12 Feb 2009 05:05
Re: AI-GEOSTATS: theoretical sill
Figen- Your description is a bit confusing without pictures, but a few points might be worth saying more generally. First, if your data are covariance stationary, then it is true that the sill is the population variance. However, the sample variance is usually a biased (downward) estimate of the population variance. Thus, you should almost always find the sill to be larger than the sample variance. Despite the bias of the sample variance, it is asymptotically unbiased as the sample domain gets larger. So whether or not you should consider the sample variance comes down to an educated choice, "do you think your domain is large enough so that the bias of the sample variance is small?" Second, It sounds, however, like there is some sort of trend in your data, thus the data are not even covariance stationary, and there won't be a theoretical variance. Of course, you can always calculate the sample variance, but this is meaningless if the data are not covariance stationary. You could spend a lot of effort trying to get the zonal anisotropy modeled well, with the sill possibly different along the major and minor axes (by picking large ranges as you suggest), but if you are interested in the predictions, then the most critical part is not to model the sill, but to properly model the semivariogram at small distances. Nicholas Nicholas N. Nagle, Assistant Professor University of Colorado Department of Geography UCB 260, Guggenheim 110 Boulder, CO 80309-0260 phone: 303-492-4794 ---- Original message ---- >Date: Wed, 11 Feb 2009 12:31:51 -0600(Continue reading)
RSS Feed