But later I read another paper by the same group of authors:
Dessai, S., Hulme, M., Lempert, R. and Pielke, R. Jr. (2009): Climate prediction: a limit to adaptation? Chapter 5 in, Adapting to climate change: thresholds, values, governance. Adger, W.N., Lorenzoni, I. and O’Brien, K. eds., Cambridge University Press, Cambridge, 530pp. A PDF file is available at http://mikehulme.org/category/academic-publications/ .

It seems that the distinction between “predictions” and “predictions” which the authors wanted to discuss is in the attitude of the users of the results of the simulations. If the users want accurate estimates of future climate, the simulations are considered as “predictions” even IPCC does not call so. But their goal is elusive. On the ther hand, if the users want to test whether a plan of adaptation is robust, they will use a broad set of projections. Then, the range of scenarios is much more important than precision of each scenario.

Dessai et al. provide convincing arguments for a robust decision making approach to climate change on the basis that predictive approaches face both theoretical and practical limitations. Another important reason for linking climate models more directly with management decision processes is that it allows us to identify key performance measures (that influence decision making) and thereby helps prioritize model improvements. In models with such a high level of complexity, there are almost an unlimited number of potential performance measures, and those that are currently in use may not be the most significant from a decision making perspective.

As a final comment, it needs to be recognized that the issues associated with model complexity and uncertainty raised by Dessai et al. are compounded many times over as we attempt to include other ecological and human components into climate models. If we are not to wander aimlessly in model parameter space, robust decision making needs to be one of the main drivers in the burgeoning field of Earth System Science.

Scott Condie
(scott.condie@csiro.au)

As for emergence vs. reductionism, my conception is somewhat different from Harrison and Stainforth.
In a model (NICAM http://www.nicam.jp/) where cloud clusters and equatorial waves are possible, super cloud clusters and Madden-Julian oscillation emerged. I do not think that they are reduced to dynamics and cloud microphysics. In a model complex enough, emergence is possible. A difficult question is whether emergence in the model is analog to emergence in the real world. About weather phenomena, we can wait and see. About climate change, can we anticipate?

I think that the issue about funding to climate modelling is a case of the following problem often found in issues of technological development. We can say that “we cannot achieve X without doing Y”. In other words, Y is a necessary condition for X. But we are not sure whether Y is a sufficient condition for X. Even if we do Y, we are not sure whether we can achieve X. Then the society must decide whether it should invest on Y under irreducible uncertainty. Scientists must make sure that the decision makers know the structure of the uncertainty.

Another issue is that there are 9 Focus Groups waiting in the wings to join the Council as full voting members, following in the footsteps of Biogeosciences. This would only increase the illogicality of the Council structure, since Focus Groups are created based on individual initiative at the grassroots level, not on any view of the logical makeup of the Council imposed from above. The Council already teeters on the edge of being too large to be an effective discussion body, and the addition of any more Focus Groups as voting members would push it over the edge.

What is needed is not to preserve a 90-year-old historical snapshot, but to trim down the Council to a manageable size (12-14 voting members in my opinion) which fairly represents the current state of Earth System Science. This would then allow vigorous discussion of important issues, something that was sorely lacking on the Council during my tenure (although there were some shining exceptions which I shall always treasure).

Note that the US Senate or the House of Lords is certainly not a logical model to follow for the AGU Council. These were bodies created as a kind of compromise–the US Senate to get the small states to sign the original Constitution, and the House of Lords to get the landed aristocracy to buy into the newly emerging democracy in England.

In the modern world of data-driven science, we believe that there is a need to emphasize data sharing as a core ethic of science and a need to develop a sustained data infrastructure supporting scientific research and applications. In that light, we urge you to consider the following points:

- Data must be freely and openly accessible, and should be made available as soon as practicable. Research results should not be published unless the associated data are readily available, ideally as part of the paper itself. This is an area where AGU, as a publisher, can show leadership in both policy and technology.

- Free and open access should explicitly recognize the need for equitable access to research communities across the globe. This could impact where data are stored, in what form, and the technologies used to serve those data.

- When data are used in a publication they should be formally cited, crediting both the data author (provider) and the data publisher. If data cannot be cited and accessed, the result should not be publishable in the scientific literature. This would require some modification of AGU’s current data citation policy (http://www.agu.org/pubs/data_policy.html). Further methods need to be developed and employed to ensure consistent and continual data reference and access.

- Data quality is increasingly important, especially in interdisciplinary research. Data providers cannot anticipate every use of their data and cannot be solely responsible for data quality. On the other hand data users cannot always be expected to fully determine the quality of data outside their field. The diverse data applications in interdisciplinary science demand greater levels of data description, characterization of uncertainty, and understanding of provider assumptions and metaphors. Formal robust semantic procedures may be helpful. New data verification methods, including the peer review of data should be explored.

- In developing the data infrastructure of the future, we will increasingly rely on robust and open standards. Specific standards should not be prescribed, but we encourage the use of relevant standards in preserving, managing and exchanging data and metadata.

- Finally, we must look to the past as well as the future. It is not enough to ensure that data being produced today are archived and available. There is also a need to identify, prioritize, prepare, and preserve historical data at risk.

Thank you for the opportunity to comment,

Mark A Parsons, on behalf of the AGU Information Technology Committee.