Determining the Earth's surface topography and deformation with interferometric synthetic aperture radar involves measurement of phase, which, for a typical coherent radar signal, can only be done modulo 2π. The cycle of ambiguity inherent in the phase measurement has to be unwrapped over all observation dimensions (e.g., azimuth, range, and time) to remove the 2π ambiguity of the phase measurements. For a time series of SAR images, useful for reducing noise in topographic applications or measuring time-varying surface deformation, the necessary steps to connect ambiguous radar phase measurements are more challenging, and the operation may be termed 3-D phase unwrapping. We describe a 3-D unwrapping approach using an extended Kalman filter. Our approach readily exploits existing information, and is robust in the presence of noise. For all tested data sets, it provides improved accuracy compared to existing approaches.