A data assimilation approach to last millennium temperature field reconstruction using a limited high-sensitivity proxy network

Paleoclimate field reconstructions using data assimilation commonly employ large proxy networks, which are often composed of records that have a complex range of sensitivities to the target climate field. This can introduce biases into reconstructions or decrease overall skill. Smaller networks of highly-sensitive proxies provide an alternative, but have not been extensively used for assimilation and their strengths and limitations are less well understood. Here, we reconstruct Northern Hemisphere summer temperature anomalies over the last millennium by assimilating the NTREND network, a spatially and temporally limited collection of highly temperature-sensitive tree-ring records. Pseudo-proxy experiments indicate that the reconstruction can be sensitive to biases in the climate model prior, so we perform 10 assimilations each using a different model prior. Reconstructed temperature anomalies are most sensitive to prior selection when the network becomes sparse in space and time, but show greater consistency as the network grows. The method also underestimates temporal variability with a reduced network or in regions distal to the proxies. The effects of network attrition emphasize the importance of analyzing temperature anomalies in conjunction with reconstruction uncertainty, which emerges naturally for spatial fields from our ensemble method. A comparison of our reconstruction and five existing paleo-temperature products reveals large differences in the spatial patterns and magnitudes of reconstructed temperature anomalies in response to radiative forcing. These extant uncertainties call for development of a renewed paleoclimate reconstruction intercomparison framework for systematically examining the consequences of network composition and reconstruction methodological choices, as well as for expanded collection of new, longer, and highly-sensitive proxy data.