[R-sig-Geo] automap model defaults

[Paul Hiemstra]

Edzer Pebesma wrote:
> Mark, please note that the linear model (with nugget), having only two
> parameters where most models have three (nugget, range, sill) or four (+
> smoothness) arises as a special case of e.g. the spherical or
> exponential model when the range becomes infinite, or at least very
> large compared to the spatial extent. I'm not sure if automap allows
> fitting of very large range values.
>   
See the following example:

data(meuse)
coordinates(meuse) = ~x+y
vgm1 <- variogram(log(zinc)~1, meuse)
vgm1$gamma = 1:15 + 0.00001*runif(15)
v1 = fit.variogram(vgm1, vgm(15,"Sph",1000,1))
v2 = fit.variogram(vgm1, vgm(15,"Lin",1000,1))
plot(vgm1, v1) #Sph
plot(vgm1, v2) #Lin

Probably the difference in the estimated covariance matrix is quite 
small between the linear and the spherical variogram. In this example 
the range is quite a bit larger than the spatial extent, so I think 
automap can handle this.

cheers,
Paul
> --
> Edzer
>
> Mark Connolly wrote:
>   
>> On 04/15/2010 07:42 AM, Paul Hiemstra wrote:
>>     
>>> Mark Connolly wrote:
>>>       
>>>> I was wondering what is behind the selection of the default models
>>>> tried by autofitVariogram (model=c("Sph","Exp","Gau","Ste")).  For
>>>> example, why was linear model left out?
>>>>
>>>> Thanks
>>>>
>>>> _______________________________________________
>>>> R-sig-Geo mailing list
>>>> R-sig-Geo at stat.math.ethz.ch
>>>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>>>         
>>> Hi Mark,
>>>
>>> These models are the models that are most often used, so it seemed
>>> logical to use these ones. But why do you want to include the linear
>>> model specifically?
>>>
>>> cheers,
>>> Paul
>>>
>>>       
>> Only because it sometimes results in a lower SSError.  I guess that is
>> my real question: why wouldn't I want to try the linear model in the
>> set?  The field I am looking at is a layered ag field in a coastal
>> plain, and the five depth layers (which extend to less than a meter from
>> the surface) I am looking at are often fairly homogeneous for specific
>> properties.  Some of property values I am fitting produce a linear model
>> as the best fit in a specific layer.  (Although sometimes when this is
>> the case, the linear model is just barely better and than the next-best
>> non-linear model.)
>>
>> I assume the selection of the default models is not arbitrary, but I
>> lack the experience to include/exclude linear models in the best fit
>> search.  Best to stick with the default?
>>
>> Thanks,
>> Mark
>>
>> _______________________________________________
>> R-sig-Geo mailing list
>> R-sig-Geo at stat.math.ethz.ch
>> https://stat.ethz.ch/mailman/listinfo/r-sig-geo
>>     
>
>   


-- 
Drs. Paul Hiemstra
Department of Physical Geography
Faculty of Geosciences
University of Utrecht
Heidelberglaan 2
P.O. Box 80.115
3508 TC Utrecht
Phone:  +3130 274 3113 Mon-Tue
Phone:  +3130 253 5773 Wed-Fri
http://intamap.geo.uu.nl/~paul
http://nl.linkedin.com/pub/paul-hiemstra/20/30b/770