So the most probable reason for results looking stepped in the SDR’s
sed_export.tif but not the NDR
n_export.tif is that SDR’s calculations include USLE, a calculation that is dependent on the on-pixel slope. NDR does not use USLE. Outside of that, both models have similar approaches to how the upstream areas of a pixel (including upstream slopes) are considered.
What’s interesting about this particular case is that your DEM is stepped. Pixel values in your DEM are all integers, and so because of the relatively fine resolution of the dataset, we end up with long, narrow plateaus that then have a higher slope at the edges of the steps, and so the slope calculations (and, ultimately, USLE) are higher at the edges of the steps and all of this results in this sort of stepped effect in the output. Neither SDR nor NDR does any interpolation, but I suspect that if you were to smooth out your DEM, you’ll see a reduction in this stepping of outputs in the
One other thing worth noting is that although this stepped effect is present, I’m only seeing it in
sed_export.tif when I force the color range in QGIS to weight towards the lower end of the pixel value spectrum (for future us, I used QGIS’s pseudocolor, Quantile color mode), and the absolute difference in pixel values across the steps are pretty small relative to some of the other pixel values elsewhere in the raster.
So all of that is just to say that the steps seem like a reasonable consequence of SDR’s usage of USLE given that your DEM has integer values (and thus lots of small, narrow plateaus) and so smoothing out your DEM should avoid this stepped appearance in SDR outputs.
Hope this helps!