3

I would like to ask you guys how would i burn streams in DEM the following images will help understand what I would like to do.

enter image description here

this is my DEM and the blue lines are my rivers. However, when I delineated flow accumulation from it, the following images resulted.

enter image description here

How would I resolve this? I tried DEM reconditioning in HEC-GeoHMS. I really believe it is stream burning, but, I don't think that it affects the DEM elevation since there has no effect on the high and low values in its layer. The high and the low values are still the same.

  • With how much are you burning down your streams? A few meters? What are your high and low values in the DEM? – Martin Apr 20 '15 at 7:23
  • I would like to note; before burining streams to DEM, try to figure out to what extent do you count on the stream polyline reliability? or differently, it might be that you needed different fill parameters during the creation of flow accumulation. Last, the first image should be focused on a mismatch between flow accumulation and stream layer, while classifying altitudes to extent; This will probably expose th region for the divergent of your results from the stream layer. – dof1985 Apr 20 '15 at 8:32
  • 3
    Lastly, just a comment, but this is a really inappropriate and unintuitive colour palette to display elevation with. It goes from red-yellow-green-blue-red, low to high. I suspect it's intended for circular variables like flow direction since it begins and ends with the same colour. It makes interpreting the topography challenging. – WhiteboxDev Apr 20 '15 at 11:52
  • @WhiteboxDev - You really should post your comments as an answer as they are very useful and descriptive! – Joseph Apr 23 '15 at 11:02
  • thank for your comments and I also think @WhiteboxDev gives the good point. .Thank You! – vklopt Apr 24 '15 at 3:21
6

Okay, I've provided this answer to try and consolidate my comments above and to serve as a resource for others contending with the issue of stream burning. As I stated in my answer to this question Shapefile and DEM: check rivers behaviour, you would expect a mapped vector stream data set to deviate from a corresponding DEM-extracted stream network because you are comparing a model with a model, rather than a model with reality. This is particularly the case in the head waters of the channel network since that is where the mapped stream network is least accurate and the definition of what a stream is is somewhat ambiguous in these small first order streams. However, your stream networks differ just as much in the higher-order streams toward the basin outlets. This is a sign that your DEM is not of a sufficient quality to be able to accurately extract a stream network. Ultimately, to do what you want to do, you would need a more accurate and higher precision DEM. However, the main issue is this, with hydrological modelling the drainage patterns and basin delineation (represented by the flow accumulation raster derived from the DEM) is as important as the extracted stream network. Knowing whether a chicken-processing facility drains into a neighbouring field or directly into an adjacent stream makes all the difference and it's the DEM-derived flow network that tells you that. While you can burn your 'more accurate' mapped stream network into the DEM, doing so will not improve your flow accumulation (drainage pattern) map in the hillslope areas. It'll only improve the flow pattern along the mapped stream network and even then it'll likely result in parallel flow where a digital stream (in the DEM) and a mapped stream are not coincident.

Burning a stream network into a DEM is generally, in my opinion, a poor and much over-used practice. There are restricted cases where it makes sense. For example, it makes sense to burn streams through road embankments, effectively reinforcing flow along the buried culvert beneath the embankment. However people often have a false sense of high-accuracy when it comes to vector stream networks. The DEM-extracted stream networks derived from many modern DEM data sources, such as LiDAR, can often be more accurate than the blue-line network off of topographic maps, at least in lower-order basins. When the DEM is of low accuracy (as is the case with your DEM) a burnt stream network can give a false sense of accuracy when in reality it doesn't do anything to improve drainage patterns on hillslopes and can create strange artifacts along flood plains. These artifacts can sometimes be lessened when you apply an away-from-stream gradient as part of the stream burning, but most stream burning tools don't offer this option. A stream-burned DEM is unlikely to have more accurately defined basin divides than a non-stream burned DEM. Stream burning will also affect slopes measured from a DEM so it is strongly recommended that you use the original DEM for this purpose (although really that is the case for any hydrological pre-processing operation on DEMs). Often times, depression breaching algorithms can offer a better solution compared with stream burning, although depression filling, the more common DEM pre-processing method, will almost certainly offer a poorer solution.

In your case, your DEM has a very large flat area near the watershed bottom (lefthand side of the image). If that is the result of extensive DEM filling then my advice would be that if you are going to burn streams in a DEM be sure to fill depressions after the burn and not before. Of course, you'd likely be better off breaching those depressions anyhow. If the flat area is just the natural topography then I'd say that in such a low-relief area the whole basis of topographically driven flow path modelling breaks down and you really can't expect your drainage patterns to be anything near realistic in these areas. In such case, you'd need a much higher precision DEM than what you have. If you do need to burn streams, then you should use a decrement value that is sufficient to cut through artifact hills resulting from positive elevation errors. Therefore a decrement value of 3-10X the DEM error (RMSE) might be appropriate but this will depend on the data and topography. And remember, it would be a good idea to apply depression breaching or a hybrid breaching/filling algorithm afterwards.

  • If you do not know the RMSE of the DEM you are working on, are there any other good rules of thumb? Something akin to "3-10x the max change in elevation between neighboring cells" ? – traggatmot Sep 9 '15 at 2:33
  • @traggatmot Good question! This topic interested me so much that after writing this answer I wrote a paper on stream burning which is currently under review. Anyhow, if you email me I'll send you the draft and you'll hopefully find your answers there. – WhiteboxDev Sep 10 '15 at 16:18

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.