NONLINEAR AND DYNAMICAL MODELS FOR TEMPERATURE RECONSTRUCTIONS FROM MULTI PROXY DATA IN BIVALVE SHELLS

Because bivalve shells are synthesized in equilibrium with the ambient environment, shells may reveal past climate information. The chemical composition of a shell depends on environmental variables such as temperature, rainfall, and food availability... the difficulty however, is the translation of the chemical signals that are measured along a shell's growth axis into environmental information. Most studies in this field try to identify one to one relationships between a chemical element (=proxy) and an environmental parameter, based on the similarity of the two seasonal patterns. However, the incorporation of chemical elements in a shell appears to be highly complex and is mostly influenced by several environmental parameters simultaneously. In this work we propose a so called multi-proxy approach. We describe the behavior of one environmental parameter (e.g. temperature) by using a combination of chemical elements (e.g. Mg, Sr, Ba and Pb). The main advantage of this multi-proxy approach is that unexplained variations in one proxy (e.g. Mg) can be explained by variations in another proxy (e.g. Sr) without fully understanding the origin of these variations. Along this work it becomes clear that many elements show nonlinear relationships with their environment. Therefore, in first instance, the multi-proxy approach is applied to nonlinear models. In a second step we introduce (linear and nonlinear) dynamical multi-proxy models, in order to deal with possible deleted proxy incorporation during the shell formation. The introduction of dynamics results eventually in the best reconstructions. The best temperature reconstructions are obtained using dynamical Mg models that result in reconstructions with an average error of 1.3°C. This is considerably better than the reconstructions obtained by classical Mg models and having average errors of 4.2°C.