Optimum sampling algorithm for the prediction of soil properties from the infrared spectra

Spectroscopy data from three datasets were included in this study. These datasets have different coverages: a European national dataset (LUCAS, n = 5639), a regional dataset from Australia (Geeves, n = 379), and a local dataset from New South Wales, Australia (Hillston, n = 384). Three sampling algorithms: Kennard-Stone (KS), conditioned Latin Hypercube (cLHS) and k-means clustering (KM) were compared against random sampling on the prediction of up to five different soil properties (sand, clay, carbon content, cation exchange capacity and pH) on three datasets. The sampling algorithms were used to select various calibration sample sizes ranging from 50 to 3000 for the continental dataset; and from 50 to 200 samples for the regional and local datasets. All calculations were conducted in R statistical language on High Performance Computing Services provided by the University of Sydney.