GNSS processing for POD suffers from systematic errors when the location of the instantaneous phase center is not known with a few millimeter accuracy. The traditional approach to model the GNSS phase center origin and variations is to determine antenna phase maps iteratively from the residuals. A more direct approach consists in modeling the antenna phase map through an expansion in well-chosen Zernike polynomials, in order to reduce potential correlations with dynamic modeling errors. This strategy has been applied to derive the GNSS antenna phase maps of several altimetry missions such as Sentinel-3A, Sentinel-6 MF and SWOT. The derived phase corrections were tested on CNES POE-F GNSS-only reduced-dynamic orbit solutions to assess their impact and validate their benefits owing to independent SLR observations.