Multivariate geostatistics for precision agriculture
In very recent times, agricultural practices have begun to exploit within-field heterogeneity to obtain higher yielding crops and more environmentally friendly fertilizer- and pesticide-application schemes. Global positioning systems installed in new farm equipment can measures the application of fertilizer and insecticides down to meters, and they provide closely spaced data on crop yields. The possibility of characterizing within-field heterogeneity of a multitude of variables has brought precision agriculture (PA) to the modern American farm. One of the goals of PA is to map variables, such as soil properties, so that the farmer can apply fertilizer or insecticides sparingly. Because yield data is plentiful and typically correlates with other variables that are difficult to measure, it is reasonable to use information on past yield to make better maps of soil and entomological variables, for example. The area of study known as geostatistics not only provides maps of predictions from noisy, incomplete data, but also maps of prediction standard errors. In this paper, we apply a multivariate geostatistical technique known as cokriging and compare it to the univariate geostatistical technique known as kriging, for some within-field measurements on corn yield and soil pH in Iowa, U.S.A.