Reconstructing Large-scale Variability from Palaeoclimatic Evidence by Means of Data Assimilation Through Upscaling and Nudging(DATUN)

Abstract

The detection of climatic changes during the last century and their attribution to increasing concentrations of atmospheric greenhouse gases and other anthropogenic activities requires a realistic estimation of the level of natural climate variability on decadal and longer time scales. These estimates can be obtained in two ways. One approach is to perform climate simulations with numerical climate models. Since it is not exactly known how realistic these models are, empirically-based estimates of climate variability are also needed. These are obtained by analysing time series from climate proxies, natural archives that contain a climate signal, and thus information about past climate evolution. The extraction of climate information from proxy data includes the analysis of data from tree rings (Schweingruber and Briffa, 1996), ice cores (Fischer et al., 1998) and corals (Draschba et al., 2000). To gain maximum insight into past climate variability, these records need to be integrated in a way that goes beyond a mere fragmentary comparison. A first step in this direction is the development of climate reconstructions that are based on a statistical combination of multiple proxy data from multiple sites (Mann et al., 1998, Luterbacher et al., 1999). Our work, which is part of the project 'Klima In Historischen Zeiten' (KIHZ, Climate In Historical Times), aims at improving the methodology for extracting the climate signal from proxy records by bringing together proxy data and numerical climate modelling in a new way. This will be achieved by using a coupled atmosphere-ocean General Circulation Model (GCM) to assimilate proxy data, with the goal of obtaining a physically consistent best guess for the large-scale states of the atmosphere during the Late Holocene with annual temporal resolution. Assimilation of observations in GCMs, for instance from surface stations, balloon soundings, or satellites, has been operationally employed for many years to find the initial conditions needed for numerical weather prediction, as well as to obtain atmospheric reanalyses. In these cases sophisticated multi-step schemes, which take into account the estimated errors and the spatial correlation structure of the observations, and suppress unphysical high frequency variability are employed. Data assimilation in GCMs has also been used for process studies and model validation, usually using the simpler, so-called nudging method, which directly relaxes the model towards local observations or large-scale target fields (e. g. Timmreck et al., 1999, Murphy, 2000). All of these strategies require a relatively precise knowledge of the target state and are not suited for assimilating proxy data or sparse instrumental data. Proxy data are available only from a few locations and typically represent climate signals that are integrated over several months to decades. In addition large uncertainties exist due to the complex relationship between climate and proxy variables, as well as due to non-climatic influences on the proxy records. Thus an assimilation method that is tailored towards applications in palaeoclimatology has been developed (von Storch et al., 2000). This so-called DATUN technique (Data Assimilation Through Upscaling and Nudging) consists of two steps. The first step is the formulation of statistical upscaling models, which link the local proxy data to large-scale circulation states, for example to the amplitudes of dominant atmospheric variability patterns, such as the North Atlantic Oscillation (NAO) (Gonzalez-Rouco et al., 2000) and the Antarctic Oscillation (AAO). A reconstruction of the strength of the summer (DJF) AAO, which is the dominant variability pattern of southern hemisphere extratropical sea level pressure, has been undertaken using tree ring chronologies from Argentina, Chile, Tasmania and New Zealand. For the purposes of this work, the AAO has been defined as the first EOF of NCEP/NCAR reanalysis sea level pressure (SLP) for the domain 15�S - 60�S. Data further south were not used because of concerns about the amount input data to the reanalysis. The chronologies were obtained from the International Tree Ring Data Bank. To select those chronologies containing an AAO signal, the chronologies were correlated with the principal component (PC) of this EOF, and those which were significant at the 5 % level were retained for analysis, a total of 10 chronologies from a pool of 59. In order to produce the reconstruction, canonical correlation analysis (CCA) was applied to the first five PCs of the chronologies and the first SLP PC, which explains 27 % of the variance of the SLP field. The first canonical pattern, which is identical to the first SLP EOF, is shown on the left side in Figure 1.