I am still interested in this topic. Where can I find theoretical information about extrapolating, interpolating.., or some other technique to calculate the gain at any point (azimuth, elevation) based on the 2 fundamental antenna pattern cuts ?
y26rin: You're basically asking to 'manufacture' information out of thin air. Unless the pattern is so simple (non-complex) that you could simply eyeball it with something like MS-Excel's curve fitting.
Can you model the antenna in NEC and then try to calibrate the results against the two cuts ? The problem is that you'll need to decide which is calibrating which...
It's pretty straight-forward, but it's also a GUESS and it ASSUMES that the pattern is non-complex. In other words, the intervening data is 'manfactured'. Never forget this!
The two slices (provided by the manufacturer) are orthogonal and provide a sample of the surface (like a couple of 'mug shots'). Just as a trained human (with antenna experience) can view the two slices, mentally interpolate, and thereby come to visualize the entire pattern; MS-Excel can be pressed into service to do the same thing explicitly with the advantage of being able to share and analyze the 'manufactured' data.
The data structure in MS-Excel will be a table in the form of a Mercator projection with (typically) azimuth (0 to 360) along the horizontal and elevation (+90 to -90) along the vertical. 5 degree steps is reasonable. The two available slices will (typically) provide a row of real data at El=0, and two columns of real data at Az=0 and Az=180.
Depending on the type of antenna, there might be some axial symmetry that can be used to fill-in the remainder of the pattern very easily. Other antennas will require a great deal of effort (possibly trial and error) to produce a reasonale guess. It might be EASIER to simply use EZNEC.
Once the data is in MS-Excel, then a sanity check can be made to see if the pattern integrates (numerically) to something reasonably close to 0dBi. Don't forget to weight the data to account for the overrepresentation of the zenith and nadir data (just like Greenland looks too big on a flat map).
I've done all of the above to analyze an array of antennas. The interpolation (of the two slices) was trivial since (luckily) each antenna was symmetrical about its boresight. The integration worked out to about 95% efficiency (a reasonable value). Once I had the data in MS-Excel, then I could start playing with the array aspects of the problem.
The same system was later analyzed using GTD/UTD (ALDAS) and the predicted patterns were extremely similar to the early predictions made using MS-Excel. This similarity is due to the nature of this particular system and it would not be true in the general case.