to-day variance in the amount of grass growing in a region and correlation between patches so that they became increasingly like their neighbours.
However, in the third model, most of the early warning signals they looked for were missing. The results of this model showed that when the grass was close to extinction, it would take on specific patterns — such
as evenly spaced patches, each about one to ten metres across. If the grass was in difficult but not so dire conditions, then it would grow in meandering parallel strips, like a maze.8 In ecosystems such as this, the patterns themselves may act as indicators of approaching tipping points, Dakos explains.
The lesson is that real-world monitoring for early warning signs would have to be tailored to each ecosystem, based on what models and real-world tests suggest are the most likely indicators to be present.
However, some argue that the haphazard appearance of warning signs in models may make them too unreliable to use at all. Alan Hastings, an ecologist at the University of California in Davis, showed last year that several early warning signals, such as variance, failed to show up as expected in some common ecological models.9 Perhaps the models that do show early warning signs are exceptional, and maybe only a limited number of ecosystems will show useful indicators, Hastings explains. “We need data from lots of real systems [to put the models to the test].”
“False alarms or no warnings are a big problem,” Dakos says. Carpenter agrees: “Before we put them into use, we need a lot more field trials.”
Even if early warning signs don’t give reliable evidence of an impending tipping point, they could still be useful to manage ecosystems better, Carpenter argues. “The real question is whether management with indicators has higher expected performance than management without indicators,” he says. “Even an imperfect indicator can have a big expected pay-off.”
Carpenter argues the effort is worth it because it could help prioritize spending on restoration and remediation work by steering efforts towards ecosystems that seem closest to problems.
Reinette Biggs, an ecologist at Stockholm University in Sweden, thinks that they could also offer hope to those working to stem the tide of ecosystem collapse. “Sometimes it seems as though things aren’t happening,” she says. Improved indicators of resilience could show that intervention is helping to pull the ecosystem back from the brink. Instead of warning signs, they would be signs of success.
1. Scheffer, M. et al. Nature 461, 53–59 (2009).↑
2. Kosten, S. et al. Glob. Change Biol. 15, 2503–2517 (2010).↑
3. Anderson, C. N. K. et al. Nature 452, 835–839 (2008).↑
4. Carpenter, S. & Brock, W. Ecol. Lett. 9, 311–318 (2006).↑
5. Drake, J. & Griffen, B. Nature 467, 456–459 (2010).↑
6. Carpenter, S. R. et al. Science doi:10.1126/science.1203672 (2011).↑
7. Dakos, V. et al. Theor. Ecol. 3, 163–174 (2010).↑
8. Dakos, V. et al. Am. Nat. doi: 10.1086/659945 (2011).↑
9. Hastings, A. & Wysham, D. Ecol. Lett. 13, 464–472 (2010).↑