HI, I wonder does anyone have experience with doing sequential gaussian simulation with krige() function in gstat?
I find it VERY slow compared to use krige() to achieve kriging function itself.. I wonder why, is that because it has to model the variogram, and do the kriging separately for each point to be simulated?
It does not model variograms on the way. It is slower than kriging because it uses the sequential simulation algorithm: for each node visited, a value is simulated, and this value is added to the conditioning data. For this reason you _have to_ limit the search neighbourhood (using the nmax or maxdist arguments) if you're simulating on more than say a few hundred nodes. Taking a very small nmax may yield fast simulations, at the expense of a good representation of the target spatial correlation structure.
so it would be N times slower to achieve the simulation than the kriging if the number of points to be estimated is N??
No. Experiment with the nmax argument. -- Edzer
______________________________________________ [EMAIL PROTECTED] mailing list https://www.stat.math.ethz.ch/mailman/listinfo/r-help
