Assessing the robustness of Antarctic temperature reconstructions over the past 2 millennia using pseudoproxy and data assimilation experiments

The Antarctic temperature changes over the past millennia remain moreuncertain than in many other continental regions. This has several origins:(1) the number of high-resolution ice cores is small, in particular on theEast Antarctic plateau and in some coastal areas in East Antarctica; (2) theshort...

Full description

Bibliographic Details
Published in:Climate of the Past
Main Authors: Klein, F, Abram, NJ, Curran, MAJ, Goosse, H, Goursaud, S, Masson-Delmotte, V, Moy, A, Neukom, R, Orsi, A, Sjolte, J, Steiger, N, Stenni, B, Werner, M
Format: Article in Journal/Newspaper
Language:English
Published: Copernicus GmbH 2019
Subjects:
Online Access:https://eprints.utas.edu.au/37760/
https://eprints.utas.edu.au/37760/1/146223%20-%20Assessing%20the%20robustness%20of%20Antarctic%20temperature%20reconstructions%20over%20the%20past%202.pdf
Description
Summary:The Antarctic temperature changes over the past millennia remain moreuncertain than in many other continental regions. This has several origins:(1) the number of high-resolution ice cores is small, in particular on theEast Antarctic plateau and in some coastal areas in East Antarctica; (2) theshort and spatially sparse instrumental records limit the calibration periodfor reconstructions and the assessment of the methodologies; (3) the linkbetween isotope records from ice cores and local climate is usually complexand dependent on the spatial scales and timescales investigated. Here, we useclimate model results, pseudoproxy experiments and data assimilationexperiments to assess the potential forreconstructing the Antarctic temperature over the last 2 millennia based on anew database of stable oxygen isotopes in ice cores compiled in the frameworkof Antarctica2k (Stenni et al., 2017). The well-known covariance betweenδ18O and temperature is reproduced in the two isotope-enabledmodels used (ECHAM5/MPI-OM and ECHAM5-wiso), but is generally weak over thedifferent Antarctic regions, limiting the skill of the reconstructions.Furthermore, the strength of the link displays large variations over the pastmillennium, further affecting the potential skill of temperaturereconstructions based on statistical methods which rely on the assumptionthat the last decades are a good estimate for longer temperaturereconstructions. Using a data assimilation technique allows, in theory, forchanges in the δ18O–temperature link through time and spaceto be taken into account. Pseudoproxy experiments confirm the benefits ofusing data assimilation methods instead of statistical methods that providereconstructions with unrealistic variances in some Antarctic subregions. Theyalso confirm that the relatively weak link between both variables leads to alimited potential for reconstructing temperature based onδ18O. However, the reconstruction skill is higher and moreuniform among reconstruction methods when the reconstruction target is theAntarctic as a whole rather than smaller Antarctic subregions. Thisconsistency between the methods at the large scale is also observed whenreconstructing temperature based on the real δ18O regionalcomposites of Stenni et al. (2017). In this case, temperature reconstructionsbased on data assimilation confirm the long-term cooling over Antarcticaduring the last millennium, and the later onset of anthropogenic warmingcompared with the simulations without data assimilation, which is especiallyvisible in West Antarctica. Data assimilation also allows for models anddirect observations to be reconciled by reproducing the east–west contrastin the recent temperature trends. This recent warming pattern is likelymostly driven by internal variability given the large spread of individualPaleoclimate Modelling Intercomparison Project (PMIP)/Coupled ModelIntercomparison Project (CMIP) model realizations in simulating it. As in thepseudoproxy framework, the reconstruction methods perform differently at thesubregional scale, especially in terms of the variance of the time seriesproduced. While the potential benefits of using a data assimilation methodinstead of a statistical method have been highlighted in a pseudoproxyframework, the instrumental series are too short to confirm this in arealistic setup.