The U.S. Army Research Laboratory has developed a 3-dimensional wind field (3DWF) model for microscale applications over complex terrain and in urban domains. The model is a high-resolution mass-consistent model with a very efficient multigrid computation method for real time applications. In the past, this type of model has usually been initialized with one or more observed wind profiles or with the profiles from a mesoscale numerical weather prediction model. We propose a new method to initialize the model using the wind observation data from a Doppler lidar. This method first retrieves the horizontal wind from the radial wind in the lidar scanning plane. The horizontal wind field is then fused in a mass consistent model using a minimization algorithm. The accuracy of the model prediction in an urban domain is tested using the data collected during JU2003 in Oklahoma City. A comparison using the traditional and the new initialization methods will also be presented.