Dear All
Apologies if the following betrays any naive assumptions but I would
be grateful for the advice nonetheless.
I am looking to do some block kriging on a pottery counts from an
archaeological survey covering the whole surface of an island (ca.
20sq.km, for details see www.ucl.ac.uk/asp). The blocks are far from
ideal in being <100m long transects (without consistent orientation
and ca. 2m wide), walked by archaeological surveyors spaced 15m
apart. At first glance, variogram-fitting and kriging prediction
based on such blocks does seem possible using gstat, but before
looking more closely, my initial concern is more theoretical (a type
of question I hope is still of interest to this list?).
The pottery densities are highly skewed towards low values and a few
high ones, to the extent that a natural log transform (after adding 1
to adjust zero values) still leaves a heavily skewed distribution.
There is probably a sampling effect in here at the low end as true
zero values are over-represented (i.e. the observed zero values
disguise a range of low potsherd intensities that would be
identifiable if surveyors were to look over a longer observation
period). In contrast, a log transform of all non-zero values does a
reasonable job of deskewing the data, but I feel it would be
deceptive to ignore the zero values in this manner.
Can anyone suggest an appropriate way forward, and perhaps also offer
any previous experience with irregular block kriging with gstat (I
have looked through the archives and found only a couple of
inconclusive threads)?
with thanks in advance
Andy
Andrew Bevan
Lecturer
UCL Institute of Archaeology
31-34 Gordon Square
London WC1H 0PY
tel: +44 (0)20 7679 1528 (internal 21528)
info: www.ucl.ac.uk/archaeology/staff/profiles/bevan
[[alternative HTML version deleted]]
_______________________________________________
R-sig-Geo mailing list
R-sig-Geo@stat.math.ethz.ch
https://stat.ethz.ch/mailman/listinfo/r-sig-geo
