A smooth interpolation of a sparse uphole survey may be the main tool available for modeling the near surface in desert areas. An uphole survey does not properly define sharp, nearsurface velocity changes associated with lithologic and topographic variations such as cliffs, canyons, or outcrops, which may be delineated by satellite imagery instead. We present a patchy interpolation technique to build a consistent near-surface model. The algorithm is based on K-means clustering to integrate geological, geophysical, and remote-sensing information. A somewhat arbitrary aspect of the method is the choice of how many clusters are used in segregating the data. Linking this parameter to surface geologic formations may be inadequate because P-or S-wave velocities are unlikely to have a sufficiently unique correlation with geologic age and lithology. We suggest an empirical criterion: improvement in the seismic stack section achieved by processing with parameters derived by various clustering choices and data-type combinations. This approach was tested using field data from Saudi Arabia.