Globality and Optimality in Climate Field Reconstructions from Proxy Data

A primary objective of paleoclimate research is the characterization of natural climate variability on time scales of years to millennia. We develop here a systematic methodology for the objective and verifiable reconstruction of climate fields from sparse observational networks of proxy data, using reduced space objective analysis. In this approach we seek to reconstruct only the leading modes of large-scale variability which are observed in the modern climate and resolved in the proxy data. Given explicit assumptions, the analysis produces climate fields, indices, and their associated estimated errors. These may be subsequently checked for consistency with parameter choices and procedural assumptions by comparison with withheld data and results from benchmark experiments. The methodology is applied to the candidate tree ring indicator data set described by Villalba et al. (1999), for the reconstruction of gridded Pacific Ocean basin sea surface temperature (SST) over the interval 1001-1990. We find that one mode of variability may be verifiably distinguished by the tree ring indicators. This mode may be interpreted as a decadal, ENSO-like pattern which influences climatic conditions in both North and South America. We speculate this pattern is recovered because large-scale SST variability influences downstream local meteorological conditions sensed by the tree ring data. Verification statistics for the proxy-reconstructed fields may be compared with those derived from calibrated red noise time series and from historical temperature and precipitation data collected near the proxy sampling sites. The results suggest that while credible reconstructions are possible, resolved variance is small and reconstruction errors are large, limiting interpretatio n to regions where there is most skill. Additional proxy data from the Pacific coast of the Americas should improve the resolution and number of reconstructed large-scale SST modes, given that the observations are unbiased, the map describing the connection between SST and the proxy data is well-defined, and appropriate observational errors have been prescribed.