Abstract: A three-dimensional variational data assimilation scheme (3DVAR) has been explored for convective scale NWP. In this scheme, a cost function is defined with a background term, an observation term, and a weak constraint term. The background error covariance matrix, though simple, is modeled by using a recursive filter. Furthermore, the square root of this matrix is used to precondition the minimization problem. In its earlier development, only radar radial velocity data could be assimilated. Recent developments include the capability to assimilate reflectivity directly in this 3DVAR framework. Based on this 3DVAR framework, a real-time, weather-adaptive analysis system has been developed for the NOAA Warn-on-Forecast (WoF) project to incorporate all available radar observations within a moveable analysis domain. The system has ability to automatically detect and analyze severe local hazardous weather events at 1 km horizontal resolution every 5 minutes in real time based on the current weather situation, and can identify strong circulations embedded in thunderstorms. The system performed very well within the NOAA Hazardous Weather Testbed Experimental Warning Program and received good evaluations from forecasters. Several real data cases are given in this paper. In the conclusion part, we shall briefly introduce two new methods - ensemble 3DVAR, and hybrid 3DVAR and EnKF schemes and discuss their potential applications for convective scale NWP.