过去2000年全球气候再分析资料数据集 (NNU-2ka Reanalysis)

该数据集为过去2000年全球温度降水同化结果。使用不同代理数据集（PAGES2k、PAGES2k+Mann_TIC、The comprehensive proxy database）结合两种同化方法离线集合卡尔曼滤波器OEnKF和混合增益模拟离线HGAOEnKF分别同化出六套过去2000年温度和降水的空间场数据。通过结合气候模型的模拟数据（如温度和降水的先验估计）与代理记录的观测数据生成历史气候状态的重建结果。每次同化过程中，75%的代理记录用于同化，25%的代理记录用于验证。所有的同化结果共有1000个重建实现（10个蒙特卡罗样本，每个样本有100个成员的集合）为了保证结果的稳健性，每个代理记录集都经过随机选择和交叉验证以确保重建结果与实际观测的温度和降水数据具有较高的相关性。结果表明在器测时代与器测温度数据能够达到一致的同化技能，外相关系数、CE（效率系数）和RMSE（均方根误差）指标对比，与器测温度相关性较好，RMSE低，CE值高。在代理数据分布较为稀疏的情况下使用HGAOEnKF方法能够更好的提高同化技能。如全球仅需30个分布合适的代理数据参与同化就能达到0.3左右的相关。

This dataset contains global temperature and precipitation assimilation results spanning the past 2000 years. Three different proxy datasets (PAGES2k, PAGES2k+Mann_TIC, and The Comprehensive Proxy Database) were used, combined with two assimilation methods: the offline Ensemble Kalman Filter (OEnKF) and the offline Hybrid Gain Analog Ensemble Kalman Filter (HGAOEnKF), to generate six sets of spatial field data for temperature and precipitation over the past 2000 years via assimilation. Reconstructions of historical climate states were produced by combining prior estimates from climate models (e.g., for temperature and precipitation) with observational proxy records. For each assimilation run, 75% of the proxy records were utilized for assimilation, while the remaining 25% were reserved for validation. All assimilation results consist of 1000 reconstruction realizations: 10 Monte Carlo samples, each with an ensemble of 100 members. To ensure the robustness of the results, each proxy record set was randomly selected and subjected to cross-validation to guarantee high correlation between the reconstructed results and actual observed temperature and precipitation data. Results indicate that the assimilation skill matches that of instrumental temperature data during the instrumental era. Comparisons based on metrics including the out-of-sample correlation coefficient, CE (Coefficient of Efficiency) and RMSE (Root Mean Squared Error) show that the reconstructed results have good correlation with instrumental temperature data, low RMSE values and high CE values. When the distribution of proxy data is sparse, the HGAOEnKF method can better improve assimilation skill. For example, only 30 appropriately distributed proxy records globally are required for assimilation to achieve a correlation coefficient of around 0.3.

搜集汇总

背景与挑战

背景概述

该数据集通过结合多种代理数据集和两种同化方法，重建了过去2000年全球温度和降水的空间场数据，验证结果显示与器测数据相关性良好，尤其在代理数据稀疏时HGAOEnKF方法表现更优。