Facilitating proxy-data interpretation of abrupt climate events using model simulations

tance between each member of the ensemble and the proxy record is measured by a cost function using reconstructed and simulated variables at the locations where the proxies are available. The best simulation, defined as the one that minimizes this cost function, is then selected as representative for this particular year or period, and used as the initial condition for the subsequent year. The procedure is repeated as many times as required in order to provide a reconstruction for the whole millennium (e. In this technique, the model is only constrained locally, at the locations where the proxy records are available. This is a clear advantage, compared to the nudging techniques, as the reconstructed spatial pattern is the result of the data assimilation procedure itself and is thus independent of any statistical method used to reconstruct patterns. A downside of the approach is the potential systematic error incurred if the model is not able to reproduce observed teleconnections between different regions. Preliminary proofs of concept using this ensemble method have demonstrated that it can efficiently yield a plausible large-scale reconstruction if only a small number of proxies are available, and yet can also reconstruct regional detail where the number of available proxy data are sufficiently large (e.g., Goosse et al., 2006). On the basis of these successful preliminary results, ongoing work is underway to investigate in greater detail the mechanisms that may be responsible for the climate changes of the past millennium. Potential refinements of the approach include a more sophisticated treatment of the uncertainties in both the proxy data and simulation results. This is a challenging issue for a number of reasons. For example, the nature of the uncertainties in proxy-derived climate records are complex, involving complicated physical or biological responses, which may yield frequency-dependent loss of climate information. Further work is necessary to characterize these uncertainties and biases more fully. Furthermore, classical assimilation methods used in meteorology and oceanography cannot be transferred directly to the analysis of the past millennium. Fortunately, focused efforts in this area are now underway, as discussed at one recent workshop on " Data assimilation to study the climate of the past millennium " (www.astr.ucl.ac.be/index. php?page=Wokshop_assim). It is reasonable to expect that significant improvements in the techniques used for data assimilation over the last millennium will be achieved in the years ahead, yielding significantly refined estimates of past changes and a better understanding of …