The Use of Spectroscopy for the Rapid Evaluation of Physical-Chemical Soil Properties in the Veneto Region (NE Italy).

Sustainable agricultural management increasingly depends on rapid and reliable methods for assessing soil physical and chemical properties, particularly in regions facing intense agricultural pressures. Conventional laboratory analysis of soils remains time-consuming, costly, and often involves the use of hazardous materials, creating a need for alternative methods. This thesis addresses these challenges by exploring and validating spectroscopic techniques for the rapid evaluation of key soil properties – specifically, soil organic carbon (SOC), texture (clay, sand, silt), and available phosphorus (POlsen) – in the Veneto region of Northeastern Italy, a highly diverse and agriculturally significant area. A comprehensive set of surface soil samples was collected over a 20-year period as part of regional soil surveys by the Veneto Regional Environmental Protection Agency (ARPAV). 400 samples, which were determined to be representative of the region’s pedological diversity, were analyzed using three spectroscopic approaches: Visible-Near Infrared (Vis-NIR), Near-Infrared (NIR), and Mid-Infrared (MIR) spectroscopy. To reflect in situ conditions, four incremental moisture levels (10%, 17%, 23%, 30% (g/g)) were introduced to a subset of 200 samples, and spectral data were acquired in the Vis-NIR and NIR range. Advanced calibration methods, including Local and ReSampling Local (RS-Local) approaches, as well as customized preprocessing combination techniques such as Multiplicative Scatter Correction (MSC), Savitzky-Golay smoothing (SG), Wavelet transformation, and External Parameter Orthogonalization (EPO), were applied. Predictive models were developed using Partial Least Squares Regression (PLSR), Random Forest (RF), and Cubist machine learning algorithms. Results indicate that Vis-NIR and NIR spectroscopy, when combined with robust calibration and correction strategies, can deliver rapid and accurate predictions for SOC and soil texture, especially under field conditions. MIR spectroscopy demonstrated strong predictive performance for SOC and texture, and showed slightly improved performance when fused with NIR spectra. However, available phosphorus prediction (POlsen) remained unreliable across all spectral approaches, reflecting inherent spectral limitations for this property.