Yes, these inaccuracies come from using a (global) DEM for catchment delineation. It also behaves strange in river deltas, when multiple branches of the river show up.
I'd like it to render a whole endorheic basin with one selection of a lake, but currently it splits the lake into two or more, e.g.
very interesting... on SRTM the example I checked (Pyramid Lake) is all at one altitude (albeit, only with 1 m resolution), while on MERIT-DEM, the lake surface differs by more than 1 m across the lake.
It's too late to edit, but I should clarify, MERIT has at least two DEM, the MERIT-DEM and MERIT Hydro.
MERIT Hydro is the one I am using at larger scales.
MERIT Hydro has "Hydraulically Adjusted Elevations". Of course the actual elevation of the surface of the lake will be set up in whatever direction the wind is blowing on a given day, but MERIT Hydro is interesting because they tweaked the elevations to make them monotonically decreasing in a way that should align with the mean direction of flow.
For visualization, MERIT Hydro is nice because it is smooth float32 (as in your image) that also downsample nicely. It's just not high enough resolution for smaller catchments. AW3D30 has 30m resolution which is much better, but is stored in int16 and can sometimes look a bit lumpy, especially with a large vertical exaggeration. For example:
One might do well to use a Voroni diagram algo whereby a middle-out where the center is the center of the obloide(globe) and you calc the volume of water as compared to the topology.
I'd like it to render a whole endorheic basin with one selection of a lake, but currently it splits the lake into two or more, e.g.
https://topography.jessekv.com/?lat=31.4299&lng=35.4724&mode...
and
https://topography.jessekv.com/?lat=31.4136&lng=35.4702&mode...