With radial wind data from 18 CINRAD/S and CINRAD/C radars in southwest China, the S-S, C-C, and C-S dual-radar wind field are retrieved and a mosaic of them is made using three-dimensional dual-radar wind retrieving technique considering standard atmospheric refraction in the dynamic earth coordinate. Then eight cloud-resolving scale numerical experiments were conducted with the WRF4.2 model and the WRFDA assimilation system to simulate a rainstorm event in southwest China on July 6-7, 2017 and investigate the impact of three-dimensional variational assimilation of dual-radar retrieved wind with different horizontal resolutions upon the heavy rainfall forecast. The results are as follows. (1) The dual-radar retrieving technique of three-dimensional winds in the dynamic earth coordinate considering the standard atmospheric refraction can be well applied in dual-radar wind retrieving of different-band radars in southwest China. (2) As to the heavy rain band from northeast Sichuan to north Chongqing, eight retrieval-wind assimilation experiments with different horizontal resolutions all significantly improved the direction, precipitation area, range, location and intensity of the heavy rain band and is more close to the real situation than the control experiment. The spatial correlation coefficient and TS score of the 24-hour accumulated precipitation at 06-30 hours valid time are significantly increased with the missing and false forecasting rates both decreased. The improvement is best for the rainfall equal to and greater than 25 and 50 mm. The forecast of wind-field assimilation experiment with a horizontal resolution of 0.25 ° is the best, where an appropriate analysis increment field relative to the background wind field is generated under the assimilation of appropriate resolution wind-field data and the analysis wind field are more close to the real. The wind field structure in the control experiment are continuously corrected in the model integration and resulted in the simulated flow field structures such as vortices, shear lines, high-altitude troughs, wind direction and wind speed being closer to the real situation. (3) As to the heavy rain band from northern Yunnan and southern Sichuan to northwestern Jiangxi, the assimilation of retrieved winds with a resolution of 0.15 ° or higher has a significant negative impact on the prediction results. The negative impacting becomes more significant when the resolution of retrieval wind increases. This is due to the data correlation caused by the assimilation of overly dense retrieval winds. The analyzed incremental fields are falsely and excessively high and ultimately affect the simulated wind field structure and the rain-band position. (4) The assimilation of retrieval wind with different horizontal resolutions has a significant impact on numerical prediction results. Overall, a horizontal resolution of 0.2 ° is the critical threshold for the thinning scheme of retrieved winds in this study, with the condition using a triply nested domain with resolutions of 18/6/2 km and conducting data assimilation in all domains. When the horizontal resolution of the retrieval wind is not higher than this threshold, it often has a positive impact on the prediction results, with a horizontal resolution of 0.25 ° being the best thinning scheme.