I have previously run my analysis using the default nearest neighbour resampling method for upscaling the the ALOS-2 1.5L from 25 to 100m resolution and geocoding, using the SAR-mosaic method (after co-registration and applying the improved lee sigma speckle filtering with a 3x3 target window). This gave acceptable looking results, but I wonder if the bilinear or cubic convolution interpolation methods might be more appropriate for geocoding to a lower resolution?
I don’t think Nearest Neighbor is appropriate when downsampling, or going from higher to lower resolution. Nearest Neighbor just uses the closest pixel to the center of the lower resolution pixel basically, so I think you would potentially be losing information. Easiest way to test, is just run multiple resampling methods and see if/how it impacts the results.
1 Like
Thanks for your response. I ran it with bilinear also, and the results look very similar, although the maximum data values across the image are reduced (and the minimum increased). I think you’re right, and it would make more sense to use another resampling method that does some averaging of the data.