In this paper, the PODEn4DVar-based radar data assimilation scheme (PRAS) was improved according to the theory of NLS (Non-Linear Least Squares)-4DVar (four-dimensional variational analysis) scheme. This work aims to deal with the application problem of PRAS under highly nonlinear conditions. As a result, a new radar data assimilation scheme, i.e. NLS-4DVar-based radar data assimilation scheme (NRAS), was developed. To evaluate whether NRAS can further improve the performance compared to PRAS, the Observing System Simulation Experiments (OSSEs) and real radar data assimilation experiments for two heavy rain events (July 8, 2010, central China;March 30, 2014, southern China) were conducted in this study. The results demonstrate that, for both the OSSEs and the real radar data assimilation experiments, NRAS can further improve the assimilation result in comparison to PRAS. By increasing iteration times, NRAS can adjust the wind field and water vapor field. This leads to further improvements on the forecast of intensity and location of the rainfall. However, with increases in the iteration times, the adjustment for the initial condition by NRAS becomes smaller, which leads to a smaller improvement on the rainfall forecast. The results indicate that NRAS can effectively deal with the application of PRAS under highly non-linear condition. With fewer iteration times, NRAS can obtain approximate convergence result. NRAS is expected to better assimilate radar data in numerical weather predictions, and thus further improve the prediction of meso-micro scale weather systems.