为研究青藏高原（简称“高原”）东部暴雨云团的局地强降水响应特征，使用FY-2卫星红外通道数据，选择降水开始前4小时直到降水结束后2小时的时段，对高原东部19次典型局地暴雨过程分两种方式进行云团分析，第一种方式为针对测站上空7×7像元范围的云团进行云顶温度变化等相关分析，第二种方式为对关键四省范围内的云团进行识别和追踪，并计算对流云团参数。对暴雨云团的雷达回波特征也进行了分析。结果表明：（1）7×7像元范围红外各通道的云顶温度变化趋势一致，降水阶段云顶温度先降后升，云顶温度梯度先升后降。云顶温度梯度极大值G_max的峰值和半小时内G_max的最大上升变化值ΔG_max均出现在强降水前（0—11 h），G_max峰值次数为1—2次，云顶温度极小值T_min的谷值多出现在强降水之前，T_in曲线斜率大值阶段对应G_max的大值阶段，ΔG_max，G_max和T_min的极值分别可达到22.3℃，48.3℃和-90.3℃。（2）用7×7范围云顶温度及温度梯度建立的降水量级预报方程能较好地模拟小时降水量随时间的变化趋势且有一定的预报提前时间，预报误差在1个降水量级内。在考虑了G_max峰值对强降水的贡献后R²由0.23提高到0.54，模拟的降水量峰值与真值峰值明显接近。（3）对流云团的识别追踪方法更简单有效，对形变较小的云团（相关系数≥0.5）的准确率为100%，对发生了合并或分裂等严重形变的云团（一般相关系数<0.5）的识别结果正确而追踪结果无效；（4）高原东部暴雨云团均为中-β—中-α尺度，水汽柱深厚但强度比低海拔地区更弱，测站暴雨开始之前多数有对流云团覆盖，若对流云团空间参数位置靠近测站，当空间距离至少小于或等于15个像距时降水或强降水将在几小时内产生。（5）暴雨云团在雷达回波图上表现为强降水超级单体风暴特征，且暴雨测站7×7范围G_max峰值时刻对应有回波顶高度（18 dBz）的梯度极大值，红外1、红外2和红外3通道的T_min谷值时刻分别对应回波顶高度极大值和垂直累积液态水含量的极大值。本研究结果对高原强对流云团的识别、跟踪及短时降水预报等具有重要参考价值。

To investigate the characteristics of rainstorm clouds to local precipitation, we analyzed 19 typical cases of rainstorm process in the Eastern Qinghai-Tibetan Plateau from four hours before precipitation to two hours after based on the cloud-top temperature data of FY-2 infrared channels. Two patterns were found, and the first was the trend of temperature at the 7×7 pixels region centered at automatic meteorological stations, and the second was the analysis of convection parameters through identification and tracking convective clouds. In addition, the characteristics of radar echo of rainstorm clouds were also analyzed. In this paper, the maximum temperature gradient on cloud top is represented as G_max, the maximum amplitude of G_max within the half-hour is expressed in ΔG_max, and the minimum temperature on cloud top is denoted by T_min. We found that the variation trend of cloud-top temperature and temperature gradient at infrared channels were consistent, and the cloud-top temperature showed a decrease before increase while the temperature gradient was the opposite in precipitation stage. One or two peaks of G_max and ΔG_max happened about 0 to 11 hours earlier than the heavy precipitation. The multiple regression equation for forecasting precipitation grade was constructed by T_min and G_max. It showed that the model can well simulate the tendency of hourly precipitation and can also predict heavy rain, and the error of hourly precipitation is within one grade of precipitation. The R² of regression improved from 0.23 to 0.54, and the precipitation peak can be better modeled after considering the contribution of G_max peak. The new algorithm of clouds identification and tracking is more simple and effective than other methods. All rainstorm cloud clusters over eastern Plateau are at mid-β to mid-α scale, and their height of water vapor columns are deep but the intensity are weaker than that of the lower altitudes. The precipitation will happen within several hours when the minimum distance between the position of clouds extremum and the meteorological station is less than 15 pixels. The rainstorm clouds have the characteristics of heavy precipitation super cell storm in radar echo, and the extremes of T_min and G_max correspond with the extremes of vertical integrated liquid water content, echo-top height and its gradient. These results are important for identification and tracking strong convective cloud clusters and short-time precipitation forecasting in eastern Qinghai-Tibetan Plateau.