[GRASS-dev] [GRASS GIS] #2719: r.watershed + D8 (unexpected results)

[GRASS GIS]

#2719: r.watershed + D8 (unexpected results)
---------------------+-------------------------
  Reporter:  180875  |      Owner:  grass-dev@…
      Type:  defect  |     Status:  new
  Priority:  normal  |  Milestone:  7.0.3
 Component:  Raster  |    Version:  svn-trunk
Resolution:          |   Keywords:  r.watershed
       CPU:  x86-64  |   Platform:  Linux
---------------------+-------------------------

Comment (by mmetz):

 Replying to [ticket:2719 180875]:
 > Dear GRASS Gurus,
 >
 > During the last days I've discovered a strange behaviour of r.watershed
 and D8. The used DEM is the result from a landscape evolution model and
 there are no depressions. I cannot use MFD as the D8-FlowDirections input
 data of a variety of tools I’ve written during the last years.
 >
 > I got obviously wrong values for the flow accumulation of r.watershed
 and D8. The number of cells in the Flow Accumulation grid at the outlet
 point was significantly smaller than the area of the analyzed catchment
 (spatial resolution in north south and east west direction is 1) also
 computed with GRASS using r.water.outlet and a conversion to Vector
 (Area).
 > I checked Flow Accumulation at the the outlet with an alternative code
 of a colleague and the contributing drainage area was exactly the value of
 the catchment size computed with r.water.outlet.

 This could only happen if the catchment is not fully covered by the DEM.
 In this case, edge cells are not contributing to the catchment because for
 edge cells it is not possible to say where the drain to.
 >
 > I further found some deviations in the flow direction grid. In some
 cases the flow routing leads to the lowest neighbouring cell which however
 might not be the steepest decent. r.watershed chooses sometimes a diagonal
 flow direction while the local channel gradient would be greater in
 horizontal direction.
 > I started some test runs with very simple geometries. I've defined a
 ramp with r.mapcalc that dips in north direction and I expected that this
 will also be shown by the flow direction of r.watershed. However, this is
 not the case. The synthetic DEM features a watershed!!! See attachment.
 It seems that r.watershed fills the region (without need). This strange
 behavior disappears by applying the "multiple flow direction" algorithm.
 >
 > I have tested with the latest versions grass70 , grass71 and grass645.
 > Here is a test simple test case (grass71) that documents the misbehavior
 of r.watershed and D8.
 >
 > {{{
 > g.region w=0 e=1000 s=0 n=1000 res=1 -p
 > r.mapcalc "DEM = 1000 - y()"
 > r.watershed -s elevation=DEM at PERMANENT accumulation=ACC drainage=DIR
 > }}}

 The correction for diagonal flow bias missed two special cases. Fixed in
 trunk r67189, please test.

--
Ticket URL: <https://trac.osgeo.org/grass/ticket/2719#comment:3>
GRASS GIS <https://grass.osgeo.org>