Radar interferograms provide precise (millimeter level) measurements of crustal deformation at fine resolution over wide areas. The dominant error source in many interferograms results from spatial heterogeneity of the wet component of atmospheric refractivity, resulting in excess path length of the radar signal propagating through the neutral atmosphere. In this report, we introduce a method to compensate for atmospheric phase artifacts in a radar interferogram using spatially interpolated zenith wet delay (ZWD) data obtained from a network of Global Positioning System (GPS) receivers in the region imaged by the radar. Our method, based on Taylor's "frozen-flow" hypothesis, uses GPS data recorded prior to and after the individual radar acquisition times to infer denser spatial networks of control points for interpolation than is suggested by the sparse and irregular spatial distribution of GPS receivers. We test this approach by correcting phase fluctuations observed in a radar interferogram over an area covered by the Southern California Integrated GPS Network stations, using maps of atmospheric delay interpolated from the denser network of GPS ZWD measurements generated by our method. We find that the largest improvement results from removing an altitude-dependent model derived from the GPS data: the uncorrected interferogram in our example has an RMS fluctuation of 16.1 mm, which falls to 8.6 mm after the altitude correction. A map derived from ZWD measurements recorded only at the instances of radar acquisition reduced overall fluctuations in the data from 8.6 to 8.1 mm RMS, while using the inferred denser network reduces the error further to 7.5 mm RMS. Most of this residual variation is due to decorrelation of the signal and does not result from atmospheric propagation, however. After smoothing to remove decorrelation noise, the residual drops from 4.7 to 3.9 mm (GPS stations only) and then to 2.7 mm RMS when the denser network algorithm is used.