I am in the midway of understanding how gstat package in R tool implements kriging method. I have understood the calculation of empirical semivariogram and fitting semivariogram models. But I did not understand how it implements the matrix inversion to calculate the weights of the kriging estimators. I have a large data set containing 50000 lat-long-precipitaion triplets. Theoretically inversion of a matrix of size 50000x50000 must be done in order to get the weights. While this large matrix takes several GBs of man memory, which is particularly impractical.
My question is that how krige function does all this within a second?
Regards,
Chandan