Direct assimilation of water isotope observations in the Last Millennium Reanalysis (LMR)

Stable water isotopes (SWIs) are the lingua franca of paleoclimate. Yet, most approaches to paleoclimate state estimation from such observations involve some form of calibration, usually to local temperature. This is true not only of traditional regressions as well as many approaches to paleoclimate data assimilation, including the current Last Millennium Reanalysis [e.g. Hakim et al., 2016; Tardif et al., 2019]. Yet it has long been known that SWI ratios in corals, ice cores, bivalves, tree cellulose or lake sediments capture climate signals that cannot be reduced to temperature. 

Motivated by the Iso2k data resource [Konecky et al, 2018], this work leverages process- based models of SWI-based observations [Evans et al., 2013; Dee et al., 2015-17] together with an isotope-enabled model (iCESM) [Brady et al., 2019] to directly assimilate water isotope observations from the PAGES 2k (2017) compilation [PAGES 2k Consortium, 2017]. In cases where linear regression performs well, our results show comparable performance; in cases where linear regression performs poorly, the process-based ice core d18O PSM shows significantly better per- formance in NINO3.4 index reconstructions compared to a linear statistical PSM, in agreement with Steiger et al. [2017]. Water isotopes also offer advantages in constraining fields other than surface temperature: directly assimilating coral d18O constrains tropical precipitation, while assimilating ice core d18O yields only limited information about precipitation. 

For broader applicability, key issues revolve around (1) the specification of observation er- ror variance, (2) correcting biases in the prior state, and (3) developing process models to take advantage of more proxy classes (notably, lake and marine sediments). We will discuss all these issues in turn.