If I load the individual tiles where the gaps appear, it seems that some of the larger tiles along the boundary include white space in the actual image. The screenshot below shows one of the tiles with white space obscuring the mosaic dataset "Footprint" boundaries for the smaller tiles.
So, from what I understand about how the ZOrder field in the "Footprint" boundary is supposed to work, if I set the value lower in the small tiles and a higher for the larger tiles, the smaller tiles (with photographic information throughout the space) should display on top and the larger tiles should be the background. So, I set ZOrder = -1 for all the smaller tiles and 1 for the larger tiles. That seemed to work, but then I when I built overviews and zoomed back out I see the same gaps, though they have shrunk a bit (compare to the first screenshot).
I found a post in the forums describing a problem that seemed to be similar to what was happening for me (http://forums.arcgis.com/threads/48759-Mosaic-dataset-overviews-appear-black-have-gaps-in-some-parts...). So, I followed the advice written in by gsumerling and recreated the mosaic again, this time running the Build Footprints tool after loading rasters into tthe mosaic, though that seems redundant since I had "Update Cell Size Ranges" and "Update Boundary" boxes checked in the Load Rasters window (I also had calculate pyramids and statistics checked, and was careful to leave "Update Overviews" unchecked). Then, I ran Build Overviews and did NOT edit the ZOrder field in the table. Got the same result as in the first screenshot.
Can anyone see where I'm going wrong in the process, here?
From what you describe in your last paragraph it sound as if your build footprints process has not worked correctly. After you run this it should eliminate the white padding. After you run check the footprints. If it is not running as you expect it might be to do with the type of imagery. Is it wavelet compressed (MrSid, JP2 or ECW) if so the ne you need to adjust the parameters as per the discussion on the following web page for lossy imagery http://resources.arcgis.com/en/help/main/10.1/index.html#//009t0000003w000000 This also updates the boundary of the mosaic which is different from the image foot prints.
Also I use the Zorder when I want to push imagery to the back. So I generally apply a value of 1 to the low resolution imagery. However, this is usually handled very well by the minps and maxps values. You can always use the calculate cell size ranges tool http://resources.arcgis.com/en/help/main/10.1/0017/00170000008m000000.htm to help in getting this right after you have loaded Imagery, built footprints, defined overviews and built overviews.
Thanks for the reply, Gordon. But running the build footprints tool (and accepting all the defaults) mungs up the footprints.
I've tried this with another set of rasters - smaller set, lower resolution, and no overlapping rasters - and those footprints get munged up as well. I tried reducing the number of vertices, and that helps but does not eliminate the problem. I did not change the defaults otherwise.
There's another problem I'm having with that second dataset, which Ill probably just write into another post.
Accepting the default values is not a good idea if you are using Lossy imagery or have irregular boundaries on the images. Looking at your screen grabs your imagery is rectangular so you should not need more than 10 verticies on the build boundary tool. Your footprints look as if you are using a value up near 80 for vetices. Please read the details about irregualr and clipped sheets on the following help file http://resources.arcgis.com/en/help/main/10.1/index.html#//009t0000003w000000 to get the best values.
If this is still failing the alternative for you would be to manually edit the footprints which have the white space, to exclude the white space. Then build the boundary and overviews.
For the white spaces, can you indicate what the RGB values are? You would want to adjust the values to account for those values. If you have some artifacts on the edge, then you should consider how much overlap you have. If you can give up some of the data, then I would recommend measuring how much would it take to remove the white and then use the shrink. This will bring the boundary of the footprint file and allow the rasters to display as desired.