This may seem quite a simple question for which I couldn't find a satisfactory answer. I have been trying to understand how resampling of raster work with ArcGIS 10.3, in both cases of refining and coarsening, especially at corner and edge cells.

For descriptive purposes, I have chosen an input raster with 4 cells of size 1x1m which I resample to a cell size of 0.25x0.25m using different resampling techniques.

Input raster

With bilinear resampling technique, I got this result: Bilinear method output raster

What I couldn't understand here is why the four cells at all four corners have the original values. If it is taking distance weighted average of four input cells, the values would be different. Or ArcGIS is adding three extra cells (of same value as the corner cell) at the corners beyond the edges of the input raster such that four closest cells of the four corner cells on out raster always have the same value of original corner cell? Something like this: enter image description here

One more thing to mention is when I used cubic method, I got the same result as with bilinear method. In this case, the input raster has only four cells; how does cubic method calculates a distance averaged value from 16 cells?


The value of the cell is located at the center of the pixel, not at its corners, and the new cells external to the original cell centers have less surrounding cells to work with. For the 4 pixels in the northwest corner, there is only one value nearer than an original-pixel-width.

For those values to be anything other than '20', one would have to extrapolate using data from more than a pixel-width away.

  • thanks @Dave X for your reply. It completely makes sense. I didn't know that ArcGIS doesn't consider cell-center values which are beyond one- cell-size-width distance on the input raster for interpolation, although there may not be other cells remaining (out of 4 and 16 interpolation cells in case of bilinear and cubic method resp.). ArcGIS help just says 4 and 16 nearest input cells, irrespective of the distance being greater or lesser than the input cell width. Could you please share, where did you find this information? I would be really grateful.
    – ShS
    Jan 18 '16 at 8:58
  • 1
    My understanding comes from the Alvy Ray Smith 1995 "A Pixel is not a little square" alvyray.com/Memos/CG/Microsoft/6_pixel.pdf and the bilinear interpolation algorithm, (following en.wikipedia.org/wiki/Bilinear_interpolation ). For bilinear interp outside of the bounds of square defined by the corner's centers, either 2 or 3 of the inputs the the bilinear interp are undefined, so one either modifies the formula to deal with missing data, (or equivalently, fills in the missing data with the nearest data point)
    – Dave X
    Jan 19 '16 at 11:33

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

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