Greg Hakim

We examine the efficacy of a novel ensemble data assimilation (DA) technique in climate field reconstructions (CFR) of surface temperature. We perform four pseudoproxy experiments with both general circulation model (GCM) and 20th Century Reanalysis (20CR) data by reconstructing surface temperature fields from a sparse network of noisy pseudoproxies. We compare the DA approach to a conventional CFR approach based on Principal Component Analysis (PCA) for experiments on global domains. DA outperforms PCA in reconstructing global-mean temperature in all four experiments, and is more cosistent across experiments, with a range of time-series correlations of 0.69--0.94 compared to 0.19--0.87 for the PCA method. DA improvements are even more evident in spatial reconstruction skill, especially in sparsely sampled pseudoproxy regions and for a 20CR experiment. We hypothesize that DA improves spatial reconstructions because it relies on local temperature correlations. These relationships appear to be more robust than orthogonal patterns of variability, which can be non-stationary. Additionally, comparing results for GCM and 20CR data indicates that pseudoproxy experiments that rely solely on GCM data may give a false impression of reconstruction skill.