气象数据是水文过程研究的关键因素，但是因为流域地形复杂，上游没有足够实测数据的气象站点，流域水文过程研究在较大程度上受到了限制。以玛纳斯河流域为研究区，研究对象选取CMFD、ERA5-Land和CFSR 3种再分析数据，构建基于物理过程的SWAT模型和基于数据驱动的AdaBoost模型，验证不同数据集在2种类型水文模型中径流模拟的适用性，选用纳什效率系数（NSE）和决定系数（R²）进行定量分析。结果表明:（1） 数据集在AdaBoost模型中的表现要好于SWAT模型，各数据集和气象站NSE与R²均有所提升，AdaBoost模型由于对输入数据的限制更少其更适用于资料缺乏地区。（2） 验证期ERA5-Land在2种模型中拟合效果均为最好（ERA5-Land+SWAT：NSE=0.83，R²=0.85；ERA5-Land+AdaBoost：NSE=0.87，R²=0.87），再分析数据集ERA5-Land可为西北实测气象资料不足的干旱区径流模拟提供参考。

Meteorological data is a crucial factor in the study of hydrological processes. However, due to the complex terrain, meteorological stations in the upper reaches of the basin are scarce, limiting the study of hydrological processes in the basin due to the lack of measured data. This paper takes the Manas River Basin, Xinjiang, China as the research area and selects ERA5-Land, CMFD, and CFSR reanalysis data for analysis. The SWAT model, based on physical processes, and the AdaBoost model, based on data-driven approaches, are constructed to verify the applicability of different datasets in runoff simulation for two types of hydrological models. The Nash efficiency coefficient (NSE) and the determination coefficient (R²) are selected for quantitative analysis. The results show that: (1) The performance of the datasets in the AdaBoost model is better than that in the SWAT model. During the verification period, the NSE and R² of the ERA5-Land dataset increased by 4% and 2%; the NSE and R² of the CFSR dataset increased by 14% and 15%; the NSE and R² of the CMFD dataset changed by -10% and 8%. The NSE and R² of the meteorological station data increased by 8% and 10%. For datasets lacking data, the AdaBoost model is more applicable due to fewer restrictions on input data. (2) In the AdaBoost model, the simulation accuracy of all datasets decreased to a certain extent during the validation period, with CMFD showing the most significant decrease and ERA5-Land the least. These results indicate that the generalization ability of the AdaBoost model is weak. (3) Using ERA5-Land, CFSR, CMFD, and meteorological station data as inputs for the AdaBoost model, the simulation results show that ERA5-Land achieved good results during the verification period. The simulation accuracy of CFSR is comparable to that of meteorological stations, while CMFD performed the worst due to its inaccurate description of meteorological data in the mountainous area of the Manas River Basin. The reanalysis dataset ERA5-Land can provide a reference for runoff simulation in arid areas with insufficient measured meteorological data in northwest China.