We are looking into more low cost GNSS rover devices which unfortunately only measure ellipsoidal heights. To use those for mapping applications, we have to post-process those heights to receive orthometric heights (MSL). I have now received an official ascii grid file with the geoid undulation values. The distance between each point feature is between approx. 750 and 900 m:
Would it be a viable approach to correct the height from ellips. heights to MSL by performing a nearest feature search and then subtract the geoid undulation from the ellips. heights? And if this is a viable way to do so in order to get proper MSL heights, would it be smarter to rasterize the grid first in order to increase the performance (ultimately this should end up in a web tool). What would be a suitable raster size if point spacing varies between 750 and 900 m?
I would say a "nearest feature search" is definitively not the way to do it. The geoid grid points are close enough that a final model/grid is probably not that dependent on interpolation method. But sure You would need to interpolate, cubic or bilinear.
For cell size the rule of thumb (I believe) says half of your shortest samling distance, i.e 750/2 meters. Probabaly wise to test grids at 375, 500, 750, 900 and 1000 meters and the 2 interpolation methods. Conforming results with the datapoints You have or better at survey monuments where You have a known geoid height.