Depth Specific Estimation of Soil Properties using vis-NIR Spectroscopy
Agricultural digitization has demanded a large quantity of soil data for site-specific or precision agriculture. Spectroscopy has shown potential as an alternative characterization of soil properties over traditional laboratory methods. It has shown potential to estimate soil properties when used in laboratory, while its use in field remains a challenge to the global research community. This thesis studied various aspects of spectroscopy for soil characterization in laboratory to field conditions, including 1) optimization of preprocessing and modelling algorithms; 2) comparison of transformation methods to eliminate environmental effects on spectral data. The combination of 1st Derivative + Gap and Random Forest performed best to predict soil properties from dry ground samples. Direct Standardization outperformed External Parameter Orthogonalisation for transforming field wet spectral data. Finally, spectroscopy was used in situ to predict 5 soil properties. The research showed a strong promise of using soil spectroscopy for in situ and depth-specific soil characterization.