Above is a one-meter resolution aerial photo from the Double Ditch Village in North Dakota on the left. The right side image has height values assigned according to grayscale values, it doesn't look bad from this height.

However, when you zoom in you can see the obvious problem.

    Basically, there are to many spikes in the data. Trying to reduce the number of grayscale values only served to obliterate any distinctive features. Therefore, I think that to successfully model micro-topography (especially sub-meter) there has to actually be micro-topography data. Of course, if micro-topography data exists, then using the vertical exaggeration function in ArcScene will help the viewer to differentiate features below ground as they correspond with the surface.

Images: Terraserver.com