[SoC] GSOC Horizon based voxel interpolation wk 7 checkin

[Tim Bailey]

Hi Folks
I worked through the weekend, I am going to have limited internet access at
the end of the week, and I am working under a rather clear mentor
ultimatum, So here is my week 7 report.

This week I implemented the voronoi operator, that had confounded me for
much of July, through a two stage iterative process.  Process A propagates
identities vertically up profiles from horizon boundaries. Process B
propagates identities horizontally away from the profiles.
I applied this to one of my experimental sandboxes that was built for a
study of late holocene co-seismic subsidence in a coastal saltmarsh. In
this circumstance coastal marshes exhibit extreme sensitivity to their
elevation above sea level.  A series of earthquakes during the past 4500
years are recorded by the sedimentary record at the coastal margins. These
earthquakes resulted in subsidence events of almost a meter and a half in
one episode, and notable movement in several other events. In addition the
introduction of tsunami, and mudflat deposits overlaying peat deposits
create a very appealing paleo environment to reconstruct using voxel
operators.The GRASS region was bound between the deepest core at 6 meters
below sea level and 3 meters above, which was clearly out of the range of
the modern salt marsh. On my computer with an amd5700 processor, 12gb ram,
with ssd, Ubuntu 13.04 and GRASS7,process A takes 12 seconds per
iteration,  and process B takes 22 seconds per iteration for a 32,000,000
voxel map. From the data that I am working with process A required 9
iterations and process B required 50 cycles. Region anisotropy was set to
10:1, where it could really comfortably be 100:1.  Nonetheless, this is not
that different from the computing demands of moderately large lidar job.
Also in this case iteration limits can be used to limit the area of
influence of a data point.

Next week will work on refining this weeks work and implement a surface
limit.

Tim Bailey
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/soc/attachments/20130731/49ee6ff5/attachment.html>