Parallel flow artifacts (“skidding”) in low-relief DEMs — alternatives to pit filling / experience with optimized pit removal?

[nvaughan22]

I’m running into an issue with drainage network extraction from a DEM where flow routing produces what I would describe as skidding — long, parallel flow paths across low-relief or nearly flat areas. I’ve experimented with different DEM resolutions and standard preprocessing (e.g., sink filling), but the issue persists.

I’ve heard prior success stories using the Optimized Pit Removal (CRWR, ~2013) tool, which uses a cut-and-fill approach to minimize topographic modification while enforcing drainage. I’m curious whether there are more modern or MATLAB-native approaches that achieve a similar goal?

My ultimate goal is fluvial inversion modeling, so preserving realistic channel geometry is important. Here are a couple examples of the drainage network issues I'm seeing:

[wschwanghart]

Interesting issue. The parallel flow lines in low/no relief areas persist when deriving the FLOWobj with the option preprocess = "carve". When using the option "preprocess = "fill", they might disappear or be less apparent. Here's an extreme example:

DEM = readexample("taalvolcano");
FD  = FLOWobj(DEM,'preprocess','carve');
A = flowacc(FD);
imagesc(log(A))

FD  = FLOWobj(DEM,'preprocess','fill');
A = flowacc(FD);
imagesc(log(A))

Looking at your example, it seems to me that the issue is more complex. The water surface has a more or less flat surface and some variations may yield an in-valley divide. Here, my approach would be to preprocess these water elevations by creating a mask (0 no water, 1 water), and then use skeletonization (see bwmorph) to identify the river's centerline. This centerline could then be used to "burn" into the DEM, i.e. lowering the elevation values at centerline. Once this is done, you could simply use FLOWobj to do the rest.

That's all I can say at that point without seeing the data. There are some more specialized tools to preprocess DEMs (e.g. Whitebox) but many approaches could be implemented using MATLAB's image processing tools.