For over a decade, future climate change impacts have been assessed using a number of standardised emission scenarios. These scenarios assume different future states of the world’s population and corresponding greenhouse gas emissions – from really optimistic to really grim. For more information on these scenarios, please see IPCC SRES Scenarios and RCP Scenarios.
It is important to note that under each of these scenarios, there is a range of uncertainty in climate change impacts under different climate model simulations. This blog entry will not attempt to explore this matter further. However, a second level of uncertainty exists. What are the probabilities of each of the scenario events occurring?
Right from the inception of these future emission scenarios, no probability was assigned to the occurrence of any particular scenario. Over time, these scenarios have been updated but this is only to reflect the increased cumulative understanding of emissions and atmospheric forcings.
It seems that gone are the days where policymakers and decision-makers are rejecting climate change as a doom and gloom prophecy, at least in most cases. Impacts of climate change are now being considered to such an extent that a whole new field of “Climate Services” has been established. This generally involves climate “specialists” collaboratively informing policymakers and decision-makers of expected climate change and to make relative vulnerability assessments.
These policymakers and decision-makers are already largely apprehensive of the uncertainty of climate change under any given scenario (Moss et al., 2010). This can only be compounded by the additional complexity of different emission scenarios. Surely it would be beneficial and helpful to understand the probabilities of the particular scenario occurrence? If this were done, could it not better inform decisions?
Furthermore, the assignment of probabilities would lead to a time evolution of climate change risk. By revising our estimates of the probability of scenarios as we progress, the accuracy of climate change forecasts would surely improve which could lead to better adaptation and mitigation strategies. In contrast, by not apportioning probability to the scenarios, we are left basing our decisions on a static state of understanding where decisions made will not be improved with time.
It seems that the general consensus amongst CSAG staff is that we are headed on the RCP8.5 route (worst case scenario). One of the CSAG staff members informed me that he found no supporting evidence of this track in the literature. Risbey (2004) admits that a few studies have identified the lack of scenario probability assignment but this has not been adequately addressed as of yet. As mentioned earlier, these scenarios are not new to the discipline of climate change. Therefore, it would appear almost irresponsible that no one has followed it up.
I am of the strong opinion that the lack of probability assignment to the various scenarios provides a major gap in climate change research and limits the potential for climate change risk management. Without the understanding of where we are headed, I feel that Climate Services are hugely undermined.
Risbey, J.S. 2004. Agency and the Assignment of Probabilities to Greenhouse Emissions Scenarios. Climate Change. 67(1): 37-42.
Moss, R.H, Edmonds, J.A., Hibbard, K.A., Manning, M.R., Rose, S.K., van Vuuren, D.P., Carter , T.R., Emori, S. et al. 2010. The next generation of scenarios for climate change research and assessment. Nature. 463: 747-756.
A thought-provoking blog Dean. My question is in tandem with Claire’s. Dean, while it is true that assigning probabilities to scenarios may give the user group a more robust information, what happens when the reality occurs outside the assigned probabilities. I also think, that assigning probabilities would make concrete adaptation more difficult because it is bound to different interpretation and meaning by decision makers and user community.
It seems like there has been more comment than blog! An excellent assessment of the power of a few words to stimulate discussion. While I will not comment on probability requirements (or not) in scenarios, I would like to ask rather why are we now seeing costs and benefits of scenarios. Would this surely not allow us to see whether a certain scenario closer to being relevant/practical or not?
I am on the run so just a quick few further comments:
I feel as though Chris and Joseph provide very valid arguments which further elucidate the challenges and perceived low value (based on expected benefit obtainable) in assigning probabilities to the scenario emissions. But a further two comments come to mind based on these responses
In the comment by Chris “around 90% of decisions that are actually being made, at local scales, with respect to climate, are short term decisions (5 to 10 year time horizons) that speak more into climate variability, risk of extremes, and the intersection of trending non-climatic stressors with various aspects of climate. Longer term climate projections are seldom actually needed!”
Does this mean we can neglect the consideration for the other 10%. Maybe less decisions are made for the longer term but this doesn’t say anything about the importance (cost if the forecasts wrong) of those decisions. Maybe a “Power Rule” exists here. Things that immediately come to mind are dam construction and urban planning (e.g storm-water drainage).The relative number of decisions made may be few but the consequences could be extremely large. That is on the hopeful assumption that dam construction and urban planning do consider longer term projections. Therefore, is it reasonable to suggest that time is better spent on understanding shorter term sources of uncertainty?
In the comment by Joseph: “No matter whether or not scientists calculate and communicate explicit probabilities for emissions scenarios, we all have some sense of what is more or less likely.” This brings me to my original interest in writing this blog. “Scientific” probability aside; DO we all have some sense of what is more or less likely? If so, what is this based on? scientific or statistical analysis, a visual inspection of trend, an educated guess, gut feeling? Is this sense of what is more likely reflected in the literature? Where are we headed?
Thank you to everyone for your insightful comments. I for one am already obtaining a better understanding of this issue as well as other important climate change concepts. Please keep them coming
Time for me to add my two cents…
When I first read the blog, my initial thoughts were in line with the questions raised by Bruce. There is no way to create objective probabilities about emissions scenarios, just as there is no way to create objective probabilities about climate model projections. Reality and our models of reality (especially in the context of the climate system) are simply not close enough. But in relation to Bruce’s first question, “can we assign probabilities?”, the answer is yes. How we assign probabilities is a different question. In situations of deep uncertainty, often expert elicitation is used to derive some sort of probabilities. Of course the probabilities are conditional but I imagine both Bruce and I would agree that it is more “probable” that greenhouse gas emissions will increase over the next 20 years than decrease. We might argue about quite how much they will increase but we can come to some approximate agreement about relative likelihood and maybe even agree on “imprecise probabilities” (to use terminology that is floating around the climate-decision literature).
No matter whether or not scientists calculate and communicate explicit probabilities for emissions scenarios, we all have some sense of what is more or less likely. This is undoubtedly affected by our world view and our knowledge of climate science. Perhaps it is best to leave it like this. Let people interpret the probabilities for themselves based on their level of knowledge and world view. However, this sounds a bit lazy too and not particularly progressive. We might be tempted to go down the route of “ruling out” scenarios, just as we do in climate modelling when we see a model that does something absurd. Maybe my previous blog on climate “possibilities” is relevant here?
However, in all of this we must be careful about the scientists role in providing “information” and I come to your comment Dean that says “By assigning probabilities to scenarios, you only give decision makers more information which could lead to more informed decisions”. More informed decisions? Well if more informed decisions is the same as decisions made having been given more information, then I guess one can’t disagree. But the notion of an informed decision implies that the information feeding the decision is reliable and relevant. I am not sure that conditional probabilities on emissions scenarios could ever be considered reliable. And then in relation to Chris’ comments about the time horizon and nature of climate services, I guess emissions scenarios aren’t all that relevant anyway so perhaps time is better spent on other sources of uncertainty relevant on shorter time horizons.
A very interesting debate. No doubt this will continue…
Interesting discussion! My response is largely focussed on the idea that climate services are “undermined” by the lack of, or inability to assign, probabilities to different projections or scenarios. Indeed, if “climate services” is viewed (as it probably is) as providing a collection of data points about future climate states, then it probably is being undermined by the probability issue.
But my growing experience in climate services is leading me to the view that its not primarily about providing future data points. Its primarily about engaging with a particular context and having a discussion about if and how climate is relevant to that context. My impression is that around 90% of decisions that are actually being made, at local scales, with respect to climate, are short term decisions (5 to 10 year time horizons) that speak more into climate variability, risk of extremes, and the intersection of trending non-climatic stressors with various aspects of climate. Longer term climate projections are seldom actually needed!
The value of longer term projections (40 to 100 years) is largely around creating awareness that society should not bet on the status quo remaining the status quo indefinitely and to explore some of the “possibility space” of what a future climate could look like. In reality, decision making processes are dominated and constrained by so many other factors (funding cycles, election cycles, conflicting interests, legal processes) that its a pipe dream to think that many decisions are actually being made based on hard quantitative, regional scale, projections, let alone any measure of probabilities.
Our hard work producing ensemble projections of future climate down to the 50km scale generally gets mangled, reformed, simplified, over-interpreted, under-interpreted, painted with glossy paint and laid on the desk of some policy maker in the form of an executive summary containing sentences like: “The western cape may be drier in the future”.
That may all sound very cynical and I suppose it is. I do still believe there is real value in producing the projections we produce. We need to keep exploring that possibility space, understanding the regional climate dynamics, feedbacks and uncertainties. There is useful information there that needs to make its way into policy and decision making. But we need to understand the nature of that decision making and be realistic about what we are actually doing.
There is lots of work and thinking to be done on this subject but I think the assigning of probabilities to particular scenarios is the least of our concerns!
In Response to Claire:
Based on my comments above, I agree with the difficulty if not impossibility of actually assigning probabilities to the scenarios.
“But let’s say probabilities could be calculated; wouldn’t such knowledge just lead policymakers to the most obvious answer? That is, adapt for the scenario with the highest probability.” This brings up another very important concept. Decision making under uncertainty . It is the sad truth that people don’t always make the most rational decisions in the face of incomplete information. However, this problem is not unique to climate change risk management. A whole field is dedicated to looking at how people make decisions called game theory.
By assigning probabilities to scenarios, you only give decision makers more information which could lead to more informed decisions. How people use that information to make decisions is a completely different problem and comes down to further concepts such as risk appetite. However, assuming probabilities could be assigned to scenarios, as in your argument, I think it would be both unethical and unjustified to withhold such information on the premise of getting people to act in a particular way (i.e to not focus on the most likely scenario)
The questions that Bruce raises are very valid and highlight just some of the complexities involved in potentially assigning probabilities to the scenarios. This is presumably why no such assignments have been made as of yet.
Bruce’s first question about conditional probability is a powerful one that I am in full agreement with. However, it raises an even more fundamental and troubling question. Are the forecasts of climate change not conditional on the scenarios implemented in the climate models? Therefore, if we cannot assign probabilities to the scenarios, how can we assign any probability to future climate change projections? Aren’t the forecasts therefore to some degree meaningless other than to show what could happen in small sample of infinite events?
My answer to Bruce’s second question, based on my comments of the first question, is a definite YES. Bruce also comments that “Would that not mean people discount / ignore perceived low probability scenarios.” I think that by basing climate forecasts on the current scenarios, climate analysts themselves are ignoring low probability scenarios. Why is there no scenario that suggests long-term global cooling? However unlikely, it is still a possible scenario (albeit not in the short-term due to the already committed climate warming). Therefore, on creating the current scenarios used in practice, some form of probability assumptions have already occurred. Is this not a bias in itself? Based on Bruce’s argument, should this hypothetical climate cooling scenario be used to inform decisions with no attribution of probability? Should people perceive such a scenario with equal consideration?
I have no rebuttal to Bruce’s third question and agree that this is a major challenge.It would be interesting to see if Social Physics applications could come into this as discussed by Philip Ball (2006) in “Critical Mass”.
In response to Gemma’s question, “to what extent could an understanding of probabilities of a particular scenario better inform decision-making?”. The probability of a particular scenario is synonymous with the probability of the projected climate change distribution of the given scenario. Therefore if we knew that that the RCP8.5 scenario was significantly more probable than the RCP4.5 scenario, decision-makers could use the RCP8.5 projections with more weight (but not neglecting the other scenarios) to better inform the adaptation and mitigation strategies. Consider this hypothetical scenario projections for a region in some future period. : There is a chance that rainfall intensity will increase by 2-4% under the RCP4.5 Scenario. There is also a chance that rainfall intensity will increase by 15-20% under the RCP8.5 scenario. (These are the types claims that can currently be made). But what if I said that there was a 20% probability that rainfall will increase by 2-4% but an 80% chance that rainfall intensity will increase by 15-20%. Surely you can make a more informed decision based on the second set of information?
So at this point, I strongly agree that the assignment of true probability is not possible, but I still firmly believe that by not having scenario probabilities, our climate change projections and consequently the decisions based on them, are severely limited
I like Dean’s argument as a thought experiment, but realistically I do not think it is possible to predict future scenarios, as well as the probability of each of them occurring. But let’s say probabilities could be calculated; wouldn’t such knowledge just lead policymakers to the most obvious answer? That is, adapt for the scenario with the highest probability. Given the uncertainties in climate models, given the (what seems to be an endless) list of uncertainty in modelling climate change, how close to the truth would further extrapolation bring one? How close to reality would those probabilities actually be? I think putting a probability on any particular scenario would be a recipe for maladaptation, in that the scenario with the highest probability would be considered by policymakers and no other scenario. What if things don’t go as expected, what if the anomaly occurs and the scenario with the least probability becomes reality? I like stats, stats can do amazing things, but I would not trust stats in the hands of someone who is looking for a definitive answer to a question that is based on so many variables – known and unknown to us.
I agree with the three questions that Bruce raises. My question for each of those, however, is who decides what the answers are for each of those? Is it government or the public, or both? Furthermore, I feel that the 3rd question raised is a fundamental one – especially in the politically correct state that society currently finds itself. Which world view is most appropriate, or the most accurate?
The 3 questions identified by Bruce are particularly relevant for such questions as “Surely it would be beneficial and helpful to understand the probabilities of the particular scenario occurrence? If this were done, could it not better inform decisions?” Which leads me to my own question: to what extent could an understanding of probabilities of a particular scenario better inform decision-making?
Interesting topic though.
This is an interesting question that raises three further questions / thoughts in my mind.
1. Can probabilities be assigned? Surely only conditional probabilities are possible … probabilities if there’s no asteroid strike, the middle east doesn’t blow up, and we don’t experience global bee colony collapse disorder. Since each of these have their own probabilities, it’s hard to see how to build a scenario probability.
2. Should we assign probabilities? Would that not mean people discount / ignore perceived low probability scenarios? As an analogy, should we ignore the low probability of a major asteroid strike in the next 50 years? (we’re statistically overdue for one).
3. Probability of emission scenario is inherently a function of one’s world view of human nature. An optimist, a pessimist, an American, an Indian, a Nigerian, will all come to different probability conclusions. So whose probability do we work with?