In order to explore the impacts of urban agglomeration in the Guangdong-Hong Kong-Macao Greater Bay Area on localized extreme heavy rainfall, this paper conducts 3DVar assimilation of Doppler radar data to simulate an extreme heavy rainfall event that occurred in the Greater Bay Area on 22 May 2020, based on the WRF-ARW mesoscale numerical model, GSI-3DVar assimilation system and ERA5 reanalysis data provided by ECMWF. The influence process and mechanism of urban area on local extreme rainfall are studied. The results show that, compared with no assimilation of any observational data, adding the assimilation of radar reflectivity and radial wind data can improve the simulation ability of precipitation, especially for the heavy rainstorm magnitude of more than 250 mm. Observation and control experiment jointly show that the urban agglomeration in the Greater Bay Area acts as a local “heat source”, increasing the temperature of the boundary layer through sensible and latent heat, resulting in significant heat island effect, which then strengthens the convective instability within lower atmosphere. On the other hand, strong friction dissipation reduces the wind speed in boundary layer so that it catches more warm and moist air within the urban area, forming stronger thermal instability and moisture convergence, and then leading to the rainfall center located at the inner part of the edge of urban area. The sensitivity experiment (i.e., removal of urban land use) further shows that friction dissipation caused by urban underlying surface affects dynamic thermodynamic environments in boundary layer below 800 hPa, resulting in stronger southwesterly and unstable atmospheric condition over downstream of the urban area in urban removal experiment. Also, the convection is lifted by the local topography, enhancing vertical upward movement, which finally results in stronger rainfall intensity than the control experiment and the location of the rainfall area more inclined to the downstream of the urban area.