First Results from The Last Millennium Climate Reanalysis Project

Abstract:

Paleoclimate proxies provide the only measured record of Earth's climate history, but they are noisy and sparse in space and time. Climate model simulations provide dynamically consistent spatial fields, but lack a direct connection to specific climate states prior to the instrumental record. Paleoclimate data assimilation (PDA) provides an optimally weighted estimate of the climate state from these two sources of information based on their error characteristics. The Last Millennium Climate Reanalysis Project (LMR) uses an ensemble-based PDA method and annually-resolved proxy records to reconstruct Earth's climate for the past 1000 years on a regular latitude--longitude grid. First results of the LMR project are reported here.

Proxy records used in the first LMR reconstructions include: trees (ring width and wood density), corals (d18O and luminescence), ice cores (d18O), and a small number of sediment and speleothem records. A component of the data assimilation approach that distinguishes it from other techniques concerns the use of proxy system models to estimate the proxy from climate variables from a model simulation. Proxies are linearly related to 2-meter air temperature in a calibration dataset. Given a prior estimate of the climate from a model (here, a randomly sampled 1000-year control integration of CCSM4), we estimate the proxy value from the linear relationship (derived independently from the model). LMR analyses are compared against both existing gridded reanalysis records and withheld proxy records.

Results show a cooling trend in global-mean air temperature during 1000-1900 CE, which derives primarily from cooling of the Northern Hemisphere extratropics, offset by weak tropical warming. There is no evidence of a Medieval Climate Anomaly. During the 20^th century, the LMR estimate of the global-mean air temperature compares very closely with other reanalysis products. Skill appears insensitive to the calibration dataset used to derive the proxy system models. Verification of spatial patterns shows coherent large-scale patterns of skill, including fields such as 500 hPa geopotential height. An essential aspect of the LMR involves error analysis, which reveals that the ensemble is very well calibrated when is error is defined by independent proxies or by gridded reanalysis fields.