## GEOSTATS: Summary: Kriging with Barriers

Expand Messages
• What a great group of folks, thank you all for the suggestions. In essense the solution is to do either 1 or 2 things. 1. Use an indicator kriging methodology
Message 1 of 1 , Apr 23, 1999
What a great group of folks, thank you all for the suggestions.

In essense the solution is to do either 1 or 2 things.

1. Use an indicator kriging methodology coding the surface
area for 1 = terrestrial sites, and 0 = aquatic sites. Krige the
trees across the entire surface using the indicator surface to
weight the solution. Only terrestrial sites should be returned

2. Krige the entire surface without an indicator surface, then use
a clipping coverage to remove the aquatic sites, leaving only the
terrestrial portion of the region.

Thanks again
Steve F.
..................................................................................................

postings from the kind folks who shed some light on my little corner
of the world.

from Ulrich Leopold

there is a possibility in SURFER 6.0 (low cost software for MS-DOS
and WINDOWS platform). You can perform interpolations in SURFER or
another geostatistical software. You can create a base map with all
your digital boundaries. This map can be used to blank your
interpolated result grid inside or outside those boundaries. You have
to define it.

If you use another geostatistical software (better decision), you
need an ascii column grid as output with the interpolated results.
X Y Z
1 2 3
1 3 3
: : :
. . .
If you just have Z but not the coordinates you need to add
them to the grid points and have to convert it to an SURFER
ascii-grid. Both can be done by 3PLOT (freeware; ai-geostat software
page). Then you can blank this SURFER grid with your base map in
SURFER and map it.

>from Tiina.
One thing you could try, is masking out of areas with lakes if you use
any GIS tool. I only have experience with Arc/Info.
Arc/Info kriging routines are not good. You could make the interpolated
surface first with some other software. Then import the datafile in
Arc/Info ( or other GIS-tool ) and create a cover of it, also creating a

cover for lakes. (Have you got access to a digitized map over lakes? Yoy

can import it in Arc/Info and make a cover.)
After that , it is simple map algebra;
Forestcover = Allcover - Lakecover.
Something like that should be possible to do with other tools as well.
Good luck!
Tiina.

> from eric j. lorup

is landcover of same class everywhere and do the trees spread through
some
mechanism that easily allows for modelling of autocorrelation? Otherwise

there would be other factors (such as soil type, areas not suitable for
tree
growing, ...) determining the distribution of the trees across your
landscape. So this will decide upon choice of Kriging method.

Re the lakes I don't have enough experience. All that came to my mind
was
the Kriging implementation in Arc/Info which, if I remember it right.
allows
for including of breakelines or borders or anything like this. But
Arc/Info
is by no means a real geostatistical tool :-)

There's other software out there, check the software list maintained by
Gregoire Dubois at:

http://curie.ei.jrc.it/software/index.htm

Kind regards
eric j. lorup

Steve Few
I discussed this problem with my wife (from Minnesota) this weekend. As
I
drive through Mn going to Milacs or Bay Lake in during the summer, I
wondered how in the world you would handle those patches of trees
between lakes.

I keep going back to the problem (ignoring scale) of breaking your State

into different pieces. The lack of continuity in the "h" vectors, the
distances (gap caused by lake milacs) are substantial. Even the small
lakes
such as Bay Lake nearby can wipe out the continuity necessary to
estimate a
reasonable semivariogram.

Breaking the Mn into pieces around big lakes, then make the big lakes
the
"center" of a region is one idea. Ignore the water, and do your
neighborhood thing as distance from center of the big lakes. On small
lakes, I would ignor the smallest ones, and do ??? Got me.

I guess you're paving the way. I haven't done any real difficult
spatial
modeling (just spatial stat class), so, sorry I can't help any further.

Steve Few
Statistician, NC DENR (Air Quality)

> Vera
I would suggest to try indikator kriging on the variable
presence-absence of lake and use it as a weighting function
for the kriged estimates of the tree density.

See Pawlowsky, Olea and Davis (1993), "Boundary assessment
under uncertainty", Mathematical Geology, 25(2), p. 125-143,
for details

> John Kern
I have dealt with this problem in many situations with other GIS folks.
I
am a statistician and often I will develop a spatial model and GIS
people
will display in ARC-Info or ARC/view etc. The problem you are
encountering
is really not a big issue as youcan go ahead and make predictions of the

numbers or density of trees where there are lakes followed by clipping
and
or intersecting the forest coverage with the lakes coverage where you
are
interested in not lakes. Making the prediction of tree density on one
pixel should have no influence on the predicted tree density on other
pixels, so you can either use a complicated kriging algorithm which
looks
at the 2000 lakes to decide whether or not to make a prediction, or
youcan
use a simple kriging algorithm to develop the tree coverage everywhere,
followed by using the power of the GIS to just eliminate unneeded
predictions in the lakes. I hope this helps.

> Gary Smith

I tend to brute force this sort of thing by having a separate indicator
data
file that assigns all grid nodes to categories (in your case each node
might
have a 1 for lakes and a 2 for vegetated landscape). Then I write a
short
piece of code that loads both the kriging results and the indicator
data
file and then have the computer assign the kriged value to the node if
the
indicator at a node is for vegetated landscape, and to return some other

sort of value if the indicator is for lakes such that when plotted, the
lakes will show up in a different color than the vegetated landscape
kriged
values. Because your kriging, and not using some sort of sequential
simulation, the fact that the estimator gives you a tree density number
where there can't be any trees does not effect the outcome where there
are
trees. So, simply ignorring the kriging results where you know there is

water shouldn't effect the validity of your assumptions.

You may get a wiser answer from someone more into geostats than I, but
this
might be worth considering

> Stephen Smith
I am surprised that you didn't get any responses (even statements that
there
is no one way of doing it). People here have run up against these
problems
when kriging snow crab, surfclams, etc around shoals, islands etc. The
only
method that I have used is to define a polygon which defines the surface
to
be kriged and then kriged over the points in that polygon only (using
poly.grid). The polygon excludes the areas where the animal can not
exist.
This ignores potential interface effects (e.g., beach, surf line).

--
*To post a message to the list, send it to ai-geostats@....
*As a general service to list users, please remember to post a summary
of any useful responses to your questions.
*To unsubscribe, send email to majordomo@... with no subject and
"unsubscribe ai-geostats" in the message body.
DO NOT SEND Subscribe/Unsubscribe requests to the list!
Your message has been successfully submitted and would be delivered to recipients shortly.