Sir — In his Commentary “What is‘dangerous’ climate change?”, StephenSchneider1 argued that — in the absence ofunambiguous expert advice — decision-makers will produce their own probabilityestimates about future climate changewithin the large range of projectionsprovided by the Intergovernmental Panelon Climate Change (IPCC) in its ThirdAssessment Report, and that this is worsethan using informed estimates provided byrelevant experts. On the other hand,Grübler and Nakicenovic2 in a subsequentCorrespondence, “Identifying dangers in anuncertain climate”, argue that providingwell-founded probability estimates isdifficult or impossible because, althoughprobabilities in the natural sciences arebased on repeated experiments andfrequencies of measured outcomes, this isnot the case in the socio-economic sciences.

However, this frequentist basis forprobabilities in predictions of an unknownfuture is not possible in the Earth scienceseither, since there will only be one realoutcome, which cannot be measured now.Probability estimates of future conditionson Earth based on modelling are notfrequentist, but essentially bayesian3, inthat they are based on prior knowledge orassumptions embodied in the variousmodels and inputs.

Various authors have shown that suchestimates for global warming arepossible1,4,5 by deriving single-peakedprobability distributions. On the otherhand, Gritsevskyi and Nakicenovic6 obtaina bimodal probability distribution ofcarbon dioxide emission amounts for theyear 2100, which they attribute to abifurcation in technological developmentpathways. If this bimodal distribution werecombined with other factors to derive aprobability distribution for globalwarming, it would almost certainly besmoother, but could still differ signifi-cantly from the other distributions.

Thus, although probability estimatesare needed, methods for derivingprobabilities require furtherdevelopment7. Some authors havesuggested that in the meantime we shouldrely on increasing robustness8 or resilience9

in developing mitigation and adaptationresponses. But without well-foundedprobability estimates, these strategies stilltacitly assume some estimates of likelihoodto limit the magnitude of their responsesin relation to costs.

In fact, a risk-management approachrequires not an assessment of theprobability of a particular amount of

greenhouse-gas emission or globalwarming at some future time, but ratheran estimate of the likelihood of exceedingan identified critical impact threshold10.This integrates the probabilities from theleast climate change up to the critical level,and is much more robust with respect tothe underlying assumptions.

In the IPCC Third Assessment Reportdiscussed in ref. 1, the uncertainties inprojections of global surface warming by2100 derive almost equally from uncertainemissions and uncertain climate science.Thus if critical thresholds for impacts lie inthe top half of the predicted range of globalwarmings, these are much more likely tobe avoided if decision-makers attempt tofollow socio-economic developmentpathways that will reduce emissions. Thevirtue of the range of socio-economicdevelopment pathways explored by IPCCin ref. 11 is that it suggests that there arevarious plausible socio-economic futuresthat will achieve this goal. Suchdevelopment pathways may still lead toappreciable warming by 2100 and beyond,which can be avoided only by stringentclimate policies.

Without probability estimates,

engineers and planners will be left needingto foster resilience and adaptive capacity,hedge their bets, delay their decisions, orelse gamble on whether humanity will godown high or low emissions developmentpathways as they adapt design standardsand zoning to climate change.

It is far more sensible to establishcumulative probability distributions toallow optimal, focused adaptation plans. A. Barrie Pittock, Roger N. Jones, Chris D. Mitchell Climate Impact Group, CSIRO AtmosphericResearch, PMB 1, Aspendale 3195, Australia 1. Schneider, S. H. Nature 411, 17–19 (2001).

2. Grübler, A. & Nakicenovic, N. Nature 412, 15 (2001).

3. Malakoff, D. Science 286, 1460–1464 (1999).

4. Webster, M. D. et al. Uncertainty Analysis of Global Climate

Change Projections, Report No. 73, MIT Joint Program on the

Science and Policy of Global Change (Cambridge, MA, 2001).

5. Wigley, T. M. L. & Raper, S. C. B. Science 293, 451–454

(2001).

6. Gritsevskyi, A. & Nakicenovic, N. Energy Policy 28, 907–921

(2000).

7. New, M. & Hulme, M. Integrated Assessments 1, 203–213

(2000).

8. Lempert, R. J. & Schlesinger, M. E. Climatic Change 45,

387–401 (2000).

9. Barnett, J. World Development 29, 977–993 (2001).

10.Jones, R. N., Natural Hazards 23, 197–230 (2001).

11.Nakicenovic N. & Swart, R. Special Report of the

Intergovernmental Panel on Climate Change on Emissions

Scenarios (Cambridge Univ. Press, Cambridge, 2000).

correspondence

NATURE | VOL 413 | 20 SEPTEMBER 2001 | www.nature.com 249

Probabilities will help us plan for climate changeWithout estimates, engineers and planners will have to delay decisions or take a gamble.

Vital parameters need to be in print Sir — In an effort to condense Letters toNature in the printed version of the journal,there is a risk that some of the criticalinformation necessary for an independentjudgement of the quality of biologicalstructural information may be omitted,appearing only in the SupplementaryInformation available in the online version.

These indicators are: the resolution ofthe structure determination; the ‘free Rfactor’ (an unbiased metric of agreementbetween the experimental X-ray diffractiondata and the derived molecular model);and the Ramachandran analysis (the onlyindependent measure of a model’s stereo-chemical reasonableness).

In some cases, this critical information— which takes little space — is excluded,whereas non-essential information, such asthe location of the synchrotron beamlineswhere the diffraction data were collectedand the names of standard programs usedto determine the structure, is included inthe printed version. It is often inconvenientto have to access the Internet while readinga paper in print, in order to obtain theseessential quality indicators.

The crystallographic community hashad a welcome change of heart onagreement about validation criteria forstructural models — critical use of theindicators I have mentioned above — andon data availability via deposition inpublicly available databases. Given thatNature has fully supported this shift, shouldthe journal not also enforce minimal,accessible reporting of a structure’s qualityindicators in the printed versions of papers,rather than allowing authors to deposit allthis information in the online-only formatof Supplementary Information?David BorhaniAbbott Bioresearch Center, 100 Research Drive,Worcester, Massachusetts 01605, USA

Nature’s policy is to include in the printedversions of its papers sufficient technicalinformation to allow an interested readerto appreciate the experiments that havebeen performed, along with the mostcritical experimental data. We wouldtherefore normally expect essentialstructural parameters to be included inthe published version of papers ratherthan as part of the SupplementaryInformation only. However, this is always assessed by Nature on a case-by-case basis — Editor, Nature.

© 2001 Macmillan Magazines Ltd