Precipitation outputs from 30 Global Circulation Models (GCMs) of the Coupled Model Inter-comparison Project Phase 6 (CMIP6) were evaluated from 1951 to 2014 over six climate zones in arid Central Asia (ACA) using the Climate Research Unit TS 4.04 (CRU) precipitation datasets as reference. An evaluation framework was constructed taking into account metrics of annual precipitation patterns, annual cycle precipitation statistics, categorical validation and long-term precipitation trend. The performance of GCMs vary from region to region. Compared to CRU, the 30 selected GCMs present considerable wet bias in the Southern Xinjiang (SX) and Hexi Corridor (HC) regions which can be larger than 100%. The simulated precipitation from most GCMs shows a consistent annual cycle shape and closer with that of observations in the western ACA, but fails to capture precipitation peak in NK and NX regions. Compared to NK, CD, TM and NX, all models and the full model ensemble have limited ability in reproducing the Probability Density Function (PDF) of observations over SX and HC. Most GCMs show limited competence to reproduce the long-term precipitation trend with high wet bias over the Hexi Corridor. Based on the comprehensive performance ranking, it is discovered that CESM2, CESM2-FV2, CESM2-WACCM and CESM2-WACCM-FV2 from NCAR, ACCESS-CM2 from CSIRO and CanESM5 from Canadian Centre for Climate Modeling demonstrate better performances across ACA. The ensemble of 30 selected GCMs has limited ability to accurately simulate precipitation according to the above four types of metrics. The results of this paper may provide scientific guidance to CMIP6 end-users when selecting the most suitable GCMs for their specific applications over ACA.

使用气候研究单位TS 4.04（CRU）对1951年至2014年间中亚干旱（ACA）六个气候带的耦合模型间比较项目第六阶段（CMIP6）的30个全球循环模式（GCM）的降水输出进行了评估。降水数据集作为参考。建立了一个评估框架，其中考虑到了年降水量模式，年周期降水量统计数据，分类验证和长期降水趋势的指标。GCM的性能因地区而异。与CRU相比，选择的30个GCM在南疆（SX）和河西走廊（HC）地区表现出相当大的湿偏向，可能大于100％。大多数GCM的模拟降水显示出一致的年周期形状，并且与西部ACA的观测值更接近，但未能捕获NK和NX地区的降水峰值。与NK，CD，TM和NX相比，所有模型和完整模型集合在通过SX和HC再现观测结果的概率密度函数（PDF）方面的能力均有限。大多数GCM在河西走廊上表现出有限的能力来重现长期降水趋势，且湿偏高。根据综合性能排名，发现来自NCAR的CESM2，CESM2-FV2，CESM2-WACCM和CESM2-WACCM-FV2，来自CSIRO的ACCESS-CM2和来自加拿大气候建模中心的CanESM5在整个ACA上表现出更好的性能。根据以上四种类型的指标，由30个选定的GCM组成的集合无法准确地模拟降水。