Abstract: High-quality three-dimensional wind field retrieval products from Doppler weather radars are essential data for studying and forecasting mesoscale weather systems. The existing 3D variational wind field retrieval method (3DVAR) mainly uses data from single or multiple radars to conduct the wind field retrieval of small regions. However, for wind field retrieval of large regions using multiple radars, challenges exist, such as high computational requirements and slow processing speeds, which obstruct operational applications. In this study, anan improved method (D-3DVAR) that can obtain high-quality large-scale wind field results is proposed by combining the dual-radar wind field retrieval method, based on the 3DVAR method, using 11 S-band Doppler weather radars and 27 X-band phased array weather radars in Guangdong Province. The optimization process consists of three steps. First, determine the wind field retrieval area based on the location of the weather system to reduce the number of radars; Second, reduce the number of radars by calculating the contribution ratio of the radar network in the dual-radar wind field retrieval; Third, during the dual-radar wind field retrieval process, only select the optimal two radial velocities for each grid point to conduct the wind field retrieval, while the remaining radial velocities do not participate in subsequent wind field retrieval. The results of wind field retrieval after each optimization step are recorded as Majorization 1-3. Taking a squall line process along the coastal area of Guangdong Province on 13 May 2022, as an example, wind field retrieval was performed on the radar network using D-3DVAR, and the retrieval results were compared and verified. The results show that the three-step optimized wind field retrieval method can obtain dual-Doppler radar wind field retrieval results for a large area of the Greater Bay Area, which can better cover the land area of the Greater Bay Area at an altitude of 2 km and most areas within the coastline at an altitude of 4 km. Compared to Majorization 1, the wind field retrieval program execution speed of Majorization 2 and 3 is increased by 3 times. The basic characteristics of Majorization 3 and Majorization 1 are consistent, with wind speed errors less than 1.7 m·s^-1 and wind direction errors within 10°. However, the wind field retrieval result of Majorization 3 can better highlight the characteristics of strong updraft and vertical vortex in the wind field of mesoscale weather systems.