Spectral soil analysis and inference systems: A powerful combination for solving the soil data crisis

Abstract

Diffuse reflectance spectroscopy (DRS) is attracting much interest in the soil science community because it has a number of advantages over conventional methods of soil analyses. The techniques are more rapid, timely, cheaper and hence more efficient at obtaining the data when a large number of samples and analysis are required. Moreover, a single spectrum may be used to assess various physical, chemical and biological soil properties. Until now, research in soil spectroscopy has focused on spectral calibration and prediction of soil properties using multivariate statistics. In this paper we show how these predictions may be used in an inference system to predict other important and functional soil properties using pedotransfer functions (PTFs). Thus we propose the use of soil spectral calibration and its predictions as input and as a complement to a soil inference system (SPEC-SINFERS). We demonstrate the implementation of SPEC-SINFERS with two examples. As a first step, soil mid-infrared (MIR) spectra and partial least squares (PLS) regression are used to estimate soil pH, clay, silt, sand, organic carbon content and cation exchange capacity. A bootstrap method is used to determine the uncertainties of these predictions. These predictions and their uncertainties are then used as input into the inference system, where established PTFs are used to infer (i) soil water content and (ii) soil pH buffering capacity together with their uncertainties. An important feature of SPEC-SINFERS is the propagation of both input and model uncertainties.