Precision Agriculture: Sampling to Estimate Soil Nutrients

Keywords: Precision farming, geostatistics, spatial variation, kriging, variogram, bulking, soil sampling, mapping

Properties of the soil vary at many different scales of spatial resolution in the landscape and even within a single field. Traditionally in agriculture it is the variation in the soil between fields that is managed, but the level of resolution that the farmer is interested in now for site specific or precision farming is that within fields. To describe the variation of soil and other properties within fields accurately we must have reliable information. The latter usually comes from sampling because we cannot examine every possible location. The spatial variation within fields comprises a very local component of less than a few metres, and one or more longer scale components of tens of metres. We can eliminate the former to ensure that the sample represents the surrounding area by taking a bulked sample. We describe how bulking can be optimised using geostatistics when we have information about the local spatial variation. This is illustrated with two case studies. To resolve the longer scale component of variation a suitable sampling intensity must be chosen. If estimation by some method of interpolation is the aim then the sampling should be such that neighbouring samples are spatially correlated. The variogram, the central tool of geostatistics, describes the correlation structure, and it can also be used for estimation by kriging. This is illustrated with one case study. Using the variogram and the data we estimated soil properties and yield (dry matter) in a field from data on a square 20-m grid for mapping. The data on the 20-m grid were then sub-sampled to provide data on square grids of 40-m, 60-m and 100-m. The effects on the estimates of both reducing the number of sampling points and increasing the sampling interval were examined visually, and by comparing the estimates with the original values statistically. Finally, an optimal sampling interval for the field studied was determined using geostatistics.

Dr M. A. Oliver and Z. L. Frogbrook, Department of Soil Science, The University of Reading, Whiteknights, Reading, UK.

36 pages, 13 figures, 7 tables, 12 refs.