Bayesian hierarchical regression analysis of variations in sea surface temperature change over the past million years. (23rd September 2016)

• Record Type: Journal Article
• Authors: Snyder, Carolyn W.
• Is Part Of: Paleoceanography. Volume 31:Number 9(2016)
• Page Start: 1283
• Page End: 1300
• Publication Date: 2016-09-23
• Subjects: local sea surface temperature responsiveness -- Bayesian hierarchical regression analyses -- deep water temperature -- proxy reconstructions -- dating uncertainty -- latitudinal temperature curve; Paleoceanography -- Periodicals; 551.46
• DOI: 10.1002/2016PA002944
• Languages: English
• ISSNs: 0883-8305

Abstract

Abstract: Statistical challenges often preclude comparisons among different sea surface temperature (SST) reconstructions over the past million years. Inadequate consideration of uncertainty can result in misinterpretation, overconfidence, and biased conclusions. Here I apply Bayesian hierarchical regressions to analyze local SST responsiveness to climate changes for 54 SST reconstructions from across the globe over the past million years. I develop methods to account for multiple sources of uncertainty, including the quantification of uncertainty introduced from absolute dating into interrecord comparisons. The estimates of local SST responsiveness explain 64% (62% to 77%, 95% interval) of the total variation within each SST reconstruction with a single number. There is remarkable agreement between SST proxy methods, with the exception of Mg/Ca proxy methods estimating muted responses at high latitudes. The Indian Ocean exhibits a muted response in comparison to other oceans. I find a stable estimate of the proposed "universal curve" of change in local SST responsiveness to climate changes as a function of sin 2 (latitude) over the past 400, 000 years: SST change at 45°N/S is larger than the average tropical response by a factor of 1.9 (1.5 to 2.6, 95% interval) and explains 50% (35% to 58%, 95% interval) of the total variation between each SST reconstruction. These uncertainty and statistical methods are well suited for application across paleoclimate and environmental data series intercomparisons. Key Points: Bayesian hierarchical regressions are applied to sea surface temperature reconstructions Local sea surface temperature response to eight climate indicators is estimated for 54 reconstructions "Universal curve" of change in sea surface temperature by latitude is stable over 400, 000 years