I have a 1m bare-earth DEM, derived from LiDAR data, representing a terraced landscape. For the sake of any subsequent use of the DEM (e.g., least-cost path, cost surface analysis, and the like) I am wondering if it is sound/necessary to 'blur' the DEM in order to de-emphasize any possible 'noise' produced by the terrace fields (e.g., sharp changes in slope values due to the terrace walls). If the answer is yes, what type of approach is better: something like a low-pass filter (I am not familiar with it, and how it can be performed in ArcGIS 10.1) or should I go back to the LiDAR data and create a bare-earth DEM from scratch with a larger cell size?

  • Can you elaborate on what you mean by noise? Maybe post a screenshot showing an example? I'm not sure I understand what you mean or would agree with treating the terraces as 'noise'. You can, of course, generalize the DEM to whatever degree you want, but if you eliminate those break lines/areas you're basically increasing the overall slope value to make up for it.
    – Chris W
    May 4 '15 at 21:50
  • Thanks for your reply. Sorry if I was not clear. Since the use of the DTM is aimed to archaeological purposes, and since we do not know if the terraces were already there 8000 years ago, I was wondering if it is possible to 'down-weight' the effect of the terrace walls on the DTM. What I am worried about is what effect (if any) can the terrace walls have on, e.g., the calculation of a least-cost path between (archaeological) sites. Hope this manages to make things a little bit clearer.
    – NewAtGis
    May 5 '15 at 17:59
  • Well, that would depend on how you calculate the cost values. Assuming slope is to be a factor, and at a large scale, then yes, presumably a sudden jump from 2% to 50% or whatever could contribute to a barrier. At the same time, if smoothing them out results in an overall slope of 20% when it's really mostly 2% with a few breaks, that's going to skew your cost as well. I would think there would be ramps or some other path between terraces to allow for a flow path. But I don't know the scale or details of what you're looking at (and would think the terraces were somewhat datable).
    – Chris W
    May 6 '15 at 7:27
  • 2
    @AndreSilva You have to consider it being represented as a raster, so the cell size (resolution) dictates slope distribution. If you have fine enough resolution to make out the terraces, then say 90% of your cells are 2% slope while 10% are something really high. If you smooth out/decrease the resolution, depending on how much, 100% of your cells are the constant top to bottom slope value or some other breakdown (50% @ 2% slope while 50% are not as high as the original 10% terraces but higher than 2%). You're right the actual overall slope doesn't change, but the representation values do.
    – Chris W
    May 11 '15 at 19:55

One would wonder if a 1m resolution is the appropriate scale to be addressing you question. Bigger is not always better.

If the scale is too fine and is picking up undesirable detail, a reasonable approach would be to figure out what scale supports your question and generalize the dem accordingly. This can be done through a resampling, with a bilinear or cubic convolution, approach.

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.