[R-sig-Geo] OLS vs likfit sill values
aelmore at usgs.gov
Tue Jul 5 15:40:48 CEST 2011
I'm trying to automate anisotropic semivariogram modeling and have done the
1) estimated isotropic parameters of NDVI data with automap
2) fed these parameters to geoR for an OLS model fit of the geoR isotropic
3) estimated anisotropy parameters for the data with intamap
4) fed the geoR OLS model parameters and intamap anisotropy parameters to
likfit to fit an anisotropic model (none of the parameters was fixed).
I'm now trying to figure out why the anisotropic model that is produced has
a sill value that's a little over one and a half times as big as the OLS
model's sill value. Additionally, both the major and minor ranges (which
are the values I'm most interested in for in my current work), are lower
than the omindirectional inputs provided, although not as remarkably
different as the sill.
The plot I've inserted below doesn't show the OLS model, but that model
basically runs right through the binned means of the omnidirectional sample
variogram (represented by filled circles in the graph), and has nugget,
partial sill, and range values of 8118, 10954, and 96, respectively. The
likfit parameters are given in the plot's text box.
Does anyone have any ideas about why I would be getting such drastically
different models from the two different fitting routines? I know that
likfit doesn't fit the experimental bins, returning instead to the cloud
values. Is likfit more sensitive to outliers or normality problems than
OLS? If you have suggested reading materials, I'm all ears (or eyes, as the
case may be).
View this message in context: http://r-sig-geo.2731867.n2.nabble.com/OLS-vs-likfit-sill-values-tp6549794p6549794.html
Sent from the R-sig-geo mailing list archive at Nabble.com.
More information about the R-sig-Geo