Prediction of Soil Properties with VIS-NIR-SWIR Reflectance Spectroscopy and Artificial Neural Networks : A Case Study on three Pedoenvironments of the Campania Region, Italy

A study was carried out to investigate whether basic soil properties can be predicted by using reflectance spectrometry GS) in the visible-near infrared- shortwave infrared (VIS-NIR-SWIR, 350-2500 nm) region using an artificial neural network (AI.IN) approach. Over 330 soil samples from three agro- forestry areas, representative of the pedo-environmental variability of the Campania region, Southern Italy, were used. The soil properties determined using conventional analyses were sand, silt, clay, organic carbon (OC), and calcium carbonate (CaCO3). Spectral reflectance (SR) measurements on soil samples were carried out under laboratory conditions, using a high resolution ASD FieldSpec spectroradiometer. Relationships between soil properties and soil SR were determined using ANN algorithms. The results obtained showed that clay content and OC can be predicted with high accuracy, while Sand and CaCO3 can be predicted with moderate to relatively high accuracy, respectively, and silt with relatively low accuracy. Improvement in the prediction of soil properties is expected using a larger number of soil samples for the training of the ANN algorithm, in combination with statistically based methods.