利用第六次国际耦合模式比较计划（CMIP6）中的18个模式， 基于欧洲中期天气预报中心第五代再分析资料（ERA5）再分析数据对青藏高原夏季降水数据进行了偏差校正， 并从平均降水和极端降水两方面评估了校正前后的CMIP6数据以及单个模式在1979 -2014年的表现。研究结果表明， 该校正方法高度依赖于用于偏差校正的ERA5再分析数据在研究区域的质量， 尽管偏差校正后的青藏高原夏季平均降水的误差和误差率上有所改善， 但在年际时间变化特征方面却不如偏差校正前的数据。大多数CMIP6模式能够较好地模拟1979 -2014年青藏高原上由西北至东南逐渐递增的平均降水空间变化特征。偏差校正前的降水数据在高原上会出现显著的高估， 误差率为60.4%， 经过偏差校正后的数据相对观测数据误差降低， 误差率为-13.9%， 并且偏差校正后的数据与ERA5的平均误差仅为0.003 mm·d^-1， 与ERA5的空间相关性高达0.999。空间趋势方面， 观测数据表明青藏高原大部分地区夏季降水在1979 -2014年呈现轻微增加的趋势， 只有东缘出现明显降低的趋势。偏差校正前后的数据都能够大致刻画出这一空间分布特征， 然而， 未经偏差校正的大多数单个CMIP6模式与ERA5的空间相关系数未超过0.5。与由独立观测降水数据的年际变化特征相比， 偏差校正前的数据高估了高原上的降水量， 而偏差校正后的数据相比观测结果则偏低。通过确定95%分位阈值选取了极端降水个例， 其集合平均极端降水空间分布与年平均降水类似， 也呈西北向东南递增的趋势。部分CMIP6模式较好地模拟了这一特征， 如MRI-ESM2-0（The Meteorological Research Institute Earth System Model version 2.0）和ACCESS-CM2（Australian Community Climate and Earth System Simulator Climate Model Version 2）， 与观测结果的空间相关系数分别为0.851和0.821。但偏差校正后的数据在空间相关性方面下降， 由偏差校正前的0.861降为0.730， 未能准确刻画高原极端降水阶梯式递增的特点。偏差校正后的极端降水数据误差分布与偏差校正前相似， 偏低区域主要集中在高原南部腹地和东部。进一步的极端降水贡献率分析结果表明， 观测结果与CMIP6降水数据均显示1979 -2014年期间极端降水贡献率变化趋势不明显。单个CMIP6模式中， EC-Earth3-Veg（European Community Earth-Vegetation model version 3）和EC-Earth3（European Community Earth Model version 3）及CanESM5（The Canadian Earth System Model version 5）在多个统计评估指标上排名靠前， 展示出较好的模拟能力； IPSL-CM6A-LR（Institut Pierre-Simon Laplace Climate Model 6A Low Resolution）在平均降水误差和极端降水的误差指标上表现出色。

We bias-corrected and assessed summer precipitation data over the Qinghai-Xizang Plateau （QXP） based on 18 models from the Coupled Model Intercomparison Project Phase 6 （CMIP6）.Our assessment of CMIP6 data， conducted for the period 1979-2014， centered on the performance of both the ensemble and individual models.We evaluated the CMIP6 data before and after bias correction， according to considering mean precipitation and extreme precipitation.The results highlight the correction method's dependence on ERA5 reanalysis data quality over the QXP.Although corrected mean summer precipitation over the QXP shows improvement in bias and bias rate， it exhibits inferior interannual time-varying characteristics compared to pre-corrected data.Most of the models were able to better simulate the spatial variability characteristics of mean precipitation over the QXP， gradually increasing from northwest to southeast from 1979 to 2014.Pre-correction precipitation data overestimates precipitation over the QXP with a bias rate of 60.4%， while corrected data is relatively underestimated with a deviation rate of -13.9%.The mean bias of the corrected data from ERA5 is only 0.003 mm·d^-1， with a spatial correlation as high as 0.999.Spatial trend analysis of observed data indicates a slight increase in summer precipitation over most of the TP from 1979 to 2014， with a significant decreasing trend only along the eastern edge.Both pre- and post-corrected data generally capture this spatial distribution， though pattern correlation coefficients of most individual uncorrected CMIP6 models do not exceed 0.5.Comparing with the interannual variability of the precipitation data obtained from observations， the pre-corrected data overestimate the precipitation on the QXP， while the post-corrected data are underestimated in comparison with the observation results.Extreme precipitation is selected by determining the 95% thresholds， a revealing a spatial distribution similar to the mean annual precipitation， increasing from northwest to southeast.This feature is well captured by some models， such as MRI-ESM2-0 （The Meteorological Research Institute Earth System Model version 2.0） and ACCESS-CM2 （Australian Community Climate and Earth System Simulator Climate Model Version 2.0）.Earth System Simulator Climate Model Version 2）， the spatial correlation coefficients are 0.851 and 0.821， respectively， compared with the observations， but the spatial correlation of the corrected data decreases from 0.861 to 0.730， failing to accurately characterize the stepwise increase of extreme precipitation on the QXP.The deviation distribution of the corrected extreme precipitation data is similar to pre-correction data， with lower areas concentrated in the southern hinterland and eastern part of the QXP.The analysis of extreme precipitation contribution shows that both the observation results and the CMIP6 precipitation data indicate that the trend of extreme precipitation contribution is not obvious during 1979-2014.Among individual models， EC-Earth3-Veg （European Community Earth-Vegetation model version 3） and EC-Earth3 （European Community Earth Model version 3） and CanESM5 （The Canadian Earth System Model version 5） ranked high in several parameters， showing better simulation capability， while IPSL-CM6A-LR （Institute Pierre-Simon Laplace Climate Model 6A Low Resolution） ranked high in the mean precipitation deviation and extreme precipitation deviation.