This forum is shutting down! Please post new discussions at

Help with DEM for SDR

I have been trying to get the SDR to work for a while now, and I think I have identified one of my many problems with the model, the DEM. I believe it has something to do with the flow accumulation, but I am not sure. Are they any steps I could follow to verify this hypothesis? 


  • swolnyswolny Member, NatCap Staff
    edited January 2018
    Hi @ruralwater -

    Can you describe what's going wrong when running the model? That might help us point you in a good direction.

    Meanwhile, I'll say a few words about how I handle DEMs, and perhaps something in there will be useful.

    - If your DEM is a mosaic of tiles, and you're using Arc, make sure you've used the Bilinear or Cubic Resampling Technique when projecting to a new coordinate system, else you'll end up with a checkerboard pattern in your derivative flow rasters.

    - DEMs pretty much always have pits/sinks that need to be filled. These will manifest in the model results as a stream network that's all chopped up in pieces. The best tool we've found for filling pits is the QGIS Wang and Liu tool. You might need to do this more than once.

    - After filling pits, and before putting a DEM into a model, I generally run Flow Direction and Flow Accumulation on the DEM to see how the stream network looks. Much easier to debug at this point, rather than running the model over and over. Of note, InVEST has a tool called RouteDEM, which will produce Flow Dir and Flow Acc outputs, as well as generate stream layers, so you can see exactly how the models will treat the DEM without actually running the hydro model.

    - Then there's the Threshold Flow Accumulation (TFA) value. This needs to be selected so that the resulting stream layer comes as close as possible to the real-world stream network. Here's where RouteDEM is handy, since it can output many streams over a range of TFAs, so you can compare to a real-world stream map.

    - Also, do look at the intermediate_outputs folder created by the model. In there you'll find flow direction and flow accumulation maps. The stream.tif map is created in the main output folder (along with sed_export.tif etc.)

    Those are the biggies. With more information on what you're seeing I can help more.

    ~ Stacie

    Post edited by swolny on
  • Hi Stacie!

    1) I did use ArcGIS to create a mosaic of tiles (Three counties @ 10m resolution) - However, I don't remember which re-sampling method was used.

    2) After I mosaicked the tiles, I ran the fill tool on the mosaicked dem in ArcGIS (see attached image). 

    3) Then I ran the Flow Direction tool on the filled mosaicked dem (see attached image).

    I don't know if you will be able to tell from these images, but should I continue to run fill to smooth out these sinks or will I have to start from the beginning re-mosaic the tiles? 

    550 x 515 - 116K
    588 x 515 - 80K
  • swolnyswolny Member, NatCap Staff
    These don't look obviously like you've got the checkerboard problem (could be more obvious zoomed into the original raster, but if you don't see them there that shouldn't be a problem.)  You generally can't tell from just looking at images if there are pits. They show up when generating streams and tracing a flow path. So I'd recommend running RouteDEM on these to generate streams and see if they are continuous and look reasonable. 

    As an aside, it looks like your DEM area doesn't represent an entire watershed, but is cut out along non-natural boundaries (administrative or something.) If you need to use a non-complete watershed, just realize that you won't be representing all of the processes contributing to the watershed or whatever point of interest might be at the outlet. For example, some sediment will just flow off the map, instead of to a stream, and so the export value won't be correct.

    ~ Stacie
  • So I'd recommend running RouteDEM on these to generate streams and see if they are continuous and look reasonable.

    I ran the RouteDEM on the above DEM and the result for flow accumulation was basically black (see attached image). The same goes for the flow direction. This makes no since to me, basically I have a watershed with a lake in the middle of it. My assumption is, that the was should eventually flow towards the lake.

    DEM area doesn't represent an entire watershed, but is cut out along non-natural boundaries (administrative or something.)

    My watershed is contained within three counties, so that's why the shape looks odd. But I read that my DEM needed to extend past the coverage areas of my watershed boundary.
    625 x 514 - 7K
  • swolnyswolny Member, NatCap Staff
    Did RouteDEM create streams? That's the big thing to look at - whether the stream network produced is continuous, not all chopped up (which is indicative of pits), and looks close to the real-world stream network. 

    ~ Stacie
  • Did RouteDEM create streams?

    No, it did not.

    I had a conversation with a person from USGS, and they postulated that the county level DEMs I mosaicked together may have different units (i.e. feet or meters) and could give really screwy results. I am going to try to use another DEM for my area of interest, and hope it gives better results. 

Sign In or Register to comment.