Comparison of different multivariate calibrations and ensemble methods for estimating selected soil properties with vis-NIR reflectance spectroscopy

Sustainable soil management requires a correct assessment of soil chemical and physical properties. Historically, this has been gained through conventional laboratory analyses, which are considered costly and time-consuming, particularly when a large number of soil samples need to be analysed. An alternative, faster and less expensive, approach is based on the use of reflectance spectroscopy in the visNIR domain. This approach implies the calibration of predictive models that relate the spectral reflectance to soil properties. The goodness of the models can be particularly influenced by the multivariate methods used. In this article, we compare the performance of different multivariate and statistical ensemble methods for estimating some basic soil properties, such as sand, silt, clay, and organic carbon in the specific pedo-environmental conditions of an important agricultural area in southern Italy.