how to choose the spatio-temporal variogram parameters for krige interpolation

Question

Recently I'm doing the spatial-temporal interpolation with R. I choose the function krigeST in package gstat,which used kriging method in spatial-temporal data. Here is my steps:

1. Firstly I create a STFDF object called stfdf:

   stfdf<-stConstruct(observation,space = list(values = 1:ncol(observation)),time = Node$Time,SpatialObj = loc,interval = T)
   

2. Then I compute the empirical variogram with the spatial-temporal data stfdf:

    var<-variogramST(values~1,data = stfdf,assumeRegular = T)
    plot(var,wireframe=T)
   

here is the image of the empirical variogram: empirical variogram

3. Finally I need to fit the empirical model with the theoretic model,here I choose the seperable model:

   sepvgm<-vgmST("separable",space = vgm(0.02, "Exp", 8000, 0),time = vgm(0, "Exp", 2, 0.001),sill = 0.02)
   sepvgmST<-fit.stVariogram(var,sepvgm,fit.method=6,method="L-BFGS-B")
   

Now,my question is how should I choose the parameters in the vgm function ? In the example above I set it to space=vgm(0.02,"Exp",8000,0) and time=vgm(0,"Exp",2,0.001) ,but it fails to fit the empirical variogram.

Does anyone have idea on this? Thanks for a lot!