Breach_depression_least_cost creates high plateaus.

[InsolublePancake]

I'm using Breach_depression_least_cost to fix a DTM. I am accessing it via the latest version of the python API (not workflows).

I apologies in advance here because I've been unable to produce a small dataset that replicates the problem. For this reason, I suspect that the issue may lie with the dataset itself, but if so I haven't been able to work out exactly how.

There are two issues. Firstly the breach depressions gives different results every time. Wildly different. Is this expected?
The other, more pressing issue, is that it creates high plateaus over topographic highs. I've included an image of several versions of the breach depressions output as seen by the profile tool in QGIS. The black line is the input DEM, all other profiles are output from breach depressions.

The output of some of them look as expected in places - e.g. filling lakes and other depressions. But why would it create these very high plateaus over the topo highs? In this example adding almost 150m of elevation.

This profile is a very small section of a very large DTM - the section was chosen entirely at random, the issue occurs everywhere.

I appreciate that there isn't a huge amount that you can do without data to work on. But perhaps can you tell me whether the variability between runs is expected? And possibly suggest some kind of data error that could prompt this behaviour.

Many thanks,
Rick

[jblindsay]

Thanks for reporting this, and for the detailed description.
What you are seeing is not expected as normal behavior, especially the large run-to-run differences.

From your description, there are two likely contributors:

1. Non-deterministic pit processing in some terrain configurations

• If many pits have similar/equal elevations, the processing order can influence downstream edits.
• Because each solved pit modifies the surface, different tie/order paths can produce different final outputs.

2. Fill fallback on unresolved pits

• If breaching cannot find an acceptable outlet (for example due to restrictive distance/cost limits), unresolved depressions may be handled by fill behavior.
• That can create broad raised flats/plateaus, including in places that look like topographic highs in profile.

Could you please run the following tests on the same DEM extent and share outcomes/screenshots?

1. Determinism check

• Run the exact same command 2-3 times with all parameters unchanged.
• Report whether outputs are bit-identical or visually different.

2. Disable filling fallback

• Re-run with fill disabled.
• If plateaus disappear, that strongly indicates unresolved-pit fill behavior is driving the highs.

3. Relax breach constraints

• Increase breach search distance and relax/remove max_cost, then compare.
• If plateauing drops substantially, unresolved pits were likely due to restrictive settings.

4. Parameter sensitivity sweep

• Keep everything fixed except one parameter at a time:
  • dist: low, medium, high
  • max_cost: strict, moderate, unlimited
• This helps isolate threshold behavior.

5. Data sanity checks

• Confirm DEM has no nodata seams/voids, extreme spikes, or mixed vertical units in the affected area.
• If possible, share min/max and histogram stats for the problematic tile.

If you can share:

• exact tool call/flags,
• WhiteboxTools version,
• and a small clipped raster reproducing the issue,
  that would make it much easier to diagnose precisely.

[InsolublePancake]

Thank you that's very helpful. I have already run the first test you suggested. I have been running each command three times with the same input/parameters. I will run the other experiments you've suggest and report back.