Geographic Information Systems Stack Exchange is a question and answer site for cartographers, geographers and GIS professionals. It's 100% free, no registration required.

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

I have a thousand raster tiles in un-projected decimal degrees from which to build a seamless mosaic in a metres-based coordinate system. The final image will be about 70,000px by 50,000px. Which is the better approach, thinking in terms of processing (time) efficiency and quality of result:

  • mosaic into a single image and then transform to the final projection?
  • or transform each tile and then mosaic?
  • (or don't mosaic at all and use a virtual raster table in the final projection?)

I can use either or any of ArcGIS Desktop, GDAL Utilities, QGIS and a have 64bit OS with 12GB RAM.

share|improve this question
I don't the second option makes any sense, there will be edges broken everywhere. Option 3 is just a way of achieving 1. Do a test with 3 and then commit that VRT to a tiled and compressed GeoTIFF if the performance is too slow. – mdsumner Mar 8 '14 at 6:26
up vote 5 down vote accepted

I prefer the third choice. A virtual raster table saves you a lot of disk space.

Mosaicking before transforming is the second best choice. Transformed single tiles have nodata edges, which need a destination format that can work with transparency.

If your raster tiles are paletted, make sure to expand them to RGB or RGBA first. This can already be done with single vrt`s. Otherwise your colours get weired.

share|improve this answer

This is a highly subjective issue, very dependent on input data. If you mosaic after projection, the edges might suffer in quality. For optimal quality, I would use a hybrid approach, of mosaicking a few tiles (3x3? 5x5?) in geographic, with a clip of some of the neighboring tiles (10-15% buffer), then project each cluster (and then clip most of the overlap back out), then either mosaic the pieces, or just manage them in a mosaic dataset from there.

You could also reverse the transformation, using a series of polygons in the target projection, deproject them, and use buffers of the target footprints to identify which tiles to mosaic, so on projection (and with a slight clip) the resulting mosaic tiles were all roughly orthogonal to the target projection (less NODATA on the boundaries, none after a nominal clip). That should save some intermediate processing time.

share|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

Not the answer you're looking for? Browse other questions tagged or ask your own question.