Climate field reconstruction uncertainty arising from multivariate and nonlinear properties of predictors

Abstract

Climate field reconstructions (CFRs) of the global annual surface air temperature (SAT) field and associated global area-weighted mean annual temperature (GMAT) are derived in a collection of pseudoproxy experiments for the past millennium. Pseudoproxies are modeled from temperature (T), precipitation (P), T + P, and VS-Lite (VSL), a nonlinear and multivariate proxy system model for tree ring widths. Spatial patterns of reconstruction skill and spectral bias for the T + P and VSL-derived CFRs are similar to those previously shown using temperature-only pseudoproxies but demonstrate overall degraded skill and spectral bias for SAT reconstruction. Analysis of GMAT spectra nevertheless suggests that the true GMAT frequency spectrum is resolved by those pseudoproxies (T, T + P, and VSL) that contain some temperature information. The results suggest that mixed temperature and moisture-responding paleoclimate data may produce actual GMAT reconstructions with skill, error, and spectral characteristics like those expected from univariate and linear temperature responders, but spatially resolved CFR results should be analyzed cautiously