The Cassandra factor
(New Scientist Via Acquire Media NewsEdge) Scientific soothsaying: seeing disasters before they strike From ecosystem collapse to climate crunch, the warnings of future catastrophe could be staring us in the face OF ALL the tragic protagonists of the Trojan wars, Cassandra cuts the loneliest figure. The soothsaying daughter of the king of Troy, she saw the warning signs that foretold the downfall of her city. But she was fated not to be believed?- and all her knowledge could not prevent the impending catastrophe.
Fast-forward a few thousand years, and disaster looms ever larger on our horizons. Climate change, ecosystem crunches, market crashes: stability seems a fragile and transitory state, threatening at any point to tip into uncertainty and chaos.
What we wouldn't give for a way to see the future and head off disasters. "There are so many massive changes that are really important for people?- the collapse of fisheries, water supply problems, desertification or species invasions?- but we are typically surprised when these changes come," says ecologist Steven Carpenter of the University of Wisconsin in Madison. "Early warnings could save a lot of money and a lot of human suffering." Carpenter is at the centre of developments that could legitimise soothsaying with solid science. As we get to grips with the dynamics of complex systems with many moving, interacting parts, from the ecosystems that Carpenter studies to financial markets and the human brain, we are beginning to see subtle similarities in how they work, and in particular how they signal stress in the lead-up to a meltdown. Look hard enough and the same precursory signals of catastrophe seem to crop up all over the place. Might they allow us to leap over Cassandra's shadow?- to foretell disaster, but also act to avoid it? The conventional view is that "tipping points" in complex systems are fundamentally unpredictable. Imagine a smouldering pile of sticks. They might smoke for a while and eventually die out, or they might suddenly kindle and burst into flames. Without knowing every detail of the pile of twigs and their surroundings?- the exact arrangement of the pile, the temperature and humidity of the environment and so on?- it is impossible to tell when, if at all, a flare up might occur.
Something bugging Carpenter's first inkling that this might not be the whole story came one night seven years ago on the Caribbean island of Tobago. He and some fellow ecologists were relaxing in a bar after a conference session, and talk turned to a computer program one of the party had written to simulate outbreaks of spruce budworm. Every few decades, numbers of this conifer-munching insect explode, bringing devastating defoliation to North American forests. But there seemed to be an irritating bug in the program itself: just before an outbreak, the virtual budworm populations showed puzzling, jagged fluctuations in their numbers.
That intrigued Buz Brock, an economist colleague of Carpenter's at Wisconsin who was also present that evening. He had been working at the mathematical interfaces of economics and ecology for two decades, and he grabbed some paper and started scribbling equations. Two hours on, he had what looked like an answer to the budworm mystery?- and a hint of a bigger discovery, too.
What he had jotted down were some calculations from bifurcation theory, the branch of applied mathematics used to characterise how a system's internal dynamics change, often abruptly, in response to mainly gradual changes around it. The budworm issue led him to see a new wrinkle in the equations. It seemed that the slow slide of a system into a new, potentially unstable state would first be reflected in subtle changes in the natural patterns of variability of the system, long before any tipping point was reached.
With budworms, it is easy to see why this might be the case. In the short term, the population of the insect would vary up and down, but natural feedbacks would act to keep it close to some fixed value. More budworms, for example, would mean more tasty meals for birds, who might shift their attentions from other food sources, pushing down budworm numbers once more.
In the longer term, however, gradual changes in outside conditions could bring more subtle effects into play. For example, over many years the foliage harbouring the budworms would grow thicker as trees grew and matured, making it progressively more difficult for birds to find the grubs. That gradual change, outwardly almost imperceptible, would first make itself known in how a budworm population fluctuated: each time the number of grubs was higher than average, it would take longer to sink back to its equilibrium value. Eventually, the long-term increase in foliage density, coupled with a natural short-term rise in the bug population, would be enough to render the birds' foraging strategy ineffective. Budworm numbers would start increasing exponentially and the system would tip abruptly into a radically different state, with catastrophic consequences for the forest.
This sort of "critical slowing down" in the response to natural perturbations was just the thing to explain the mysterious fluctuations in the virtual budworm populations (see diagram, page 40). It was not a new idea; something similar had been spotted in systems from atoms emitting laser light to patterns of neural activity in squid brains. But Brock was the first to suggest that it might be a consistent early warning sign of events to come. "That's when we started wondering if Buz hadn't hit on something much bigger," says Carpenter. If critical slowing down could be used to predict tipping points in budworm populations, why not also in other systems where these slowing downs occur? "The more excited we got, the more we also wondered, are we kidding ourselves?" says Carpenter.
If they were, they weren't the only ones. Around the same time, climatologists Hermann Held, Thomas Kleinen and Gerhard Petschel-Held at the Potsdam Institute for Climate Impact Research in Germany were noting something similar. They were modelling how additional fresh water, for example from melting ice caps, would alter the flow of the ocean currents that carry energy and dissolved materials around the globe. As meltwater increases, this normally highly predictable flow starts to become less regular in time and space, before eventually, at much higher meltwater influx, ceasing altogether (Ocean Dynamics, vol 53, p 53).
Such effects are not just confined to computer models. In 2006, Chih-hao Hsieh, then at the Scripps Institution of Oceanography in La Jolla, California, and his colleagues reviewed a 50-year database of fish larvae populations. Leaving overall declines aside, they found that the variability in species exploited by commercial fishing?- which are more prone to devastating crashes?- matched the changes predicted by theory, even when those populations had not yet been so overfished as to collapse (Nature, vol 443, p 859).
Vasilis Dakos and Marten Scheffer of the University of Wageningen in the Netherlands and their colleagues, meanwhile, were looking into eight cases of abrupt changes in Earth's past climate. These ranged from the transition from a balmy tropical state to a colder climate with ice caps 34 million years ago to an event 5000 years ago when the north African landscape switched abruptly from a savannah dotted with lakes to desert. In each case, the researchers identified a sudden increase in the autocorrelation of the temperature record in the time leading up to the transition. Autocorrelation is a mathematical sign of critical slowing: it reflects the predictability of a time series, or how closely its behaviour correlates with what it did in the recent past. As a system approaches a tipping point and its responses to natural perturbations grow slower, that autocorrelation grows larger (Proceedings of the National Academy of Sciences, vol 105, p 14308).
Induced collapse What could be the final proof of a pervasive predictive effect came last September. Ecologists John Drake of the University of Georgia in Athens and Blaine Griffen of the University of South Carolina in Columbia induced populations of zooplankton to crash by slowly reducing the food available to some of them. In those populations destined for extinction, the predicted signs of critical slowing in their fluctuations showed up as much as eight generations before extinction (Nature, vol 467, p 456).
"This is a real landmark," says Scheffer, who was not involved in the research. "It reveals theoretically predicted signals in a real biological system in a controlled experiment." So can we use this to our advantage? Perhaps. Scheffer and others readily admit that moving from occasional laboratory success to consistent practical application will be far from easy.
One problem is how to deal with the possibility of false predictions - something that previous experience has shown to be a tough nut to crack (see "All in the mind", left). Changing tack in areas such as climate change and biodiversity to avoid assumed tipping points can incur huge costs, so policy makers will demand total confidence that when an early warning sign comes it means they must act. And even if we can guarantee total confidence in the predictions, will we be able to spot the warning signals far enough in advance to make a difference? To find out, Carpenter, Brock and ecologist Reinette Biggs of Stockholm University in Sweden developed a computer model of a fishing ground in which they varied fishing and shoreline development policies and watched for their effects on the virtual fish stocks. Shoreline development can have a profound effect on fish numbers by depriving fish of natural habitats and increasing harmful run-offs. But once developed, a shore is not easy to undevelop, at least quickly. In this case, the warning signs of critical slowing and increased autocorrelation did indeed show up, but too late for any change in development policy to feed through and avert a collapse. "If you wait for clear evidence of negative environmental impacts, you may well be too late to do anything about it," says Brock.
In the case of simple overfishing, however, the outcome was more positive. If policies such as fishing moratoriums were implemented immediately after an early warning was received, the collapse of fish populations could be prevented. (Proceedings of the National Academy of Sciences, vol 106, p 826). This suggests that with the right high-precision data to hand we can recognise and avert impending catastrophes. "With the rapid growth in high-frequency environmental monitoring equipment, this may be more possible in the future," says Biggs.
The same possibilities and caveats apply to the largest complex system where advance, undeniable warning of impending disaster would be very useful: the global climate. In a recent review, Tim Lenton of the University of East Anglia in Norwich, UK, alongside Scheffer and others, has listed the threats most likely to cause major shifts in Earth's climate, such as the melting of the polar ice caps. Any kind of concrete early warning would be valuable, says Lenton, and better data is the key to spotting them.
Even if we were to see incontrovertible signs in time, that does not necessarily mean we will be able to agree on the right actions, says John McNeill, an ecological historian at Georgetown University in Washington DC. "Unless we know those things in convincing detail and with near-unanimity, human collective-action problems bedevil effective action," he says.
Our understanding of tipping points and the warnings we receive of their approach is undoubtedly improving in leaps and bounds. But predicting whether we can use that to sharpen up our act may be the hardest call of all. n Mark Buchanan is a freelance writer based in the UK Defoliation occurs when budworm numbers pass a tipping point FINANCIAL WHIRL Might financial markets also give consistent warning of their collapse? That possibility awakes intense interest given the profits to be made by any financial whizz kid who can see the future.
So far, the science is ambiguous. Economist John Geanakoplos of the University of Yale, working with physicists Doyne Farmer of the Santa Fe Institute and Stefan Thurner of the Medical University of Vienna, has studied a model of competition between hedge funds, which try to attract investors by earning higher returns than their competitors. They found that this competition can drive the market past a tipping point, triggering a sudden, self-amplifying spiral of losses. Long before that collapse occurs, however, the researchers observed ever more skittish flows of money in and out of the funds.
These model predictions seem to fit closely with the actual behaviour observed before a real-world disastrous event in August 2007 known as the "quant meltdown". But in general, reliable evidence for early warnings in financial systems is scant. Where money is at stake, any successful method of prediction would quickly invalidate its own forecasts, as investors change their strategies to avoid its predictions.
All in the mind One area where tipping points have a long history is in medicine, especially in studies of that most complex of systems, the human brain. As long ago as the 1970s, researchers at the McDonnell Douglas Astronautics Company in West Huntington Beach, California, built a pocket-sized EEG device that used a statistical analysis of patterns in the brain's electrical activity to give early warning of epileptic seizures?- the point where the normal patterns of neuronal firing give way to excessive or synchronous activity, with incapacitating effects.
Their aim was to give people with epilepsy sufficient warning of an attack that they could cease or avoid dangerous activities?- driving a car, for example, or crossing the road. Ultimately, the plan was to develop an implant that could deliver electric shocks or a drug infusion to head off a seizure.
In a small study limited to five people, their technique accurately predicted about 90 per cent of the seizures tens of seconds to minutes before they came. But it also frequently predicted seizures that never came. That risked causing unwarranted alarm and could have subjected someone to unnecessary treatments. Since then, decades of effort have gone into developing more specific ways to foresee seizures that might overcome this problem by looking at more complex features of the patterns of electrical activity within the brain. Success has been limited.
In fact, proving any kind of practical predictive ability to be more successful than random guessing isn't as easy as you might expect. When it comes to blind studies of the effectiveness of different algorithms for predicting seizures, for example, there is still no clearly accepted definition of what actually constitutes the onset of a seizure. "The idea seems simple enough," says Brian Litt of the University of Pennsylvania in Philadelphia. "You make a prediction, and then within some predetermined period a seizure either occurs or does not. But in practice you face difficult questions like 'Did a seizure really occur?'." (c) 2011 Reed Business Information - UK. All Rights Reserved.
IoT in the Channel
Privacy for IoT
Privacy for IoT
Solutions: Smarter Track
AI Use Case
AI Use Case
On Premise Strategies: CBRS and Private LTE
On Premise Strategies: CBRS and Private LTE