Authors: Bill Bryson
How large are the scientific uncertainties, though? People often say that meteorologists’ inability to predict weather credibly beyond about ten days bodes ill for climate projection over decades. This misses a key difference between the instantaneous state of the atmosphere – weather – versus its time and space averages – climate. Even though the evolution of atmospheric conditions is inherently chaotic and the slightest perturbation today can make a huge difference in the weather a thousand miles away and weeks hence, large-scale climate shows little tendency to exhibit chaotic behaviour (at least on timescales longer than a decade). Good models can thus make reasonable climate projections decades or even centuries ahead if the processes forcing change are large enough to detect above the
background ‘noise’ of the climate system – the unpredictable part. The Intergovernmental Panel on Climate Change (IPCC)’s laboriously compiled projections combine such modelling with scenarios for greenhouse gas emissions based on different assumptions about economic growth, technological developments, and population increase.
1
These scenarios, despite major differences in emissions, show paths for global temperature increase that do not diverge dramatically until after the mid twenty-first century. This has led some to declare that there is very little difference in climate change across scenarios, and therefore, emissions reductions can be delayed many decades. That is a big mistake. It takes many decades to replace current polluting energy systems. There is also delay between emissions and temperature change due to the thermal inertia in the climate system caused by the large heat capacity of the oceans. After the mid twenty-first century, there are large differences based on emissions over the next few decades in the projected temperature increases – and the risks of associated dangers – for the late twenty-first century and beyond. Some of these risks imply irreversible changes.
Much of the uncertainty contributing to the ranges of projected future temperature increase derives from the so-called climate sensitivity. How much warming can we expect a given amount of greenhouse gas to cause? It is often estimated as the equilibrium global mean surface temperature increase due to a doubling of atmospheric CO2 from pre-industrial levels of about 280 parts per million. The IPCC estimates that it is ‘likely’ (there is a 66–90 per cent chance) that the climate sensitivity is between 2 and 4.5 °C and roughly a 5–17 per cent chance that it is above 4.5 °C (with the remainder being the chance it is less than 2 °C). They also offered a ‘best guess’ of 3 °C climate sensitivity.
Many studies have produced probability distributions for climate sensitivity with a long right-hand tail, meaning that high climate sensitivity values, while relatively unlikely, still register a probability of a few per cent or more. One example is displayed in figure 1, which shows a very uncomfortable 10 per cent chance that the climate sensitivity is higher than 6.8 °C. The median result – that is, the value that climate sensitivity is as likely to be above as below – is 2.0 °C, while there is a 10 per cent chance the climate sensitivity will be 1.1 °C or less. Like all model dependent studies, the detailed numerical values should not be taken literally, but the overall message must be taken seriously.
Our uncertainty goes beyond scientific understanding of the scale and distribution of climate changes from any single scenario of increasing greenhouse gases to include the trajectory of human development and our adaptive capacity. Moreover, future greenhouse gas emissions are heavily dependent on policy choices worldwide. But we do know that if we wait to act until an increase in undesirable impacts occurs, the inertia in the climate system and in the socioeconomic systems that produce greenhouse gas emissions will have committed us to even more severe impacts stretched out over many decades to centuries.
We cannot eliminate all of the important scientific uncertainties, but we can be more precise about their extent. That, however, is only part of the scientists’ job. We also have a responsibility to communicate all of this as well
as we can. Communicating this complex systems science to policy-makers and the public is difficult. Too often, confusion reigns when an advocate for strong policy cites a well-established severe outcome as the most important consideration, and another advocate from some enterprise institute disliking public control of private decisions cites speculative components of the systems analysis as if that is all there were. Not surprisingly, politicians, media, and just plain folks get frustrated by this ‘duelling scientists’ mode of presentation, an unfortunate staple of the mainstream media.
Professional training also leads too many scientists to ‘bury our leads’, as American journalists would put it, rather than finding effective ways to communicate complex ideas. Being straightforward and understandable is a challenge given the strong scientific tradition of full disclosure, which makes us lead with our caveats, not our conclusions. But what I call the ‘double ethical bind’ – be effective in public communication even if that means there isn’t enough space or time to present all of the caveats – is not unbridgeable. It calls for the scientist to develop a hierarchy of products ranging from sound-bites on the evening news to get our findings headlined on the agenda, to short but meatier articles in semi-popular journals like
Scientific American,
to more in-depth websites, to full-length books in which that smaller fraction of the public or policy worlds that actually want the details about the nature of the processes and how the state of the art has evolved can find them. Yes, it is very time-consuming to produce websites or long books with the details, but it is also necessary for those in complex systems science fields like climate science to simultaneously be effective in public messaging, where all the details are not feasible to communicate, but the longer backup materials can honestly separate the components of the science that are well established from those best characterised as competing explanations and from those which are still speculative.
The Royal Society and my own National Academy of Sciences (if less boldly, I think) have moved into this realm with clear statements of the potential risks of climate change. An evolving series of pronouncements
include the joint statement of 2001 of the Royal Society with fifteen other national science academies on the science of climate change.
3
The statement of June 2005 on global response to climate change by the science academies of the G8 nations and of China, India and Brazil stressed that the scientific understanding of climate change is now sufficiently clear to justify prompt action.
4
There followed the May 2007 statement on sustainability, energy efficiency and climate protection of the national science academies of the same countries plus Mexico and South Africa
5
and most recently the June 2009 joint statement calling for the transformation of the G8+5 nations’ energy strategies.
6
In addition, I always push at our annual US National Academy membership meetings for us to be more publicly oriented, but it comes slowly. I am glad that our new NAS President, Ralph Cicerone, is committed to communicating quality science in the public interest. It is also encouraging that President Obama’s new science adviser, John Holdren, is more in the mould of former UK government adviser and Royal Society President Lord May than some previous science advisers in the US who tended to carry the administration’s message to the science community, rather than the other way around, as in the case of May or Holdren.
Along with climate projections, scientists also have to explain how systems science gets done. We cannot usually do traditional ‘falsification’ controlled experiments. What we can do is assess where the preponderance of evidence lies, and assign confidence levels to various conclusions. Over decades, the community as a whole can ‘falsify’ earlier collective conclusions – like the sporadic suggestions in the early 1970s that the world would cool. But in systems science it sometimes takes a score of years to even discover that certain data were not collected or analysed correctly, as well as continuing to identify new data, and such discoveries are rarely by individuals but by teams and even assessment groups.
When I first got involved in discussing the range of outcomes in climate change, I didn’t understand Bayesian versus frequentist statistics, but in fact that was the heart of the matter – how to deal with objectivity and subjectivity in modelling and in projections.
As Bill Bryson mentions in the Introduction, the English clergyman and mathematician Thomas Bayes (circa 1702–61) formulated an approach to probability now called Bayesian inference. His key theorem was published posthumously in 1764. In essence, it expresses how our knowledge base – and prejudices – establish an
a priori
probability for something (that is, a prior belief in what will happen based on as much data and theory as is available). As we further study the system, obtaining more data and devising better theories, we amend our prior belief and establish a new,
a posteriori
probability – after the fact. This is called Bayesian updating. Over time, we keep revising our prior assumptions until eventually the facts converge on the real probability.
Since we cannot do experiments on the future, prediction is wholly a Bayesian exercise. This is precisely why the Intergovernmental Panel on Climate Change produces new assessments every six years or so, since new data and improved theory allow us to update our prior assumptions and increase our confidence in the projected conclusions.
That confidence still falls short of certainty for most aspects of the problem. For example, there is only maybe a fifty-fifty chance of sea levels rising many metres in centuries to come. The conclusion cannot be objective, since the future is yet to come. However, we can use current measurements of ice sheet melting. We can compare them with 125,000 years ago, when the Earth was a degree or two warmer than now and sea levels were four to six metres (thirteen to twenty feet) higher. Because that ancient natural warming had a different cause (changed orbital dynamics of Earth around the Sun) from recent and near future warming caused primarily from current anthropogenic greenhouse gas increases, we can’t say with high
confidence that a few degrees of warming from greenhouse gases will also cause a four-to-six-metre rise in sea levels. But it undoubtedly indicates an uncomfortable Bayesian probability of something similar to that happening in the next few centuries. This indeed was the conclusion of the Synthesis Report of the IPCC’s Fourth Assessment in 2007, for exactly those reasons.
Some statisticians and scientists are leery of Bayesian methods. They prefer to stick only with empirical data and well-validated models. But what do you do when you don’t have such data? One example is found in clinical trials in cancer treatments, a subject in which I have had a very personal interest. The ‘gold standard’ is a double-blind trial where half the patients receive a placebo and the other half receive the drug being tested, and neither the patients nor the researchers know who got what. After five or ten years, if there is a statistically significant difference between the recovery rate of drug and placebo, the trial is declared successful. The trial isn’t designed to pinpoint individual differences. Even if we knew the odds of recovery for the average person from different treatments, there is a wide spread in individual responses. So medicine should try to tailor treatments to the individual’s idiosyncrasies. That makes some doctors – and many insurance companies – nervous. Likewise, some scientists and many policy-makers are nervous about Bayesian inferences based on the best assessment of experts, preferring hard statistics. But as there are no hard statistics on the future, Bayesian methods are all we have. They are certainly better than no assessment at all and hoping that everything will work out fine with no treatment. If we care about the future, we have to learn to engage with subjective analyses and updating – there is no alternative other than to wait for Laboratory Earth to perform the experiment for us, with all living things on the planet along for the ride.
While we have refined our models, it has also taken decades to develop the right approach to these scientific realities, and to find the language to
convey them properly to policy-makers. In the global climate policy discussion, the most important assessments have been produced by the Intergovernmental Panel on Climate Change, in an extraordinary exercise which involves thousands of scientists reviewing the latest evidence. Ever since the IPCC was founded in 1988, I have pushed hard for a cultural change in the assessments. As I have said, overcoming uncertainties, the traditional approach of what the philosopher Thomas Kuhn
7
called ‘normal science’, will take an unforeseeably long time. Climate systems science demands a shift to managing uncertainties instead.
That means we scientists, and policy-makers, grappling with climate change impacts are dealing with risk management. As the sea level rise example indicates, outcomes cannot be assessed with high confidence in many important cases, but the probable range can often be estimated.
Risk-management framing is a judgment about acceptable and unacceptable risks. That makes it a value judgment. As with the Bayesian approach to probability, many traditional scientists are uncomfortable with that. I am one of them, but I am more uncomfortable ignoring the problems altogether because they don’t fit neatly into our paradigm of ‘objective’ falsifiable research based on already known empirical data.
Systems science also alerts us to the possibility of ‘surprises’ in future global climate – perhaps extreme outcomes or tipping points which lead to unusually rapid changes of state. By definition, very little in climate science is more uncertain than the possibility of ‘surprises’. But it is nevertheless a real one. Even so, it took several long rounds of assessment just to get IPCC to mention surprises, let alone discuss formal subjective probabilistic treatment of such potentially irreversible, large changes.