1

I'm looking into some global DEMs (like ETOPO1 - http://www.ngdc.noaa.gov/mgg/global/global.html), and noticed that the statistics for the elevation values can change quite a bit after reprojecting the raster.

The original data comes in a LatLong pseudo-projection. Its stats are:

from GRASS 7.1 r.univar
min: -10803
max: 8333
mean: -1892.40422534
stddev: 2649.98339303

from gdalinfo -stats
min: -10803.000
max: 8333.000
mean: -1892.404
stddev: 2649.983

Now if I project the raster to a cylindrical equal-area projection:

GRASS
min: -10803
max: 8333
mean: -2382.28934158
stddev: 2508.93105538

gdalinfo
min: -10803.000
max: 8333.000
mean: -2382.289
stddev: 2508.931

So the projection process is causing this, although I'm using a nearest neighbor method, so the values shouldn't change.

I keep thinking on the elevation values as a big array of values, and that these values shouldn't change. Right?

I would expect some change with other interpolators, but not nearest neighbor. Or am I missing something really basic here?

In any case, this raises the question: which one is correct? (ignore the fact that the mean value doesn't really mean much due the bimodal distribution of elevations on Earth)

Carlos Grohmann
  • 103
  • 5
  • 2
    Those statistics are based on pixels. The number and spatial distribution of pixels will change significantly when going from lat-lon to a CEA projection. With that, the statistics calculated on the pixels will change. – Mikkel Lydholm Rasmussen Nov 17 '15 at 13:29
  • See also http://gis.stackexchange.com/questions/64150/why-are-raster-z-values-changed-when-reprojecting – Anna Nov 17 '15 at 14:52
  • See also all the answers in the grass-user mailing list for the same question: https://lists.osgeo.org/pipermail/grass-user/2015-November/thread.html#73319 – markusN Nov 28 '15 at 23:54

0 Answers0