Caution: only tested with
MSFS 2002 and 2004
 

Mesh Overview: Resolution
The images below show the effect of taking data elevation measurements of the same real-world terrain at 90m and 30m intervals.
 
low resolution source data high resolution source data

earth's surface - 90 m data spacing

earth's surface - 30 m data spacing

Even at this stage, a good deal of detailed information about the shape of the earth is lost in the sampling process. At least using data available to the public.

source data available for processing

source data available for processing

These figures suggests what the world looks like when reconstructed using the available data. This is the theoretical ideal we can hope for in the sim.

sim terrain data
resample spacing approximately same as original

sim terrain data
resample spacing greater than original
rarely used, but still good

The developer specifies the final resolution for the sim using an LOD value, but has no further control over the locations of the samples used to create the new data. MSFS uses its own internal grid system, which does not match any commonly available source resolution, so some additional accuracy is lost at this stage. We do not know the details of the conversion process, but the green shaded areas suggest the nature of the additional distortion added here. This is a bit like recording the original data from the real terrain, with a bit of math thrown in. Notice that the location of the peaks and valleys generally shift at each stage in this process. This makes it very difficult to insure proper alignment with other features in the sim.

sim terrain data
closer resample spacing
closer to model but not to actual surface

sim terrain data
closer resample spacing
closer to both model and actual surface

Even with extremely small sample spacing and including sample points precisely at the tops of peaks and bottoms of valleys, this is the best we can hope for. The LOD/final resolution does little to help if the source data is not accurate. Too high an LOD ("oversampling") can actually distort the data!

Screenshots readily support these diagrams.