How can we obtain non-biased probability distributions (i.e. error bars) for various aspects of Earth System model projections as climate sensitivity, sea-level rise or MOC strength?

Earth System models used for decadal and centennial projections are usually manually tuned in a high dimensional parameter space to fit current day’s climate. For physically and statistically reliable PDFs of climate quantities of interest at the end of this century we need methodical development and improvement of: automatic tuning processes, valid training data sets & methods, automatic initialisation & bottom up and top down validation attempts. Political decision methods are based on costs and benefits derived from these resulting PDFs and political action/inaction is decisive in this decade. We need to improve the quantitative aspects of the Earth System modeling enterprise substantially to justify massive welfare redistributions for the coming generations. The suggested technical development extends the questions concerning climate sensitivity and sea-level rise because it addresses general methodological difficulties in Earth System Modeling.

One comment

1. cornford says:

   at

   I agree this is a key issue – quantification of uncertainty is central to any question we wish to ask about the earth system. Without this, no action can be justified, and sceptics will always have a route out. All aspects of the modelling require proper treatment of uncertainty, which is a research agenda in itself.