陆面过程模型需要高质量的气象强迫场数据，气象数据不确定性对陆面过程模拟结果影响显著。本研究基于CMIP6全球气候多模型模拟数据集，使用贝叶斯模型平均法（Bayesian Model Averaging，BMA）对多模型数据进行后处理，以量化并校正气候模型预估误差。BMA多模型融合数据再经过Schaake洗牌法（Schaake Shuffle）处理，可在保持气候数据趋势性和随机性的同时，也保留空间分异规律和多个变量之间的内在相关性。利用Schaake Shuffle方法，可建立月尺度低分辨率数据与3小时尺度高分辨率数据之间的对应关系，实现气象数据时间和空间上的降尺度，生成全球气温、降水，中国区域气温、降水、气压、风速、湿度、下行长短波辐射的多模型集合数据。

除了可以用于陆面过程和水文模拟之外，多模式气象数据集也可直接用于气候分类，分类结果能够直观展示未来气候变化的空间格局。本研究运用BMA融合CMIP6多模型数据，进行多个历史与未来情景的柯本气候分类，揭示了未来不同排放情景下的气候类型变化趋势。

本研究主要结论如下：

（1）本研究选用的CMIP6各模型模拟气温的精确度优于降水，且在气温模拟中，各CMIP6模型间准确度更相似，模型间的差异性明显小于降水变量。模型在不同地区的模拟质量有显著差异，在地形起伏的地区模拟偏差较大。BMA输出的结果模拟质量在气温和降水两个变量上优于任何选用的单个模型，且有效改善了模型在不同地区模拟质量的差异。

（2）原始CMIP6模型数据及BMA输出的数据经过Schaake Shuffle方法处理后再通过分位数映射（Quantile Mapping）方法进一步进行偏差校正，以获得高分辨率多模式气象数据集。与历史数据对比验证的结果表明，本研究生成的数据既能保持空间分异特征和多变量间的相关性，又能降低气候模型模拟误差，还保持了集合成员的随机性与趋势性。本研究生成的高时空分辨率多模型融合气象强迫场数据有助于提高陆面过程模型的性能，有助于支撑气候变化相关研究。

（3）本研究以BMA方法生成的全球气温、降水数据为输入，生成未来多个情景的柯本气候分类，统计全球范围内柯本气候分类的变化。结果表明，未来热带气候类型（A）、干旱气候类型 （B）面积增加，寒冷气候类型（D）、极地气候类型（E）面积减少，温暖气候类型（C）面积基本持平，全球气候向变暖方向发展，夏天的温度将变得更高，气候变化敏感区主要在高海拔地区、高纬度地区。基于BMA方法生成的气象数据进行柯本气候分类的准确性比基于单一模型数据进行柯本气候分类的准确性更优，这得益于BMA方法在提高数据模拟质量及降低不确定性的优势。

Land surface models require high-quality meteorological forcing data, and the uncertainty in meteorological data significantly affects the simulation results of land surface processes. In this study, based on the CMIP6 global climate multi-model simulation dataset, the Bayesian Model Averaging (BMA) method was employed to post-process the multi-model data to quantify and correct the estimation errors of climate models. The BMA multi-model fusion data was then processed using the Schaake Shuffle method to preserve both the trend and randomness of climate data while retaining spatial  heterogeneity and intrinsic correlations between multiple variables. By utilizing the Schaake Shuffle method, a correspondence between monthly low-resolution data and 3-hourly high-resolution data can be established, achieving temporal and spatial downscaling of meteorological data. A multi-model ensemble dataset of global temperature and precipitation, as well as a regional dataset for China including temperature, precipitation, air pressure, wind speed, humidity, and downward longwave/shortwave radiation, were generated in this research.

In addition to being used for land surface process and hydrological simulations, multi-model meteorological data sets can also be directly used for climate classification. The classification results can intuitively demonstrate the spatial patterns of future climate change. In this study, BMA was applied to integrate CMIP6 multi-model data and performed Köppen climate classification under multiple historical and future scenarios, revealing the trends of climate type change under different emission scenarios.

The main conclusions of this study are as follows:

(1)The accuracy of temperature simulation by the selected CMIP6 models  is better than that of precipitation, and the accuracy among different CMIP6 models is more similar in temperature simulation, with the differences between models being significantly smaller than those for the precipitation. The simulation quality varies significantly across different regions, with larger biases in areas with complex terrain.  The simulation quality of the results output by BMA is superior to any of the individual models selected for both temperature and precipitation, and effectively improves the regional differences in simulation quality.

(2)The original CMIP6 model data and the data output by BMA are further processed through the Schaake Shuffle method and then further bias-corrected by the Quantile Mapping method to obtain a high-resolution multi-model meteorological dataset. Comparison with historical data shows that the data generated in this study can maintain spatial heterogeneity and the correlations between multiple variables, reduce climate model simulation errors, and retain the randomness and trend of ensemble members. The high spatiotemporal resolution multi-model integrated meteorological forcing data generated in this study can help improve the performance of land surface models and support climate change-related research.

(3)Using the global temperature and precipitation data generated by the BMA method as input, Köppen climate classifications for multiple future scenarios were generated, and changes in Köppen climate classifications on a global scale were statistically analyzed. The results indicated that the areas of tropical climate type (A) and arid climate type (B) will increase in the future, while the areas of cold climate type (D) and polar climate type (E) will decrease. The area of temperate climate type (C) will remain stable overall, and global climate will trend towards warming, with higher temperatures in summer. Climate change sensitive areas are mainly in high-altitude and high-latitude regions. The accuracy of Köppen climate classification using meteorological data generated by the BMA method is superior to that based on single model data, benefiting from the advantages of the BMA method in improving data simulation quality and reducing uncertainty.