| Popis: |
All potential temperature proxies (e.g. ?18O, Mg, Sr,...) suffer of one common problem: the proxy signal is strongly influenced by non-temperature forcers such as salinity, metabolism and shell growth. The use of multi-proxy models offers a solution for that problem. However, the use of classical linear multiple regression models are not always appropriated since some data show substantial non-linear relationships between temperature and elemental ratios. In previous research we demonstrated that the so called Weight Determination by Manifold Regularization approach can be used to develop a non-linear multi-proxy model. Now a new question rise: "Which proxies should be integrated in the multi-proxy model?" We evaluated four trace element records (Mg/Ca, Sr/Ca, Ba/Ca and Pb/Ca) measured in the shell of the common blue mussel Mytilus edulis. Different proxy combinations were evaluated over a salinity range between 15 and 32. Our findings highlight Mg/Ca ratios as the most powerful paleothermometers but we indicate that its reconstruction performance is significantly improved by combining it whit other elemental ratios into a multi-proxy model. We assume that disturbances on Mg/Ca profile due to growth and food availability can be explained by de variations in the Sr/Ca profile and the Ba/Ca profile, which results in better temperature reconstructions using a Mg/Ca, Sr/Ca and Ba/Ca the WDMR. Although Pb/Ca ratios seem to contribute slightly positive to the final reconstruction performance of a four proxy model, we discourage the use of Pb/Ca in a multi-proxy model for temperature reconstruction since the element is known to be strongly influenced by antropogenetic forcers. Using Mg/Ca, Sr/Ca and Ba/Ca the WDMR model gave a root mean squared error of ±2.21ºC for a temperature reconstruction based on a shell sampled independently in time and location. |
