passing 'cutoff' to the 'autoKrige' function ('automap' package)

Dear list,

Is there any way to pass the 'cutoff' argument (or 'width', or other
arguments of the 'variogram' function of the 'gstat' package)  to the
'autoKrige' function of the 'automap' package ?

I tried with:

   

x.ok<- autoKrige(zinc~1, meuse, meuse.grid, miscFitOptions=list(maxdist=100,  cutoff=500, width=50) )
     

but both arguments are ignored.


I also tried with:

   

x.ok<- autoKrige(zinc~1, meuse, meuse.grid, maxdist=100,  cutoff=500, width=100)
     

but I got the following error message:

"Error in gstat(formula = formula, data = locations, model = model,
beta = beta,  :
  unused argument(s) (cutoff = 500, width = 50)"


Some session info:

R version 2.12.2 (2011-02-25)
automap_1.0-9
gstat_0.9-79


Here there is a reproducible example (for the 'autofitVariogram'
function, because I didn't find a way to pass the 'cutoff' and 'with'
arguments to the 'autoKrige' function):

------------ START ---------------

library(automap)

# Data preparation
data(meuse)
coordinates(meuse) =~ x+y
data(meuse.grid)
gridded(meuse.grid) =~ x+y

# Computing the 'boundaries' value for the 'variogram' function,
# in order to properly compare its results with those of the
'autofitVariogram' function

input_data<- meuse

x<- coordinates(input_data)[, 1]
y<- coordinates(input_data)[, 2]
scale_number<- (0.35 * sqrt((max(x) - min(x))^2 + (max(y) - min(y))^2)/100)
boundaries<- c(2, 4, 6, 9, 12, 15, 25, 35, 50, 65, 80, 100) * scale_number

# variogram computation. Case 1, default parameters
vgm.gstat<- variogram(zinc ~ 1, locations=meuse, boundaries=boundaries)
vgm.automap<- autofitVariogram(zinc ~ 1, input_data=meuse)

# These variograms are the same
print(vgm.gstat)
print(vgm.automap$exp_var)

# variogram computation. Case 2, with 'cutoff' and 'width'
vgm.gstat2<- variogram(zinc ~ 1, locations=meuse, cutoff=500, width=50)
vgm.automap2<- autofitVariogram(zinc ~ 1, input_data=meuse,
cutoff=500, width=50)

# These variograms are NOT the same
print(vgm.gstat2)
print(vgm.automap2$exp_var)

# variogram computation. Case 3, with 'cutoff', 'width' and 'boundaries'
vgm.gstat3<- variogram(zinc ~ 1, locations=meuse, cutoff=500,
width=50, boundaries=boundaries)
vgm.automap3<- autofitVariogram(zinc ~ 1, input_data=meuse,
cutoff=500, width=50)

# These variograms are the same
print(vgm.gstat3)
print(vgm.automap3$exp_var)


------------ END ---------------


Looking at the examples, I think that the problem arises because the
'boundaries' argument superimpose its values to the 'cutoff' one, both
in the 'variogram' and 'autofitVariogram' functions.

However, whereas in the 'variogram' function the boundaries are
optional, in the 'autofitVariogram' they are computed internally (in a
smart way), but its maximum default value disable any shorter 'cutoff'
passed to the 'autofitVariogram' function.


In any case, I think that even after solving the issue related to
'cutoff' and 'boundaries', it will still be impossible to pass the
'cutoff' argument to the 'autoKrige' function, because the '...'
argument in 'autoKrige' is passed (internally) to the 'krige' function
and not to 'autofitVariogram' one. Am I right ?

I'm not sure if this could be considered a bug, but I thought it could
be useful to report this behaviour, for the next release of 'automap'.


Kind regards,

Mauricio Zambrano B.


   

Hi,

The miscFitOptions parameter does not include the 'variogram' paramters. I
will look into adding them, together with the option to set the boundaries
manually. However, autofitVariogram at this stage defines a fixed set of
boundaries, so I'm not sure if the parameters you mention will resort any
effect even when passing them on. An option at this stage would be to use
the normal krige function combined with autofitVariogram. autofitVariogram
passes on parameters to variogram() using the ... .

Could you create a new issue in the bitbucket.org account which stores
automap [1]? This way I can keep track of the issues that need to be fixed
for autofitVariogram.

2011/3/26 Paul Hiemstra<p.hiemstra at geo.uu.nl>:
   

Hi,

The miscFitOptions parameter does not include the 'variogram' paramters. I
will look into adding them, together with the option to set the boundaries
manually. However, autofitVariogram at this stage defines a fixed set of
boundaries, so I'm not sure if the parameters you mention will resort any
effect even when passing them on. An option at this stage would be to use
the normal krige function combined with autofitVariogram. autofitVariogram
passes on parameters to variogram() using the ... .

Could you create a new issue in the bitbucket.org account which stores
automap [1]? This way I can keep track of the issues that need to be fixed
for autofitVariogram.
     

Thanks Paul for your answer.

I created the issue in bitbucket.org, as an 'enhancement', not a bug.

If you manage to include 'cutoff' and other  'variogram' paramters
into the 'autofitVariogram' function, a way to use its value could be
(after the definition of the 'scale_number'):


   

     scale_number = (0.35 * sqrt((max(x) - min(x))^2 + (max(y) -
         min(y))^2)/100)
     if (!is.na(cutoff))
       scale_number<- min(scale_number, cutoff/100)
     boundaries = c(2, 4, 6, 9, 12, 15, 25, 35, 50, 65, 80, 100) * scale_number


Do you think it may work ?
   

In the mean time the combination of 'autofitVariogram' with 'krige'
that you mention is a nice way to overcome the issue with cutoff.

Thanks again for such a useful package.

Cheers,

Mauricio

   

On 03/28/2011 10:47 AM, Mauricio Zambrano wrote:

2011/3/26 Paul Hiemstra<p.hiemstra at geo.uu.nl>:

Hi,

The miscFitOptions parameter does not include the 'variogram' paramters.
I
will look into adding them, together with the option to set the
boundaries
manually. However, autofitVariogram at this stage defines a fixed set of
boundaries, so I'm not sure if the parameters you mention will resort any
effect even when passing them on. An option at this stage would be to use
the normal krige function combined with autofitVariogram.
autofitVariogram
passes on parameters to variogram() using the ... .

Could you create a new issue in the bitbucket.org account which stores
automap [1]? This way I can keep track of the issues that need to be
fixed
for autofitVariogram.

Thanks Paul for your answer.

I created the issue in bitbucket.org, as an 'enhancement', not a bug.

If you manage to include 'cutoff' and other ?'variogram' paramters
into the 'autofitVariogram' function, a way to use its value could be
(after the definition of the 'scale_number'):

Thanks for creating the issue, I'm not sure when I'll come round to fixing
the issue though ;). I'll probably fix the problem by giving the user the
freedom to change all inputs to variogram(), including boundaries. But I'll
figure that out later.

? ? scale_number = (0.35 * sqrt((max(x) - min(x))^2 + (max(y) -
? ? ? ? min(y))^2)/100)
? ? if (!is.na(cutoff))
? ? ? scale_number<- min(scale_number, cutoff/100)
? ? boundaries = c(2, 4, 6, 9, 12, 15, 25, 35, 50, 65, 80, 100) *
scale_number


Do you think it may work ?

Could be an option.

cheers,
Paul