[GRASS-dev] [GRASS GIS] #1429: r.watershed large grids

GRASS GIS GRASS GIS developer
1

#1429: r.watershed large grids
-------------------------+--------------------------------------------------
Reporter: dnewcomb | Owner: grass-dev@…
     Type: enhancement | Status: new
Priority: normal | Milestone: 7.0.0
Component: Default | Version: svn-trunk
Keywords: Large grids | Platform: Linux
      Cpu: x86-64 |
-------------------------+--------------------------------------------------
tried r.watershed on a large elevation grid (51000 rows, 133000 columns,
6783000000 cells) with the following options:

  r.watershed -m elevation=nc_20ft_ncfpm@statewide threshold=500000
accumulation=nc_flow_accum_wsh@statewide drainage=nc_direct_wsh@statewide
basin=nc_wshed_label_wsh stream=nc_stream_seg_wsh
half_basin=nc_half_basin_wsh memory=8000

and got the following output:

SECTION 1 beginning: Initiating Variables. 5 sections total.
SECTION 1a: Mark masked and NULL cells
segment_format(fd,1,-803458658,1,4096,8): illegal value(s)
seg_open(): illegal configuration parameter(s)
segment_format(fd,1,-803458657,1,262144,16): illegal value(s)
seg_open(): illegal configuration parameter(s)
SECTION 1b: Determining Offmap Flow.
Subprocess failed with exit code 11
category information for [nc_flow_accum_wsh@statewide] in [statewide]
missing or invalid
category information for [nc_direct_wsh@statewide] in [statewide] missing
or invalid
category information for [nc_wshed_label_wsh] in [statewide] missing or
invalid
category information for [nc_stream_seg_wsh] in [statewide] missing or
invalid
category information for [nc_half_basin_wsh] in [statewide] missing or
invalid
(Sat Aug 27 11:38:42 2011) Command finished (45054 sec)

GRASS 7.0 svn pull from 8/17/2011

looking at raster/r.watershed/seg/cseg.h

--
Ticket URL: <http://trac.osgeo.org/grass/ticket/1429&gt;
GRASS GIS <http://grass.osgeo.org>

2

#1429: r.watershed large grids
-------------------------+--------------------------------------------------
Reporter: dnewcomb | Owner: grass-dev@…
     Type: enhancement | Status: new
Priority: normal | Milestone: 7.0.0
Component: Default | Version: svn-trunk
Keywords: Large grids | Platform: Linux
      Cpu: x86-64 |
-------------------------+--------------------------------------------------

Comment(by mmetz):

Replying to [ticket:1429 dnewcomb]:
> tried r.watershed on a large elevation grid (51000 rows, 133000 columns,
6783000000 cells)

For larger grids, GRASS must be compiled with LFS: configure --enable-
largefile, and quite a bit of free disk space must be available where the
current location is living.

Apart from that, 6783000000 cells are currently too much, see recent
similar post in the user ml [0]. The current limit is 2^31^ - 1 =
2,147,483,647 cells. This can be overcome in trunk, but requires quite a
few modifications both in the segment library and in r.watershed.

Markus M

[0] http://lists.osgeo.org/pipermail/grass-user/2011-August/061724.html

--
Ticket URL: <http://trac.osgeo.org/grass/ticket/1429#comment:1&gt;
GRASS GIS <http://grass.osgeo.org>

3

#1429: r.watershed large grids
--------------------------------------+-------------------------------------
Reporter: dnewcomb | Owner: grass-dev@…
     Type: enhancement | Status: new
Priority: normal | Milestone: 7.0.0
Component: Raster | Version: svn-trunk
Keywords: r.watershed, Large grids | Platform: Linux
      Cpu: x86-64 |
--------------------------------------+-------------------------------------
Changes (by neteler):

  * keywords: Large grids => r.watershed, Large grids
  * component: Default => Raster

Comment:

I tried with the current GRASS 7.svn (r59279, approximately 7.0.0beta1)
and
r.watershed worked smoothly with 9 billion pixels (77 hours, way more
pixels
than in the original report):

{{{
GRASS 7.0.svn (nc_spm_08_grass7):~ > g.region rast=elevation -p res=0.15
projection: 99 (Lambert Conformal Conic)
zone: 0
datum: nad83
ellipsoid: a=6378137 es=0.006694380022900787
north: 228500
south: 215000
west: 630000
east: 645000
nsres: 0.15
ewres: 0.15
rows: 90000
cols: 100000
cells: 9000000000

GRASS 7.0.svn (nc_spm_08_grass7):~ > time -p r.watershed
elevation=elevation \
   threshold=500000 accumulation=nc_flow_accum_wsh
drainage=nc_direct_wsh \
   basin=nc_wshed_label_wsh stream=nc_stream_seg_wsh \
   half_basin=nc_half_basin_wsh memory=42000 -m
SECTION 1 beginning: Initiating Variables. 5 sections total.
SECTION 1a: Mark masked and NULL cells
  100%
SECTION 1b: Determining Offmap Flow.
  100%
SECTION 2: A* Search.
  100%
SECTION 3a: Accumulating Surface Flow with MFD.
  100%
SECTION 3b: Adjusting drainage directions.
SECTION 4: Watershed determination.
  100%
SECTION 5: Closing Maps.
Closing stream segments map
  100%
Closing basins map
  100%
Closing half basins map
  100%
Closing accumulation map
  100%
Closing flow direction map
real 277828.64
user 59672.08
sys 4481.16

--> 77.2 hours for

rows: 90000
cols: 100000
cells: 9000000000
}}}

Peak memory usage was 10GB (checked with munin).

Special kudos to Markus Metz for making this possible.

(I suppose the ticket can be closed at this point)

--
Ticket URL: <http://trac.osgeo.org/grass/ticket/1429#comment:2&gt;
GRASS GIS <http://grass.osgeo.org>