Uk'37, derived sea surface temperatures and coccolith abundances in North Atlantic sediment cores

The UK37' index has proven to be a robust proxy to estimate past sea surface temperatures (SSTs) over a range of time scales, but like any other proxy, it has uncertainties. For instance, in reconstructions of the Last Glacial Maximum (LGM) in the northern North Atlantic, UK37' indicates higher temperatures than those derived from foraminiferal proxies. Here we evaluate whether such warm glacial estimates are caused by the advection of reworked alkenones in ice-rafted debris (IRD) to deep-sea sediments. We have quantified both coccolith assemblages and alkenones in sediments from glaciogenic debris flows in the continental margins of the northern North Atlantic, and from a deep-sea core from the Reykjanes Ridge. Certain debris flow deposits in the North Atlantic were generated by the presence of massive ice-sheets in the past, and their associated ice streams. Such deposits are composed of the same materials that were present in the IRD at the time they were generated. We conclude that ice rafting from some locations was a transport pathway to the deep sea floor of reworked alkenones and pre-Quaternary coccolith species during glacial stages, but that not all of the IRD contained alkenones, even when reworked coccoliths were present. We speculate that the ratio of reworked coccoliths to alkenone concentration might be useful to infer whether significant reworked alkenone inputs from IRD did occur at a particular site in the glacial North Atlantic. We also observe that alkenones in some of the debris flows contain a colder signal than estimated for LGM sediments in the northern North Atlantic. This is also clear in the deep-sea core studied where the warmest intervals do not correspond to the intervals with large inputs of reworked coccoliths or IRD. We conclude that any possible bias to UK37' estimates associated with reworked alkenones is not necessarily towards higher values, and that the high SST anomalies for the LGM are unlikely to be the result of a bias caused by IRD inputs.