Improved calibration of organic SST proxies via community-driven Bayesian, spatially-varying regression

Improving the calibration of SST proxies is fundamental to providing accurate estimates of past changes in sea-surface temperatures. Existing calibrations assume spatially and temporally constant regression terms, but this may not adequately capture the influence of both known and unknown secondary environmental factors on proxy response. As an alternative, we propose a BAYesian, SPAtially-varying Regression model (BAYSPAR) for general application to marine organic geochemical SST proxies. The calibration model treats regression parameters as slowly-varying functions in space and allows for a full propagation of errors in both the proxy and the SST field. Initial application of the technique to the TEX86 proxy demonstrates that it yields better-behaved residuals than previous calibrations and therefore improves SST estimates in certain regions. Two different prediction models allow users to apply to the calibration to either Neogene or "deep-time" data, the latter of which uses an analog approach. Traditionally, calibrations for SST proxies are updated incrementally via individual publications over a period of many years, and in some cases the coretop collections that form these calibrations are left unarchived. To facilitate both up-to-date prediction and data archiving, BAYSPAR will be designed to reflect community-based improvements in knowledge and data in real time via a semi-autonomous updating process. Users may enter new coretop data into a portal on the web, and after a screening procedure, the data will be added to the calibration model, which will then be autonomously updated and made available to the users. In this way, calibration of SST proxies becomes a community-driven process.