3
HYDROLOGICAL PROCESSES Hydrol. Process. 21, 1127 – 1129 (2007) Published online 6 March 2007 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/hyp.6573 Probabilistic climate scenarios may misrepresent uncertainty and lead to bad adaptation decisions Jim Hall* Tyndall Centre for Climate Change Research, School of Civil Engineering and Geosciences, Newcastle University, UK *Correspondence to: Jim Hall, Tyndall Centre for Climate Change Research, School of Civil Engineering and Geosciences, Newcastle University, UK. E-mail: [email protected] Received 31 August 2006 Accepted 5 September 2006 Development of climate scenarios at global and regional scales is set upon a course towards probabilistic representations of uncertainty. The motive for this probabilistic approach is clear. The presentation of scenarios without quantification of relative likelihood, as typified by the Intergovernmental Panel on Climate Change’s (IPPC) Third Assessment Report (TAR) (Houghton et al., 2001), has been criticized as providing policy makers with insufficient information upon which to base decisions (Pittock et al., 2001; Schneider, 2001, 2002). If probabilities are not provided then, it is argued, decisions will be made with implied assessments of relative likelihood that depart, perhaps significantly, from experts’ best judgement. A rational approach to dealing with climate risks will, it is argued, weigh up probabilities and consequences. This applies to adaptation decisions, e.g. in relation to water resource management, as much as it applies to the question of mitigation policy, where it has been most actively discussed. Considerable progress has been made since the TAR in exploring the parameter space of general circulation models (GCMs) of the global climate and in conditioning model outputs with climate observations and paleoclimate data. It is therefore argued that the climate community is ready to provide probabilistic climate projections, which will be much more informative than the results published in the TAR. The UK Climate Impacts Programme (UKCIP) is committed to publishing probabilistic distributions in its next climate scenarios for the United Kingdom, which will supersede the four (Low, Medium Low, Medium High, High) scenarios published by UKCIP in 2002 (Hulme et al., 2002). As with global probabilistic scenarios, these forthcoming UKCIP scenarios will contain a great deal more information than the previous four discrete projections, which were presented without any indication of their relative likelihood, and were based upon only one GCM. But are these probabilistic scenarios in danger of providing more information than is actually warranted by the available evidence? While the motive for probabilistic scenarios is clear, there are several concerns that accompany their use for impact assessments and adaptation studies. Incompleteness Probabilistic scenarios can only represent a fraction of the total uncer- tainties in climate projections. They are, on the whole, presented as being conditional upon a given emissions scenario, which is appropriate if the aim is to inform mitigation policy but is less helpful for adaptation deci- sion makers. Moreover, probabilistic scenarios have been developed on the basis of relatively small numbers of GCM integrations (Tebaldi et al., 2004), sometimes only from one GCM (Murphy et al., 2004). An excep- tion to the former criticism is the work of Stainforth et al. (2005), who have been able to do several thousand model runs, but using only one version of a Hadley Centre model. The number of RCM runs available for generation of downscaled hydrological projections (see e.g. Fowler et al., 2005) is also quite modest. Given the relatively sparse sampling Copyright 2007 John Wiley & Sons, Ltd. 1127

Probabilistic climate scenarios may misrepresent uncertainty and lead to bad adaptation decisions

Embed Size (px)

Citation preview

HYDROLOGICAL PROCESSESHydrol. Process. 21, 1127–1129 (2007)Published online 6 March 2007 in Wiley InterScience (www.interscience.wiley.com). DOI: 10.1002/hyp.6573

Probabilistic climate scenarios may misrepresent uncertaintyand lead to bad adaptation decisions

Jim Hall*Tyndall Centre for Climate ChangeResearch, School of Civil Engineeringand Geosciences, NewcastleUniversity, UK

*Correspondence to:Jim Hall, Tyndall Centre for ClimateChange Research, School of CivilEngineering and Geosciences,Newcastle University, UK.E-mail: [email protected]

Received 31 August 2006Accepted 5 September 2006

Development of climate scenarios at global and regional scales is setupon a course towards probabilistic representations of uncertainty. Themotive for this probabilistic approach is clear. The presentation ofscenarios without quantification of relative likelihood, as typified by theIntergovernmental Panel on Climate Change’s (IPPC) Third AssessmentReport (TAR) (Houghton et al., 2001), has been criticized as providingpolicy makers with insufficient information upon which to base decisions(Pittock et al., 2001; Schneider, 2001, 2002). If probabilities are notprovided then, it is argued, decisions will be made with impliedassessments of relative likelihood that depart, perhaps significantly,from experts’ best judgement. A rational approach to dealing withclimate risks will, it is argued, weigh up probabilities and consequences.This applies to adaptation decisions, e.g. in relation to water resourcemanagement, as much as it applies to the question of mitigation policy,where it has been most actively discussed.

Considerable progress has been made since the TAR in exploring theparameter space of general circulation models (GCMs) of the globalclimate and in conditioning model outputs with climate observations andpaleoclimate data. It is therefore argued that the climate communityis ready to provide probabilistic climate projections, which will bemuch more informative than the results published in the TAR. TheUK Climate Impacts Programme (UKCIP) is committed to publishingprobabilistic distributions in its next climate scenarios for the UnitedKingdom, which will supersede the four (Low, Medium Low, MediumHigh, High) scenarios published by UKCIP in 2002 (Hulme et al.,2002). As with global probabilistic scenarios, these forthcoming UKCIPscenarios will contain a great deal more information than the previousfour discrete projections, which were presented without any indication oftheir relative likelihood, and were based upon only one GCM. But arethese probabilistic scenarios in danger of providing more informationthan is actually warranted by the available evidence? While the motivefor probabilistic scenarios is clear, there are several concerns thataccompany their use for impact assessments and adaptation studies.

IncompletenessProbabilistic scenarios can only represent a fraction of the total uncer-tainties in climate projections. They are, on the whole, presented as beingconditional upon a given emissions scenario, which is appropriate if theaim is to inform mitigation policy but is less helpful for adaptation deci-sion makers. Moreover, probabilistic scenarios have been developed onthe basis of relatively small numbers of GCM integrations (Tebaldi et al.,2004), sometimes only from one GCM (Murphy et al., 2004). An excep-tion to the former criticism is the work of Stainforth et al. (2005), whohave been able to do several thousand model runs, but using only oneversion of a Hadley Centre model. The number of RCM runs availablefor generation of downscaled hydrological projections (see e.g. Fowleret al., 2005) is also quite modest. Given the relatively sparse sampling

Copyright 2007 John Wiley & Sons, Ltd. 1127

J. HALL

and, more importantly, the small number of mod-els used in these analyses, it is to be expected thattotal uncertainty will be underestimated. More signif-icantly, all of the models employed contain significantsimplifications (e.g. through parameterization of sub-gridscale processes) of the known physics and, more-over, do not represent all of the processes that havebeen postulated as being significant to climate (e.g.some processes of abrupt climate change). The prob-abilistic outputs of analyses based upon these modelsare therefore highly conditional upon the model runsand indeed the statistical method used to computeprobability distributions. This must be made clear todecision makers.

Representation of Severe Uncertainty

There is a long tradition in probability, which datesback to Laplace, that in the absence of informationabout the distribution of probabilities, uniform distri-butions should be adopted. In other words, ignoranceshould be represented with uniform probability dis-tributions. Several thinkers, including Keynes (1921),have demonstrated how indiscriminate use of uniformdistributions can lead to absurd conclusions. Yet uni-form probabilities are still widely used to representsituations of severe uncertainty. For example, in Mur-phy et al. (2004) uncertainties in key parameters wererepresented by uniform distributions over parameterranges obtained from experts, on the grounds thatno further information was available with which toconstruct prior distributions. Conventional probabil-ity theory does not have an adequate representationof complete ignorance.

Bounded Rationality

A further criticism of the probabilistic approachcomes by consideration of how decision makers makeuse of probabilistic information. Flood defence andwater resource engineers, among others, are well prac-ticed at using probabilistic information (for example,relating to weather variables such as rainfall) to opti-mize the expected returns from design decisions. Theirapproach typically involves ‘integrating out’ proba-bility distributions in order to obtain the expectedvalues of performance variables, which then providethe objective function for optimization of, for exam-ple, flood defence crest level or reservoir capacity.Precisely the same approach can be applied to proba-bilistic climate information, which if undertaken with-out proper consideration of the residual uncertaintiesmentioned above, could lead to bad adaptation deci-sions, which prove to be far from optimal. A repre-sentation of uncertainty based upon less strong prob-abilistic assumptions (Kriegler and Held, 2005; Hallet al., 2006) might identify options that perform bet-ter under a wide range of conditions (Lempert et al.,2004; Dessai and Hulme, 2006).

Attitudes to Ambiguity

Ellsberg (1961) demonstrated that decision makersare systematically averse to gambles to which theydo not know the odds. In other words, decision mak-ers are prepared to pay more to take a bet wherethey know the odds are 50 : 50 than they will pay fora bet, with the same two outcomes, where they donot know the odds. Figure 1 shows a number of pub-lished probability distributions for climate sensitivity(the equilibrium global mean temperature increase

1

0.9

0.8

0.7

0.6

0.5

0.4

0.3

0.2

0.1

00 1 2 3 4 5

Climate sensitivity (°C)

6

Forest et al. (2002): uniform priorForest et al. (2002): expert prior

Knutti et al. (2002)Tol & de Vos (1998)Murphy et al. (2004): unweighted

Murphy et al. (2004): weighted

Andronova & Schlesinger (2001): GTAS

7 8 9 10

Cum

ulat

ive

prob

abili

ty

Figure 1. Some published cumulative probability distributions for climate sensitivity

Copyright 2007 John Wiley & Sons, Ltd. 1128 Hydrol. Process. 21, 1127–1129 (2007)DOI: 10.1002/hyp

INVITED COMMENTARY

from doubling atmospheric CO2 concentrations rela-tive to preindustrial levels). It is clear not only thatthe response of the climate to radiative forcing isuncertain, but that the scientific community is notagreed on how uncertain it is. To reduce this uncer-tainty to a single probability distribution in order togenerate probabilistic scenarios would seem to mis-represent the state of current scientific knowledge. Ifdecision makers are not presented with informationupon the severe ambiguities in probabilistic climateinformation, then they will not understand the uncer-tainties and they will not be able to take decisionsthat reflect their legitimate aversion to ambiguities.It has been suggested (Kriegler et al., 2006) that theprecautionary principle is an expression of ambiguityaversion. Given full information about uncertainties,decision makers may be inclined to adopt precaution-ary approaches.

The arguments outlined above should not be seenas reason to put a stop to the process of generatingprobabilistic scenarios and using them in assessmentsof the potential impacts of climate change. However,they do indicate that considerable caution is requiredin the interpretation of probability distributions andin their use for adaptation decision making in, e.g. thefield of flood defence or water resource management.

References

Andronova NG, Schlesinger ME. 2001. Objective estimation of theprobability density function for climate sensitivity. Journal of Geo-physical Research-Atmospheres 106(D19): 22605–22611.

Dessai S, Hulme M. 2006. Robust adaptation decisions amid climatechange uncertainties: a case-study on water resources managementin the East of England. Global Environmental Change (in press).

Ellsberg D. 1961. Risk, ambiguity, and the Savage axioms. QuarterlyJournal of Economics 75: 643–669.

Forest CE, Stone PH, Sokolov AP, Allen MR, Webster MD. 2002.Quantifying uncertainties in climate system properties with the useof recent climate observations. Science 295(5552): 113–117.

Fowler HJ, Ekstrom M, Kilsby CG, Jones PD. 2005. New estimatesof future changes in extreme rainfall across the UK using regional

climate model integrations. 1: assessment of control climate. Journalof Hydrology 300(1–4): 212–233.

Hall JW, Fu G, Lawry J. 2006. Imprecise probabilities of climatechange: aggregation of fuzzy scenarios and model uncertainties.Climatic Change. DOI: 10.1007/s10584-006-9175-6.

Houghton JT, Ding Y, Griggs DJ, Noguer M, van der Linden PJ,Dai X, Maskell K, Johnson CA. 2001. Climate Change 2001: TheScientific Basis. Contribution of Working Group I to the ThirdAssessment Report of the Intergovernmental Panel on Climate Change.Cambridge University Press: Cambridge.

Hulme M, Jenkins GJ, Lu X, Turnpenny JR, Mitchell TD, Jones RG,Lowe J, Murphy JM, Hassell D, Boorman P, McDonald R, Hill S.2002. Climate Change Scenarios for the United Kingdom: TheUKCIP02 Scientific Report . Tyndall Centre for Climate ChangeResearch: Norwich.

Keynes JM. 1921. A Treatise on Probability . MacMillan: New York.

Knutti R, Stocker TF, Joos F, Plattner GK. 2002. Constraints onradiative forcing and future climate change from observations andclimate model ensembles. Nature 416(6882): 719–723.

Kriegler E, Held H. 2005. Utilizing belief functions for the estimationof future climate change. International Journal of ApproximateReasoning 39(2–3): 185–209.

Kriegler E, Held H, Bruckner T. 2006. Climate protection strategiesunder ambiguity about catastrophic consequences. In Decision Mak-ing and Risk Management in Sustainability Science, Kropp J, Schef-fran J (eds). Nova Science Publishers: New York.

Lempert R, Nakicenovic N, Sarewitz D, Schlesinger M. 2004. Char-acterizing climate-change uncertainties for decision-makers. ClimaticChange 65(1–2): 1–9.

Murphy JM, Sexton DMH, Barnett DN, Jones GS, Webb MJ,Collins M. 2004. Quantification of modelling uncertainties in a largeensemble of climate change simulations. Nature 430(7001): 768–772.

Pittock AB, Jones RN, Mitchell CD. 2001. Probabilities will help usplan for climate change—Without estimates, engineers and plannerswill have to delay decisions or take a gamble. Nature 413(6853): 249.

Schneider SH. 2001. What is ‘dangerous’ climate change? Nature411(6833): 17–19.

Schneider SH. 2002. Can we estimate the likelihood of climaticchanges at 2100? Climatic Change 52(4): 441–451.

Stainforth DA, Aina T, Christensen C, Collins M, Faull N, Frame DJ,Kettleborough JA, Knight S, Martin A, Murphy JM, Piani C, Sex-ton D, Smith LA, Spicer RA, Thorpe AJ, Allen MR. 2005. Uncer-tainty in predictions of the climate response to rising levels of green-house gases. Nature 433(7024): 403–406.

Tebaldi C, Mearns LO, Nychka D, Smith RL. 2004. Regional prob-abilities of precipitation change: a Bayesian analysis of multimodelsimulations. Geophysical Research Letters 31(L24213).

Tol RSJ, De Vos AF. 1998. A Bayesian statistical analysis of theenhanced greenhouse effect. Climatic Change 38(1): 87–112.

Copyright 2007 John Wiley & Sons, Ltd. 1129 Hydrol. Process. 21, 1127–1129 (2007)DOI: 10.1002/hyp