青藏高原东部及周边地区地势呈阶梯状分布， 是天气预报模式预报的难点。为了了解新版GRAPES_Meso模式在该区域的预报性能， 利用统计方法对比了GRAPES_Meso模式新版（V3.1模式）与旧版（V2.5模式）在2011年夏季（6-8月）预报结果， 还对比了V3.1模式在高、 中、 低三种地形高度的预报差异， 得到以下主要结论： （1）同化系统升级后， 各要素的误差明显减小， 初始场质量有了较大提高， 尤其在对流层中高层， 各要素的改善幅度依次为相对湿度>纬向风>经向风>位势高度>温度； （2）对于降水预报， 无论是6 h还是24 h累加降水， V3.1模式的TS评分都高于V2.5模式， 尤其是大雨和暴雨预报， V3.1模式能明显减小漏报率、 提高日平均降水准确率， 但对空报率和云南以西的虚假降水改善不明显； （3）V3.1模式预报的位势高度、 温度、 风场和相对湿度误差垂直分布廓线与V2.5模式的不同， V3.1模式可以改进整层的经向风、 纬向风和相对湿度、 对流层中高层的位势高度、 500 hPa以下的温度， 但位势高度误差增长速度大于V2.5模式， 对流层中低层的南风预报偏强， 500 hPa以上相对湿度误差仍很大； （4）V3.1模式在不同地形高度的预报结果对比发现， 除700 hPa的位势高度和温度外， 三种地形高度的各要素误差都是在初始时刻差异最小， 随着预报时长增长， 误差差异不断增大， 这表明V3.1模式各要素的误差增长受地形影响明显。

The eastern Qinghai-Xizang Plateau and its surrounding areas, which have a character of a ladder-like topography distribution, are not only the strong signal areas for many significant weather systems upstream, but also the difficulty of many weather forecast models. In order to understand the prediction performance of GRAPES_Meso V3.1 in these areas, two forecast outputs of GRAPES_Meso V3.1 and V2.5 in  summer (from June to August) of 2011 are compared using statistical methods. The main conclusions are shown as follows: (1) Since assimilation system is upgraded, the initial fields have been greatly improved, and the mean error (ME) and root mean square error (RMSE) of various elements are reduced obviously, especially in the middle-upper levels of troposphere. According to the error-decreasing amplitude, various elements are sorted from large to small as follows: relative humidity, zonal wind, meridional wind, geopotential height, temperature. (2) TS scores of the V3.1 model, either every 6 h or 24 h accumulated precipitation forecast, are higher than that of V2.5 model, especially 25 mm and 50 mm precipitation forecast. The missed event rate of V3.1 model has improved significantly, and the average daily rainfall is better than that of V2.5 model, but it can not improve the false precipitation in the west of Yunnan. (3) The vertical-error profiles of V3.1 model for geopotential height, temperature, wind and relative humidity forecast are different with V2.5 model. V3.1 model mainly improves the whole-layer meridional wind, zonal wind and relative humidity, as well as geopotential height in middle-high-layer troposphere and temperature below 500 hPa. It is noteworthy that the geopotential height errors of V3.1 model are growing faster than that of V2.5 model and new model forecast stronger south wind than true wind in low-layer troposphere. (4) Compared forecast results in three different terrain height of V3.1 model, it is found that, in addition to the geopotential height and temperature on 700 hPa, the differences of the same element error are small in the initial moment, but gradually expand with forecast length increasing. Those show that the errors of V3.1 model are affected significantly by topography.