A novel Bayesian multilevel regression approach to the reconstruction of an eastern Mediterranean temperature record for the last 10,000 years

Abstract

Climate reconstructions derived from proxy records for individual sites often fail to incorporate existing regional information, which may help to determine uncertainties and express variability within specific reconstructions or commonalities between datasets. Such an understanding is crucial when examining past human-environment interactions. Taking the eastern Mediterranean as our case study, we present here a new air temperature model for the last 10,000 years that utilises data from 33 previously published proxy-based independent reconstructions, using a novel fully Bayesian approach that applies multilevel regression models of individual temperature datasets grouped together into 300 year long consecutive sub-intervals. A Bayesian multilevel approach allows the model to share information between each regression model from the individual datasets and the 300 year grouped regression models. The results demonstrate commonalities between individual datasets derived from different sources, and embed the uncertainties within the model. Our results establish that the eastern Mediterranean region was consistently warmer than the 20th century, except for two short intervals at the end of the Early Holocene (between 8400 and 8250 yrs BP) and the start of the mid-Holocene (between 7800 and 7650 yrs BP). We also identify changes within temperature associated with both the 8.2 ka and 4.2 ka BP events, however our findings identify regional warming in the eastern Mediterranean, rather than cooling often associated with the 4.2 ka BP event. Our results are comparable with previous large scale hemispheric reconstructions, demonstrating that our model is a robust candidate for temperature reconstructions within a confined region, which can range from mesoscale up to macroscale.