At short wavelengths, the adjacency correction causes strange effects at borders

[rese_support]

For geocoded imagery the number of adjacency zones should be set to N=1.Reason: The problem with adjacency correction at image borders is that information outside the scene is required, as the algorithm employs a moving window corresponding to twice the adjacency range, centered on each pixel.So an assumption is needed to estimate this missing information. If you set N=2 (or any N > 1) the program will calculate the average reflectance in two zones, and the outer zone in touch with the geocoded background will generally have a different average reflectance than the inner zone, even if a reasonable assumption for the background reflectance is being made (e.g. average non-background scene reflectance). For longer wavelength bands the influence of the adjacency effect decreases, so the effect cannot be seen.So our recommendation is to use N=1 for geocoded data. However, you can use up to N=5 and compare it to N=1 results, but you have to omit the scene borders for a quantitative evaluation. Then you will find that the effect of N=1 versus N=5 or N=2 is usually very small.