Abstract: The Variational Doppler Radar Analysis System (VDRAS) has been further developed through improving the quality control and preprocessing of radar data, the warmrain parameterization scheme of a cloudscale model, the interpolation and analysis method of meso-scale first guess, and the calculation schemes of meso-scale background for both coldstart and warm-start cycles, and applied to the real-time retrieve dynamical and thermal fields at low levels in support of the Beijing 2008 Olympics. The developed VDRAS retrieves the three-dimensional thermodynamical fields nearly related to the convective storm initiation and evolvement, including the horizontal wind, vertical velocity, convergence and divergence, perturbation temperature and its gradient, as well as the 12min time increment of those fields at low-level by assimilation to the Beijing and Tianjin S-band radar data via the 12-min rapid update cycling mode using 4DVar and the cloud-scale model with a simplified warm-rain parameterization scheme. The analysis of real-time retrieval results for the two storm cases during the 2008 Olympics shows that retrieved dynamical fields can indicate horizontal convergence, vertical updraft, storm outflow and their change with time at low-level. The analysis of retrieved thermal fields can indicate cold pool and its change with time, as well as the location and intensity of gust front nearly related to storm development. The VDRAS retrieval is very consistent with the conception model concerning the relationship between the storm development and the cold pool, gust front, convergence and updraft that found by many former studies. Preliminary comparison and verification on the retrieved wind and temperature fields from the VDRAS against observations from a boundary layer wind profiler, the Auto Weather Stations and a groundbased radiometrics indicate the retrieval results are close to observations. Theretrieved wind can reveal the vertical shear of low-level wind that can also be observed from the profiler and is a very important factor with respect to storm development, and the retrieved temperature can reveal distinct changes in the air temperature with time that can also be observed from the AWSs while a storm passes through. The retrieved low-level wind speed is smaller than observations from the profiler and has the maximal bias of -1.5 m/s and the maximal root mean square (rms) error of 2.5 m/s. The maximal bias and rms error of the retrieved wind direction are 20 and 45 degrees against observations from the profiler, respectively. The retrieved lowlevel temperature has the maximal bias of -1.9 Celsius degrees and the maximal rms error of 2.8 Celsius degrees, and is lower than observations from the radiometrics.