My notes from reading 'An evaluation of void filling interpolation methods for SRTM data'.

The CIGAR version of SRTM contains a number of improvements, mostly around filling voids. I probably won’t be using SRTM for gpxz, but I like their thorough void-filling methodology.

Here’s my notes from reading the paper on the dataset:

Reuter H.I, A. Nelson, A. Jarvis. (2007). An evaluation of void filling interpolation methods for SRTM data. International Journal of Geographic Information Science (21:9, 983-1008).

  • SRTM contains 3,000,000 voids, including up to 9% of Nepal
  • Voids especially in mountainous and desert regions
  • Uses EGM96 vertical datum
  • Process done with tiles with large overlap.
  • They plan to use ASTER DEM to fill holes in the future
  • In an early version they had cliffs between tiles. Need to check for this, and avoid by using large overlaps.
  • Released 2008 so sources might be dated.
  • Nice map of voids, coloured by proportion
  • Log plot of void size
  • Use gtopo30 as filler
  • Quantitative classification of voids into 15 categories
    • Plains, grouped by altitude
    • Hills
    • Mountains
  • Geostatistical methods have a lot of parameters
  • Multiple error metrics.
  • Just from their sample imagery, IDW, spline, and kriging all look good. IDW and kriging look similar.
  • Kriging and spline are usually the best interpolation method.
    • Kriging always best for flat, spline bad for flat.
    • Kriging fine for mountains, spline sometimes a bit better.
    • Kriging only best for small voids.