If you are using a commercial baseline processing software it should produce a covariance matrix that you could use for your estimation parameters.

Dear Dr.Londe

my goal is calculating this covariance matrix by MATLAB software.I must evaluate these semivariogram parameters in order to use in ordinary kriging interpolation.

Thank you very much for your answer.

If you have multiple data sets you might compute an empirical variogram for each one ('semivariogram"  is outdated, incorrect terminology since 1988). The parameters of the variiogram (theoretical model, empirical variograms do not have any parameters) are based on certain stationarity assumptions, are those different for the different data sets (strictly speaking stationarity is not a property or characteristic of data). Bottom line, why do you want to compute empirical variograms for the separate data sets and what information do you expect to obtain from the separate empirical variograms, how would you use the information for the separate empirical variograms. Note that the only way you could compute a variogram would be if you knew the multivariate probability distribution which is extremely unlkely.

Be careful about the term "covariance matrix", it is probably not relevant to either empirical or theoretical variograms.

Dear Dr.Myers

Actually, I am doing research for my dissertation now which its subject is Network Real Time Kinematic Processing (NRTK).

As you know in empirical formulas for example in exponential model we must evaluate sill,range and nugget parameters. I want to create kriging interpolation matrices by variogram in order to interpolate ionospheric and tropospheric errors (My desired values for interpolation) for special baseline in the network. Note that the desired network consists five permanent GPS stations. Actually, In NRTK algorithm we need to select one GPS station as a master reference station. In this algorithm at first we need to evaluate correlations between master station and other reference stations. After doing this job, we want to interpolate our desired values for special baseline. It should be mentioned that the special baseline is the baseline between master and rover stations.

So, how can I  use variogram and specially empirical formulas in order to interpolate ionospheric and tropospheric errors for special baseline?

Thanks for your attention. I need your help.

http://www.int-arch-photogramm-remote-sens-spatial-inf-sci.net/XL-1-W5/441/2015/isprsarchives-XL-1-W5-441-2015.pdf

This paper  may answer your question.

As I understand your problem, you are wanting to interpolate corrections and your data  consists of the corrections at the baseline stations. Is that correct?

Dear Dr.myers

I want to ask if you have any ideas about evaluating empirical models, such as exponential model. 

How can I calculate sill,range and nugeet parameters in order to substitute in exponential model?

The exponential variogram does not have a true range, instead you look for the distance (on the plot of the empirical variogram) at which the variogram value is 95% of the sill.  The sill is the height of the horizontal line that the plot of the empirical variogram  is asymptotic to. You can't really "calculate" the parameters although you might try a least squares fit against the empirical variogram cloud.  Unfortunately this is non-linear least squares so getting an analytical solution. More over it should probably be a weighted least squares and there are several possible weighings to use (no one choice is best possible for all applications) Most geostatistical packages will use this approximate range correctly but be aware that if you write your own code it is more complicated.

Again you can not "calculate" the nugget effect (except possibly by least squares), it is the magnitude of the jump discontinuity at distance zero and is determined by the shape of the empirical variogram for short distances. The difficulty is that for any data set there is always shortest distance between two data locations and hence no plotted points for shorter distances. Moreover in most data sets, the number of data pairs for short distances is small, often too small to get a good plot and hence a good picture of the shape of the variogram for short distances.

Once you have a preliminary estimate for these parameters try cross validation to see how sensitive the cross validation statistics are to changing one or more of the parameters. You can't find the "best possible" but you can improve your estimates