[R-sig-Geo] variogram problem

rubenfcasal rubenfcasal at gmail.com
Thu Jun 25 10:05:04 CEST 2015


Hi Adeela,

> This is the the data set and want to make variogram for Cd15.
Start with an exploratory analysis of the data. For example, using the 
geoR package:
lyallpur <- read.csv('lyallpur.csv')

library(geoR)
datgeo <- as.geodata(lyallpur, coords.col = 2:3, data.col = lyallpur$Cd15)
# as.geodata: 2 replicated data locations found.
#WARNING: there are data at coincident or very closed locations, some of 
the geoR's functions may not work.
# Use function dup.coords() to locate duplicated coordinates.
# Consider using jitterDupCoords() for jittering replicated locations

summary(datgeo)
#$duplicated.coords
#   dup    long     lat SrNo  Cd15  Cd30
#15   1 72.9229 31.5025   15 0.087 0.015
#16   2 72.9229 31.5025   16 0.021 0.012
#11   3 72.9810 31.5052   11 0.034 0.008
#12   4 72.9810 31.5052   12 0.020 0.011

plot(datgeo)

There are replicated measurements (not 'standard' in geostatistics...). 
Consider combining them  (easy way...) or introducing a measurement error...

> The distance between each sample location was 4 cross 4 km.

This does not seem to be true...

There are also two locations farapart from the others,I would recommend 
to leave them out in the initial analysis.

lyallpur <- lyallpur[lyallpur$lat > 31.35,]
datgeo <- as.geodata(lyallpur, coords.col = 2:3, data.col = 
log(lyallpur$Cd15))
plot(datgeo)

The log transform seems appropriate, the distribution of the transformed 
values is almost symmetric.

> Please suggest me how to allocate initial guess about nugget, sill and 
> range parametrs.

It seems that you have to fit a spatial trend first...

If the results obtained with the traditional procedures are 
unsatisfactory (e.g misspecification of the parametric model),you could 
try the npsp package (e.g. Fernandez-Casal R. and Francisco-Fernandez 
M., 2014, Nonparametric bias-corrected variogram estimation under 
non-constant trend, Stoch. Environ. Res. Ris. Assess, 28, 1247-1259).

 >WARNING: self-advertising detected...

Best regards, Ruben Fernandez.



More information about the R-sig-Geo mailing list