Evaluating climate field reconstruction techniques using improved emulations of real-world conditions

Pseudoproxy experiments (PPEs) have become an important framework forevaluating paleoclimate reconstruction methods. Most existing PPE studiesassume constant proxy availability through time and uniform proxy qualityacross the pseudoproxy network. Real multiproxy networks are, however,marked by pronounced disparities in proxy quality, and a steep decline inproxy availability back in time, either of which may have large effects onreconstruction skill. A suite of PPEs constructed from a millennium-lengthgeneral circulation model (GCM) simulation is thus designed to mimic thesevarious real-world characteristics. The new pseudoproxy network is used toevaluate four climate field reconstruction (CFR) techniques: truncated totalleast squares embedded within the regularized EM (expectation-maximization) algorithm (RegEM-TTLS), theMann et al. (2009) implementation of RegEM-TTLS (M09), canonical correlationanalysis (CCA), and Gaussian graphical models embedded within RegEM(GraphEM). Each method's risk properties are also assessed via a 100-membernoise ensemble. Contrary to expectation, it is found that reconstruction skill does not varymonotonically with proxy availability, but also is a function of the type andamplitude of climate variability (forced events vs. internal variability).The use of realistic spatiotemporal pseudoproxy characteristics also exposeslarge inter-method differences. Despite the comparable fidelity inreconstructing the global mean temperature, spatial skill varies considerablybetween CFR techniques. Both GraphEM and CCA efficiently exploitteleconnections, and produce consistent reconstructions across the ensemble.RegEM-TTLS and M09 appear advantageous for reconstructions on highly noisydata, but are subject to larger stochastic variations across differentrealizations of pseudoproxy noise. Results collectively highlight theimportance of designing realistic pseudoproxy networks and implementingmultiple noise realizations of PPEs. The results also underscore thedifficulty in finding the proper bias-variance tradeoff for jointlyoptimizing the spatial skill of CFRs and the fidelity of the global meanreconstructions