Estimate what's in the ground, and how sure you are.
Model grade, soil properties or contamination across a site from your drill-hole or core samples, with co-kriging for a correlated variable and logratio variograms for compositional data.

Drilling is expensive, decisions are large, and estimates must hold up to review.
How much of the variation in soil organic carbon across one field is spatially structured, and over what distance?
- Start from the samples840 soil samples of organic carbon (SOC, %) from one 26-hectare field in southern Ontario.How to: Importing data
- Fit the structureThe fitted variogram has a nugget of 0.52 and two Cubic structures, with ranges of 419 m and 825 m.How to: Variography
- Krige and cross-validateOrdinary kriging across the field; the build script's leave-one-out run gave an RMSE of 0.637 and a correlation of 0.783.How to: Kriging
Carbon is spatially correlated out to about 419 m, with a second structure to 825 m and a nugget of 0.52 for variation finer than the sampling or measurement error. The kriged surface and its variance map give the estimate and where it is weak.
Ranges are in Web Mercator units, about 1.38 times the ground distance at this latitude, so the 825 m range is about 596 m on the ground. Over one field the kriged values are unaffected.
