Prasad,
I have interpolated the log values
for the SEA population densities(it is standard
for concentration data in geostatistics) it is being done
for several reasons -
-reduction of gradients when the data change over
several magnitudes
-preventing the function going to negative values
(when you unlog negative values of log data you get
data close to zero, but never negative)
I would suggest to add zeroes into the ocean
probably along the shoreline if you have tools to do that.
(e.g. I needed to do bathymetry from insufficient data
so i have taken the shoreline, put the zeroes as
attributes and interpolated the data from the bottom
of bay + the points defining the shore and the results
looked very good I have used v.prune to reduce the
number of points on shoreline)
this will prevent the high values from one island influence
the values on another island as the function will always
go to zero as it enters the ocean. And of course, mask
then the ocean out to get exactly zeroes there.
And ALWAYS use smoothing, I have done some analysis
of crossvalidation errors, small smoothing (0.05)
will keep your surface very close to your data and it
improves the predictive capability of s.surf.tps
significantly. I wiler to
GRASSclippings and I hope we will be able to add
an automatic choice for the optimum choice of
tension and smoothing for the next release.
If you have too much trouble with that, you can send me
the data after September 1.
I hope this helps,
Helena