A multiproxy approach to reconstructing sea surface conditions using coral skeleton geochemistry

[1] We use elemental ratio (Mg/Ca, Sr/Ca, U/Ca and Ba/Ca) and oxygen isotope data from a Porites lutea coral head collected from offshore Amedee Island, New Caledonia (22degrees29'S, 166degrees28'E) to assess the degree to which changes in these geochemical variables reflect variations in sea surface conditions. We have assessed the robustness of each geochemical proxy by comparing 25 years (1992-1968) of monthly geochemical variations with a local record of sea surface temperature (SST) and with the appropriate 1degrees by 1degrees grid box from a global gridded SST data set. We conclude from our comparison that the most consistent proxies of monthly SST variations at this site are delta(18)O (r(2) = 0.84; p = <.0001) and Sr/Ca (r(2) = 0.84; p = <.0001). The fidelity of the coral-based Sr/Ca-SST proxy was assessed via calibration-verification exercises at New Caledonia and Rarotonga. The paleoclimate'' accuracy of the coral Sr/Ca-SST technique is judged to be robust, and the precision of the technique is estimated to be similar to0.3degrees +/- 0.5 (2sigma) based on the calibration-verification exercise. In contrast, monthly variations in Mg/Ca and U/Ca ratios are only moderately correlated with monthly SST variations (r(2) = 0.55; p = <.0001 for Mg/Ca and r(2) = 0.47; p = <.0001). The strength of the correlation between SST and Mg/Ca and U/Ca varies as a function of time, suggesting that variations in these ratios are not a simple function of SST variations. This variability in goodness of fit between U/Ca-SST and Mg/Ca-SST may ultimately limit the use of these ratios as coral paleothermometers. Averaging or stacking individual proxy SST records (Sr/Ca, Mg/Ca and U/Ca) to generate a composite proxy SST record does not necessarily improve the proxy record if the signal-to-noise ratio in each proxy is highly variable. For example, stacking of Sr/Ca-Mg/CaU/Ca records at New Caledonia produces a proxy SST time series that has greater standard error than the individual Sr/Ca-SST time series (cf. ordinary least squares standard error of 1.16degreesC versus 0.86degreesC). Ba/Ca variations have little correlation with SST, Sr/Ca, Mg/Ca, or U/Ca variations at New Caledonia. In conclusion, variations in oxygen isotopic composition of seawater, calculated by removing the Sr/Ca-based temperature component of the delta(18)O signal, agree reasonably well with observed variations in sea surface salinity, especially at the interannual timescale.