We combine GPS/MET data from 29 occultations and IGS ground data collected from 160 stations around the world to perform stochastic tomography of the ionosphere with a 4¿20¿20 global grid of voxels extending from 200 to 650 km above the mean surface of the Earth. A correlation functional approach that limits the spatial high frequency content of the images is used, and a Kalman filter is applied in the time direction. The combination of ground and occultation data and the use of smoothing techniques is robust enough for vertical resolution in this four-layer model analysis. We discuss the role of noise on the choice of the correct range of eigenvalues in the inversion problem, and the impact of occultation data, showing that ground data alone is insufficient for vertical resolution even in a three-layer, noise-less simulation.¿ 1997 American Geophysical Union