Uncertainty and Climate Risk Assessment
March 22nd, 2007 <-- by Michael Mastrandrea -->
Risk is often thought of as the product of consequences and likelihood—what can happen, and what are the odds of it happening. Both of these factors are important in determining whether and how we address specific risks. For example, even though the consequences of an asteroid colliding with the Earth would be catastrophic, the likelihood of it happening is extremely low in a time frame relevant to human society, and therefore, while we pay attention to the possibility, we generally focus on other concerns that may have fewer consequences, but have a much higher likelihood of occurring.
Projections of future climate change and climate impacts are inherently uncertain. To be clear, the question is not whether the climate will continue to warm, driven by greenhouse gas emissions from human activities. The overwhelming scientific consensus is that temperatures will continue to rise globally as a result of greenhouse gas emissions. Rather, the question is how much the climate will warm, how fast, and what the impacts will be. In very general terms, climate policy is about managing risk: assessing the potential impacts of climate change, judging how likely it is that various impacts will occur, and determining how our policy choices (discussed broadly here) will affect those risks. Uncertainty is a critical factor in assessing both climate risks and the effectiveness of different policy strategies.
There are two general sources of uncertainty in projecting future climate change: what we do and how the natural climate system responds. Uncertainty over what we do incorporates a broad range of social factors: possible trajectories for economic development, population growth, utilization of carbon-free energy sources, and other societal factors that affect greenhouse gas emissions. These sources of uncertainty become especially important beyond the next few decades.
The second component of uncertainty is the response of the climate system to increasing greenhouse gas concentrations—how much the climate will warm, how that warming will affect other processes like rainfall patterns and the natural uptake of carbon by the ocean and growing vegetation, how changes will be distributed across different regions, and so on. Even if we could predict exactly how emissions will play out in the future, we would still be faced with a range of possible climate change and a range for the severity of impacts.
These ranges, however, are dependent on the level of greenhouse gas emissions. Policy decisions can strongly influence the first source of uncertainty (future emissions), but will have little influence on the second source (climate response to emissions). We cannot know precisely what the severity of impacts will be for a specific trajectory for future emissions, but we can confidently say that the severity will be reduced if emissions are reduced. Policy makers must weigh the pitfalls of decision making under uncertainty. Extensively committing resources early may prove excessive, if climate change falls at the low end of the range. On the other hand, delaying or modestly committing resources may fail to prevent severe impacts, if climate change falls at the high end. A further complicating factor is that greenhouse gases emitted into the atmosphere can remain there for decades or even centuries, depending on the type of greenhouse gas. Therefore, simply waiting to see if impacts will be severe before making a decision will lock in even more severe impacts even if immediate action is taken at that point.
At an international level, a risk management approach is exemplified by the United Nations Framework Convention on Climate Change (UNFCCC), drafted in 1992 at the Earth Summit in
The authors of the UNFCCC only provided general guidelines for determining what “dangerous” means in this context. Article 2 states that, “Such a level should be achieved within a time-frame sufficient to allow ecosystems to adapt naturally to climate change, to ensure that food production is not threatened and to enable economic development to proceed in a sustainable manner.” But these are not the only criteria that can be used, nor do they suggest a single answer. Clearly, different people in different parts of the world, with different values and vulnerability will view different levels or types of climate impacts as “dangerous.” In future posts, we will explore the implications of the UNFCCC goal and how it relates to other climate policy efforts, as well as looking more closely at the implications of uncertainty for climate policy and methods for assessing climate risks.
March 22nd, 2007 at 11:05 pm
I am interested in the topic of uncertainties. In my specialty of weather forecasting, I note an increasing interest in the part of the meteorological community to, in an understandable fashion, communicate the uncertainties of a forecast to the intended audience.
I recently wrote about ensemble forecasting in our local newspaper, and how if the solutions converge, that suggests a degree of confidence in the forecast, and if the members of the ensemble diverged, the forecaster would have little confidence in selecting one solution over the others. There, of course, is the further complication that an ensemble of forecasts run on one of the models may not behave similarly or converge on the same solution as an ensemble run on another model. That introduces yet more uncertainty into the forecast. I also suggested that on the 5 day scale of a day to day forecast, small errors in our understanding of how some processes work (such as clouds, and their radiative properties) might not be very noticeable, but the same lack of knowledge on the century scale of a climate change model may be huge. I related it to a bank savings deposit that is made, and then left alone. The interest earned on the deposit in the first 5 days is miniscule, but over the larger part of a century, would become very large.
I recently presented a talk at the Annual Meeting at the 14th Symposium on Meteorological Observation and Instrumentation concerning inadvertant inaccuracies that enter the climate record due to varying platforms, a lack of standardization of what is defined as a max/min temperature (some platforms take a 5 minute average of 5 consecutive 1 minute average temperatures using a 10 secon sampling rate while others are taking the highest instantaneous reading based upon a 16 second sampling rate), a lack of standardization of sampling frequencies and response times. I found that this could reduce the observed daily temperature range by as much as 3F. Also, errors entered in due to the protocols of moving a measurement from the sensor to the display screen. My recorded presentation can be found at:
http://ams.confex.com/ams/87ANNUAL/techprogram/paper_114918.htm
I see that the current BAMS had excellent articles regarding the observation of the properties of low, thin liquid clouds and high, thin ice crystal clouds. It is clear that a great deal of uncertainty exists with these topics observationally, and certainly presents problems in trying to correctly parameterize these processes into the climate models.
Clearly, these are all examples of clear uncertainties in observations, instrumentation, definitions, and model physics and parameterizations.
Since I am in front of a television audience, how can I really understand the actual magnitude of the uncertainties inherent in a climate change forecast, how does this differ from what is presented as the range of possible scenerios presented by the IPCC, and how can I intelligently discuss this, and differentiate what is meant by “consensus” vs “uncertainty”?
[Response: You ask a good and very important question about uncertainty in climate projections. The first important point is that they are fundamentally different than weather forecasts. While the small-scale and short-term details of local weather cannot be predicted beyond ten days or so as I’m sure you well know, large-scale climate changes are not primarily random on decadal to century scales. We can say, for instance, that on average, winter in the Northern Hemisphere will be about 15ºC colder than summer, and about 5ºC colder in the Southern Hemisphere where the ocean moderates temperature to a greater extent. We can be confident that this will generally be true 50 or 100 years from now, because this is a direct response to the Earth’s orbit around the sun. Others have treated this question of why climate projections are possible in far more detail than I can, for instance here.
As I mentioned in my post, the key sources of uncertainty in climate projections are the level of future emissions and the response of the climate to those emissions (specifically, the increased atmospheric concentrations they induce). A key parameter is the “climate sensitivity,” or how much the climate will warm for a given increase in atmospheric greenhouse gas concentrations. The “consensus” in this case is that the climate will warm, and the “uncertainty” is how much it will warm. To use your bank account analogy, uncertainty in the climate sensitivity is somewhat like uncertainty in the interest rate, with temperature being the account balance. If we knew the interest rate exactly, we could project how much our balance would grow. But uncertainty in the interest rate produces a range of possible growth rates, and a range of possible values, say, for the balance in 50 years (like the range of projections from different climate models running the same emissions scenario). Where this analogy breaks down is that if we stopped adding greenhouse gas emissions to the atmosphere (stabilizing concentrations), the balance (temperature) would not grow indefinitely, it would also stabilize, so perhaps we need to think of a declining interest rate.
A general rule of thumb is that the uncertainty in human-induced emissions of greenhouse gases is roughly equal to the uncertainty in the response of the system. This of course will change over time, and in my view is not quite an equivalent comparison, since we generally have control of the level of emissions in the long-term and therefore can determine what this component will be. But one simple way to separate these sources of uncertainty is to look at the IPCC temperature projections for different scenarios, available here. The range of temperature increase from different climate models for 2100 for the lowest IPCC scenario used, B1, is 1.1-2.9ºC (this is not the lowest possible scenario, of course). The range for the highest, A1FI, is 2.4-6.4ºC. In 2100, CO2 concentrations for A1FI are about double concentrations for B1, and the lower and upper bounds of these ranges have roughly the same ratio. -Mike Mastrandrea]
March 24th, 2007 at 10:05 am
I’d like to thank Dr. Mastrandrea for his kind reply. I’m wondering about the impact of the uncertainties that I mentioned (uncertainties in describing the role of certain processes such as clouds and aerosols either explicitly or by parameterization in the models, as well as the uncertainties that I refered to observationally concerning these proceses as described in the BAMS articles on thin liquid and ice clouds, and my talk on temperature measurements among differing platforms) on the results of the models.
[Response: Identifying and reducing uncertainty is a two step process. First scientists work to accurately characterize the full range of possible responses. Then they work to further constrain that range (that is to reduce the window of uncertainty surrounding the possible range of responses). For example, in estimating climate sensitivity, the key is to identify all the plausible positive and negative feedbacks that could accompany increasing greenhouse gas concentrations. If you assume the maximum negative feedbacks and the minimum positive feedbacks, you get the low end of the plausible range. If you assume the minimum negative feedbacks and the maximum positive feedbacks, you get the high end of the range. The next step is to try to better understand the potential range of feedbacks so that you can bring the high and low estimates closer together. Data is helpful for this. So are past ice age/interglacial cycles because we have a good idea of how much the temperature changed along with how much greenhouse gas concentrations and the ice sheets changed. -phiggins]
The range of numbers arising from the suite of models such as B1 and A1FI suggest the uncertainties involved using the assumptions that are built into these models. My question of how to grasp the “real” uncertainty concerns how the central values, and range about the central values might change as we eventually include items such as clouds more confidently in the models. How real are the present ranges? Do we have an idea of the magnitude, and in what direction, that the added knowledge would change these values and ranges?
[Response: Again, the key is not to focus on just the best guess of climate sensitivity (or emission scenario, or roll of clouds and aerosols), but to examine and constrain the whole range of each. For the temperature ranges projected for the end of the century, roughly half of the uncertainty is due to the physical climate system (how much warming occurs for a given change in greenhouse gas concentrations) and half due to how much we emit. If we had better understanding of how clouds respond, we would constrain the range. We cannot say whether it would raise the lower end or lower the upper end. The safest bet would be a little of each.
I should also point out that the possibility always exists that we’ll find a new feedback (positive or negative) that we haven’t accounted for. That would shift the range of plausible outcomes. We should view projected ranges as a methodical examination of what we know and understand. A great deal of effort goes in to that and it accounts for our considerable understanding of the climate system. But it is neither an infallible assessment nor a blind guess. Of course, from a policy perspective the range of uncertainty (and the possibility we’re missing something) cuts both ways: the problem could end up being better or worse than we might expect.
Thanks for these terrific questions. Please let me know if any of explanation remains unclear and I’ll take another stab at it. -phiggins]
March 26th, 2007 at 4:44 pm
While Al Gore got all the press and face time when he testified before Congress last week, the truly important testimony came from Bjorn Lomborg after Al left the building. Mr. Lomborg, who accepts the consensus opinion of the science without question, has been arguing for years that carbon mitigation programs are a huge waste of resources, and that the world would be much better off 100 years from now, if we concentrated on solving existing problems today.
[Response: Lomborg would appear to misunderstand fairly basic economics then. The reason is that individuals who emit greenhouse gases generally receive all the benefits for their polluting activities but don’t pay the costs of climate change. Instead, those costs are distributed among everyone. As a consequence, the damages of climate change do not directly influence incentives of polluters even though those costs still affect the overall economy. In short, that translates into an economically damaging subsidy to pollute. Reducing that subsidy through some form of pollution fee would actually tend to benefit the economy as a whole.
Lomborg’s point about spending limited resources on climate change is correct: such spending may, or may not, be better allocated to other programs. But that argument doesn’t hold for mitigation efforts that seek to reduce (or eliminate) the emission subsidy by bringing the social costs of climate pollution into the price paid by those who pollute. That can be expected to benefit the economy.
I can’t quite agree with you that Lomborg accepts the consensus scientific opinion. His book generally confuses and misrepresents the science of climate change, in my view. The main reason for that, I believe, is that Lomborg tends to give more weight to and be less skeptical of a relatively small group of unrepresentative papers than most climate scientists. -phiggins]
We do not have to speculate on the effectiveness of the type of carbon mitigation programs that are proposed by many of the advocates of the issue. A recent analysis of the effectiveness of the Kyoto Protocol revealed that the European Union has spent about 300 billion dollars trying to implement the treaty, resulting in a possible net reduction in global temperature of 0.003 of a degree C by 2050! That makes the Pentagon’s purchase of $1,000 toilet seats look like a smart buy! We at least have something to show for the toilet seat purchase.
Even if we could increase the efficiency of these carbon mitigation programs, there will never be a way to quantify their effectiveness! Not only do you have the two aspects of uncertainty that Dr. Mastrandea explained above, but you add the additional uncertainty of what problems could have been addressed with the resources that went to carbon mitigation. The question whether or not carbon mitigation programs will ever be ‘worth it’ will always have a theoretical answer. Since there are no definable bench marks of success, the definition of success will be defined by the prevailing political view at the time. I am 95% certain that this will lead to a huge bureaucracy, wielding large amounts of power and influence, while generating more harm than good. (The evidence for this claim will come out in a few months, as soon as I can make sure that all the evidence I have collected agrees with the above statement:-))
[Response: Again, it is incorrect to equate all mitigation options with the spending of limited resources. Because of the pollution subsidies (described above) one can even say the opposite: by not accounting for the social costs of pollution we are wasting limiting resources by encouraging pollution when we could instead be allocating those resources to more beneficial programs. That said, you’re right that we can never fully quantify policy effectiveness (or the full magnitude of the pollution subsidies that I describe), which brings us back to decision making under uncertainty, the topic of the post. Unfortunately, that is probably true for almost all policies. -phiggins]
The alternative to carbon mitigation policy is to address problems that already exist, but would likely become much worse in a warming world. We need to improve susceptibility to extreme weather events by building smarter and planning communities better. We need to develop more efficient uses of the global water supply. We need to prevent the spread of disease with improved sanitation in the third world, and the proper use of pesticides. We need to empower the global economy by promoting stable, representative governments around the world, promoting the spread of (information) technology and keeping the global markets free to operate under a reasonable rule of law. And, yes, we need to develop non-carbon based fuels and become more energy efficient overall, not because it will make the climate stable, but because oil will become more expensive as it becomes harder to find.
In other words, we need to move the emphasis away from carbon mitigation to developing policy that addresses problems that already exist, but could become worse with future warming. These ‘adaptation’ policies would result in immediate definable benefits that will continue to grow into the future. We can target these policies to the specific areas where they are needed, as all climate change is essentially local. We will be able to rapidly adjust and adapt these policies as technologies improve and global situations change. We will also be able to tell if the policy is effective in a real world sense, and adjust it accordingly; advantages we will not have with carbon mitigation policies.
The real advantage of adaptation policies are that they provide a benefit no matter what future climate change brings. If it turns out that the net effect of a doubling of atmospheric CO2 is one degree C or less, Europe has already wasted $300 billion that it can scarcely afford, while accomplishing absolutely nothing!
So why does carbon mitigation seem to be the only game in town? Because carbon mitigation policies empower activists, governments, the UN and, to some extent, the atmospheric science community. Adaptation policies, on the other hand, address the issues without forwarding the ideologies of environmental activists, empowering the United Nations, adding additional powers to governments or mandating a never ending call for a ‘better understanding of climate change’!
[Response: Building adaptive capacity makes sense as does a full discussion of all of society’s options, which we began here. -phiggins]
I would like to call upon the AMS to start emphasizing adaptation policies over carbon mitigation policies when offering counsel to elected officials. No amount of carbon mitigation will save New Orleans from a repeat hurricane disaster, but adaptation policies certainly could. The AMS, and the scientific community in general, will carry much more credibility into the future if we can show that our recommendations have resulted in significant benefits; something that we will never be able to do solely with carbon mitigation policy.
Finally, if we put all of our eggs in the ‘carbon mitigation’ basket, and the Earth actually starts to cool in the next 20 years, who will ever take us seriously again?
April 13th, 2007 at 3:31 pm
Thanks to Dr. Mastrandrea and Higgins for their informative replies to my questions dealing with uncertainties. While I still find myself feeling that the degree of confidence attached to the quantification of the range of possible scenerios is perhaps too high, the statement by Dr. Higgins that “We should view the projected ranges as a methodical examination of what we know and understand. … But it is neither an infallible assessment nor a blind guess.” frames the picture well.
I found the following paper useful:
http://blue.atmos.colostate.edu/publications/pdf/maccracken2002.pdf
Do the changes made to how the range of scenerios were calculated in the 2001 assessment apply to the 2007 edition? Is there an easy way to express what the current models treat in a more thorough or explicit manner than were available in the 2001 report?