古气候数据同化能够有效融合气候模式模拟与代用资料，已成为重建过去气候变化的重要方法。作为基础的先验数据集对同化结果具有重要影响，但目前针对不同模式输出作为先验数据集对同化结果的差异化影响，尚缺少系统性对比与定量评估研究。基于离线古气候数据同化框架，系统评估了13 个耦合气候模式输出作为先验数据集对北半球过去千年年均温度场重建的影响。对1880—2000 年观测时期重建的结果显示，同化后的重建数据质量在时空维度上均获提升：时间相关系数由0.54~0.88 提高至0.86~0.89，泰勒技巧评分由0.31~0.78 提升至0.71~0.80，且不同重建结果间的离散程度缩小；空间维度上，同化有效提升了北半球年际变率的同步性并校正了多数区域的平均态冷暖偏差，但改善幅度呈现区域差异特征，主要受代用资料覆盖密度与先验场初始质量的共同约束，模式固有偏差是制约特定区域功效系数提升的关键因素。通过多维度秩评分体系，将重建结果划分为高、中、低性能组进行评估后发现，各组均能清晰捕捉中世纪暖期、小冰期和现代暖期的演变特征，并反映出了极地放大效应及强火山事件冷却响应，其中高性能组表现较为稳健。研究证实了耦合模式比较计划的标准输出作为同化先验数据集的可行性，也为先验优选提供了定量依据，以期为理解过去千年气候演变特征提供数据基础。

Abstract： Paleoclimate data assimilation effectively integrates climate model simulations with proxy records and has emerged as a key approach for reconstructing past climate variability. The prior dataset fundamentally shapes assimilation outcomes, yet systematic comparative assessments of how different model outputs perform as prior inputs remain scarce. We applied an offline paleoclimate data assimilation framework to systematically evaluate the influence of prior datasets derived from 13 coupled climate models on annual mean temperature field reconstructions across the Northern Hemisphere over the past millennium. Evaluation over the 1880-2000 observational period revealed consistent improvements in reconstruction quality across both temporal and spatial dimensions: temporal correlation coefficients increased from 0.54~0.88 to 0.86~0.89, Taylor skill scores rose from 0.31~0.78 to 0.71~0.80, and the spread across reconstructions narrowed substantially. Spatially, assimilation enhanced the coherence of interannual variability across the Northern Hemisphere and corrected mean-state biases in most regions. The magnitude of improvement varied regionally, primarily reflecting the combined constraints of proxy network density and prior field quality, with model-inherent biases constituting a key limiting factor for efficiency coefficient gains in specific regions. A multidimensional rank scoring system classified all reconstructions into high- , medium- , and low-performance groups. All groups clearly captured the Medieval Warm Period, the Little Ice Age, and the Modern Warm Period, and reproduced both polar amplification and the cooling signatures of major volcanic eruptions, with the high-performance group demonstrating the greatest robustness. This study confirms that standard Coupled Model Intercomparison Project (CMIP) outputs serve as viable prior datasets for paleoclimate data assimilation, provide a quantitative basis for prior selection, and offer a data foundation for advancing our understanding of climate variability over the past millennium.