They begin by noting that the system "must be modeled as an interactive whole," and that "because of the complexity of the process and interactions involved, high-performance computing is absolutely essential."  As they go on to elaborate, however, today's climate models are sorely lacking in this "absolutely essential" characteristic, as they are also deficient in many other important properties, which clearly implies that even our best climate models are not yet up to the task required of them, i.e., accurately predicting the future evolution of earth's climate.

O'Neill and Steenman-Clark note, for example, that there are "considerable gaps in knowledge about the interactions among the sub-systems," and that "current models include only a limited set of the necessary components," which leads us to ask: Are we way off-base in concluding that if today's climate models have "gaps in knowledge" large enough to be described as "considerable," ought not those gaps be filled before one puts much credence in the predictions of the models?  And what about the models possessing a limited set of the necessary components?  Wouldn't one want them to have all of the necessary components before their predictions were deemed correct?

Two examples of the coupling of subsystems that are "poorly treated at present," say O'Neill and Steenman-Clark, are the coupling of changes in atmospheric chemistry with climate and the coupling of the biosphere with climate.  Moreover, they note that "individual subsystems like the atmosphere exhibit enormous complexity in their own right," and that "an increase of high-performance computer power of several orders of magnitude is needed to make significant progress."  This being the case, we again are forced to ask: Are we way off-base in concluding that if we need "several orders of magnitude more computer power" to merely make "progress," is there not a very real likelihood that current climate models are nowhere near being able to produce an accurate description of earth's future climate?

In addition to the maddening complexity of the planet's climate system and the great gaps that exist in our knowledge of its workings, the lack of sufficiently fine spatial resolution is another enormous hurdle that stands in the way of accurate climate change predictions via numerical model calculations.  With respect to the fast and dramatic climate changes that are thought to be linked to similar changes in the thermohaline circulation of the world's oceans, for example, O'Neill and Steenman-Clark say that "predicting rapid change reliably will require coupled models of the atmosphere and ocean with much finer spatial resolution than is used at present."  An "imperative," as they thus put it, is to bring "much greater high-performance computer resources to bear on the problem to allow the Gulf Stream and related circulations to be adequately simulated."  And if that need is truly imperative, as they say, we ask ourselves yet again: Are we way off-base in our belief that this need should be satisfied before we start turning the world's economy upside down in an effort to forestall model-based predictions of catastrophic global warming?

Then there is O'Neill and Steenman-Clark's statement that "it is widely recognized that the representation of convection, clouds and their interactions with radiation is one of the greatest weaknesses of current climate-prediction models," which is also a consequence of insufficiently-fine spatial resolution.  And what is their prescription for solving this problem?  They say that "a major drive in climate modeling must be to reduce the impact of uncertain parameterizations, such as that of convection, by resolving important processes to a greater extent," which clearly requires you-know-what and which prompts us to ask yet one more time: Are we way off base in demanding that the models resolve these processes before we start letting them make our decisions for us?

Of course we're not off-base; our questions and their implied answers are right on the mark.  The nature of the well-chosen words so aptly employed by O'Neill and Steenman-Clark leave no doubt about it - computer modeling of earth's climate, as far as it has come, still has a long, long way to go before it is up to the task of accurately defining future climate.  And until it gets there, the three of us have no intention of letting an inadequately programmed computer usurp the responsibility we have to do our thinking on the vitally important issue of carbon dioxide and global change.  The stakes are just too high.

Reference
O'Neill, A. and Steenman-Clark, L.  2002.  The computational challenges of Earth-system science.  Philosophical Transactions of the Royal Society of London, Series A 360: 1267-1275.