Enhanced warming in the Arctic (Arctic amplification, AA) in the last decades has been linked to several factors including sea ice and the Atlantic Multidecadal Oscillation (AMO). However, how these factors contributed to AA variations in a long-term perspective remains unclear. By reconstructing a millennial AA index combining climate model simulations with recently available proxy data, this work determines the important influences of the AMO and anthropogenic greenhouse gas forcing on AA variations in the last millennium, leading to identification of a significant downward trend of AA on top of a sustained strong AMO modulation at the multidecadal scales. The decreased AA during the industrial era was strongly associated with the anthropogenic forcing, proving the emerging role of the forcing in reducing the AA strength.

Climate change simulations based on climate models are inevitably uncertain. This uncertainty typically stems from parametric and structural uncertainties in climate models as well as climate forcings. However, combining model simulations with instrumental observations using appropriate statistical methods is an effective approach for describing this uncertainty. In this study, the authors applied Bayesian model averaging (BMA), a statistical postprocessing method, to an ensemble of climate model simulations from the Paleoclimate Modelling Intercomparison Project phase 3 (PMIP3) and phase 5 of the Coupled Model Intercomparison Project (CMIP5). Uncertainties, weights, and variances of individual model simulations were estimated from a training period using the National Centers for Environmental Prediction–National Center for Atmospheric Research (NCEP–NCAR) reanalysis dataset. The results presented here demonstrate that the BMA method is successful and attains a positive performance in this study. These results show that the selected proxy-based reconstructions and simulations are consistent with BMA estimates regarding climate variability in the past 1000 years, though differences can be found for some periods. The authors conclude that BMA is an effective tool for describing uncertainties associated with individual model simulations, as it accounts for the diverse capabilities of different models and generates a more credible range of past climate change over a relatively long-term period based on multimodel ensemble simulations and training data.

Constructing suitable tree ring width (TRW) proxy system models (PSMs) is an emerging research focus in paleoclimate data assimilation (PDA). Currently, however, it is unknown as to which TRW PSMs are optimal for practical PDA applications. This study proposes an artificial neural networks (ANN)‐based TRW PSM and compares its performance with those of existing TRW PSMs, including linear univariate model, linear multivariate model, and physically based VS‐Lite model. The results show that ANN‐based TRW PSM is more suitable for practical PDA applications than other three TRW PSMs in terms of performance and universality. Overall, the performances of the four TRW PSMs in PDA can be ranked as follows (from best to worst): ANN, linear multivariate model, linear univariate model, and physically based VS‐Lite model. In addition, the results of our study not only indicate that the ANN model is a really effective tool for constructing TRW PSM in practical PDA applications but also imply that the ANN model has the potential to provide new insights into the construction of other types of PSMs (e.g., speleothem δ¹⁸O PSM) when physics of the climate‐proxy relationships cannot be described fully in advance.

The spatiotemporal variations of the climate of the last millennium provide the context for understanding the recent climate. However, there are few 1000-year-long spatiotemporally continuous climate reconstructions at hemispheric to global scales. Here, a millennial, spatiotemporally continuous and annually resolved air temperature dataset for the Northern Hemisphere (NH) was reconstructed using paleoclimate data assimilation (PDA) approach based on an expanded tree ring proxy dataset and artificial neural network-based proxy system models. Verifications show the PDA-based reconstruction agrees well with instrumental temperature, previous proxy-based reconstructions and the Last Millennium Climate Reanalysis version 2 (LMRv2) dataset in space and time domains. Since the PDA-based reconstruction is spatiotemporally continuous fields without missing value over the whole NH, which is a distinct advantage of the PDA-based reconstruction over other proxy-based NH-averaged temperature reconstructions, we focussed on analysing the spatial variations of the NH temperature during the last millennium, for example, the primary modes of temperature variabilities and the spatial variabilities of the NH temperature in three typical climatic epochs of the millennium. Results show that the first-leading mode (explained variance of 72.50%) of the NH temperature variability is a spatially consistent mode, while the second-leading mode (explained variance of 10.60%) exhibits a different pattern that is different from the first-leading empirical orthogonal function mode, particularly in the Arctic, which shows a dipole structure anomaly. The 20th century was the warmest, with 94.17% of the NH displaying positive anomalies. The 11th century was the second warmest century in the record, with 70.26% of the NH showing a positive anomaly. The 19th century was the coldest, with negative anomalies covering 86.97% of the NH. Northwest North America and the Northern Europe-Barents Sea are two distinctly anomalous centres during these three centuries. Overall, the reconstructed air temperatures provide a new and reliable dataset for studying the millennial NH temperature variations.