[gdal-dev] Fwd: gdalwarp -r average bug?

[Stephen Roecker]

Out of curiousity the other day I compared the results of gdalwarp (-r
average) against  the raster R package aggregate(fun=mean) function
for aggregating a raster to a coarser resolution. I was suprized how
different the results of gdalwarp were from raster. When zooming in
and manually averaging the overlapping cells, the results of gdal were
off.

The elevation difference between the raster and gdal aggregated
rasters only averaged 0.66 meters, but had a max of 12 meters. Also
when subtracting the raster and gdal aggregated rasters, the resulting
subtracted layer looked like a hillshade, suggesting the GDAL
aggregated raster was shifted. However the raster and GDAL rasters
overlapped perfectly, so I can only assume the shift occured during
the aggregation process.

Can someone explain gdal's behavior to me? Why the difference, is this
a bug in gdal? gdalwarp claims it's averaging all the overlapping
cells except the NA. That doesn't seem to be the case. FYI I'm using
GDAL 1.10.1

See a reproduceable R example below.

Stephen

library(gdalUtils)
library(raster)

src_dataset <- system.file("external/tahoe_lidar_bareearth.tif",
package="gdalUtils")
test <- raster(x=src_dataset)

writeRaster(test, "test.tif", overwrite=T)
gdal_setInstallation(search_path="C:/ProgramData/QGIS/QGISDufour/bin",
rescan=T, verbose=T)
gdalwarp(srcfile=src_dataset, dstfile="test_gdal.tif", of="GTiff",
r="average", ot="Float32", tr=res(test)*3, overwrite=TRUE,
verbose=TRUE)

aggregate(x=raster(src_dataset), fact=3, filename="test_raster.tif",
format="GTiff", NAflag=-99999,
  progress="text", overwrite=T)