Ocean State Estimation for the Last Glacial Maximum : Combining Models and Proxy Data via Data Assimilation

Abstract

Investigating past climate states is essential to understand the global climate system and to validate climate models. Data assimilation can be used to obtain estimates of past climate and ocean states that are consistent with model physics as well as with proxy data. The Last Glacial Maximum (LGM, 19-23 ka) was a time interval when the climate was substantially different from today. Even though primary boundary conditions are comparatively well known, the large-scale patterns of the global ocean circulation, especially the strength of the Atlantic Meridional Overturning Circulation (AMOC), remain uncertain. Most studies indicate the presence of a shallower North Atlantic Deep Water (NADW) and a more extensive Antarctic Bottom Water (AABW) during the LGM. However, previous studies using proxy data, models, or a combination of models and proxy data via data assimilation show dissimilar results regarding the AMOC strength. As of yet, only a few state estimates of the LGM ocean obtained from combining models and proxy data exist. To date, no state estimate exists that is based on global surface data as well as on global data from the deep ocean and it is unclear how robust previous results of ocean state estimation are and which influence the assimilation of additional data would have. Furthermore, the adjoint method, which has been used to obtain previous ocean state estimates, requires the "adjoint" of the model code, which is not easily obtained for most models. In this thesis a new technique for ocean state estimation is developed that combines a Kalman smoother method with a state reduction approach. The new technique and the adjoint method are used to obtain estimates of the ocean state during the LGM constrained by global annual and seasonal sea surface temperature reconstructions and by data on the oxygen isotopic composition of calcite from benthic and planktic foraminifera. The estimates are, therefore, constrained by global surface as well as deep-ocean data. The new technique does not require an adjoint and is very computationally efficient if the control space is small. It is tested through pseudo-proxy data experiments and, additionally, it is used to investigate the influence of data from within and outside of the Atlantic Ocean on state estimates. The results of the state estimation for the LGM indicate that SST and oxygen isotope data alone do not necessarily support a shallower NADW and a more extensive AABW during the LGM. The results from ocean state estimation depend strongly on the assimilated proxy data and the experimental design. More proxy data would be required to obtain reliable ocean state estimates. However, the pseudo-proxy data experiments indicate that proxy data at LGM coverage are sufficient to reconstruct the global ocean circulation if no or only a small model error is present. Additional adjoint sensitivity experiments show that especially data from the deep North Atlantic and the global deep Southern Ocean would be important to constrain the AMOC strength. This thesis provides a step forward in using data assimilation for paleo-state estimation with different climate models and highlights the need for more proxy data to obtain realiable reconstructions of the LGM ocean state.