Page 1 of 1

[SOLVED] Add measurement uncertainty for kriging

PostPosted: Thu Oct 18, 2018 11:47 am
by AliciaGressent
Hi all,

I would like to consider measurement uncertainty for kriging. Have you done this before?
Is there any option in the kriging function to add this uncertainty?

Thanks a lot,
Best,

Alicia

Re: Add measurement uncertainty for kriging

PostPosted: Wed Oct 24, 2018 9:29 am
by Fabien Ors
Hi Alicia

There is an option in RGeostats which offers you the possibility to take into account the variance of measurement error (VME) while kriging.
You have to add a variable in your input database using the locator "v" which contains VME values for each sample.

Here is an example with two points (the first one with no measurement error and the second one with a big one).
Hope this helps !

Code: Select all
library(RGeostats)
# Variogram model (exponential)
model = model.create(2,range=20)
# Targets grid
grid = db.create(nx=c(100,100))
# Neighborhood (unique)
neigh = neigh.create(type=0)
# Input dataset
dat = db.create(x1=c(20,30),x2=c(50,50),z1=c(-1,1))
# Simple kriging without measurement error
res1 = kriging(dat,grid,model,neigh,uc="",mean=0)
plot(res1,title="Kriging without VEM",pos.legend=1)
plot(dat,add=T,pos.legend=7)
# Add measurement error on the second point only
datm = db.add(dat,error=c(0,100),loctype="v")
# Simple kriging with measurement error
res2 = kriging(datm,grid,model,neigh,uc="")
plot(res2,title="Kriging with VEM",pos.legend=1)
plot(datm,add=T,name.prop="error",pos.legend=7)


KrigWithoutVEM.png
KrigWithoutVEM.png (7.17 KiB) Viewed 7743 times

KrigwithVEM.png
KrigwithVEM.png (7.57 KiB) Viewed 7743 times

Re: [SOLVED] Add measurement uncertainty for kriging

PostPosted: Mon Sep 02, 2019 10:51 am
by AliciaGressent
Hi Fabien ,

I know I posted this VME question long time ago and it works pretty well but I was wondering how this VME is implemented in the kriging? Could you please tell me more about this?

Thank you very much,
Alicia

Re: [SOLVED] Add measurement uncertainty for kriging

PostPosted: Wed Sep 04, 2019 1:43 pm
by Fabien Ors
Dear Alicia,

The variance of measurement error of each sample data is added to the corresponding diagonal term of the covariance matrix before solving the kriging system.

Hope this helps,
Best regards