Reassessment of recommendations for processing mammal phosphate δ18O data for paleotemperature reconstruction

Abstract

The stable oxygen isotope composition of phosphates (δ18OP) extracted from vertebrate bone and teeth material is widely used for paleotemperature reconstruction over the Quaternary period. However, inconsistency in analytical and statistical procedures used by different researchers for recalculation of mean air temperatures from measured δ18OP values makes integration of different data sets challenging. Therefore, consistency in both stable isotope data processing and the application of appropriate statistical procedures is highly desirable. This paper evaluates the uncertainty associated with two most commonly used regression methods, least-squares inverted fit and transposed fit. We assessed the performance of both fits by designing ‘real-life’ experiments using various data sets. We used both regression models to reversely calculate the differences between actual recorded in ‘real-life’ values and values calculated from the regression models. Our calculations clearly show that the mean absolute errors are always substantially higher for the inverted fit and the transposed fit always returned mean values closer to the values measured in ‘real life.’ In all studied scenarios, the predictive models performed better than causal models, returning mean absolute errors by 12–65% lower. In the paper, we also examined the other main sources of uncertainty, concerns about calculation of δ18OP of phosphates from δ18OC of carbonates, averaging of δ18OW in precipitation and highlight major challenges in the field of phosphate stable oxygen isotope biogeochemistry. The estimation of paleotemperatures from isotope records could be improved by employing globally consistent calculation procedures.