In order to understand the physical features of the output results of localized mesoscale WRF (Weather Research Forecast) model before the occurrence of short-time severe rainfall in Yunnan Province and to reveal the effect of physical quantities on the prediction of short-time severe rainfall,WRF numerical prediction model was used to simulate five short-time severe rainfall events in the main flood season (June-August) of Yunnan in 2016.Using the high spatial and temporal resolution data output from the model,some physical quantities of water vapor,dynamic and unstable condition classes were calculated for 85 samples in five processes 6 hours before the occurrence of short-time severe rainfall.The distribution characteristics of physical quantities and their relationships with short-time severe rainfall are analyzed by box-line graph,and the thresholds of physical quantities are determined by empirical cumulative distribution function graph.Research shows that water vapor quantity sample values distribute intensively,and the values increase gradually with the approaching of short-time severe rainfall.The median and average values of 6 km vertical wind shear of dynamic class change little with time,and the threshold values of 6 km vertical wind shear of all times are lower than 12 m/s,which indicates that the vertical wind shear is weak before short-time severe rainfall.In the unstable condition class,the convective available potential energy (CAPE) sample data have greater dispersion,so CAPE has no instruction significance for short-time severe rainfall.The data of LI index,K index and 700 hPa pseudo equivalent potential temperature samples have less dispersion,and the upper and lower limits of the median value,average value and threshold of K index increase significantly one hour before the occurrence of short-time severe rainfall.The concentration of K index data is the highest one hour before short-time severe rainfall,and the larger K index value corresponds well with the short-time severe rainfall.The method of estimating the location of short-time severe rainfall by the physical quantity threshold value can improve the prediction performance of Yunnan localized WRF model to short-time severe rainfall.