Climate Field Reconstruction via Data Assimilation

3 We examine the efficacy of a novel ensemble data assimilation (DA) technique in climate field 4 reconstructions (CFR) of surface temperature. We perform several pseudoproxy experiments 5 with both general circulation model (GCM) and 20th Century Reanalysis (20CR) data by 6 reconstructing surface temperature fields from a sparse network of noisy pseudoproxies. We 7 compare the DA approach to a conventional CFR approach based on Principal Component 8 Analysis (PCA) for experiments on global and Antarctic domains. Both methods reproduce 9 time series of global mean temperature with similar magnitudes and correlation coefficients 10 despite the fact that reconstruction skill can vary considerably by data set; for a comparison 11 global-mean temperature reconstruction, DA (PCA) correlations of 0.44 (0.32) for 20CR data 12 contrast with 0.89 (0.82) for GCM data. Important differences involve spatial reconstruction 13 skill: DA particularly outperforms PCA in sparsely sampled pseudoproxy regions and for 14 20CR. Average DA correlations for a 20CR reconstruction are a factor of 1.8 greater than 15 PCA globally and 2.1 greater in sparse pseudoproxy locations. Antarctic reconstructions 16 are consistent with the global results in that both DA and PCA have similar regional17 mean temperature reconstructions and that DA shows higher spatial skill throughout the 18 domain. We hypothesize that DA improves spatial reconstructions because it relies on local 19 temperature correlations; these relationships appear to be more robust than orthogonal 20 patterns of variability, which can be non-stationary. Additionally, our results indicate that 21 pseudoproxy experiments that rely solely on GCM data may give a false impression of 22 reconstruction skill. 23