Exploring the spectral patterns of chemical elements in soils through proximal VIS-NIR spectroscopy and ML (a pilot study: Armenia)

Abstract. Armenia has limited agricultural soil resources, which varies in type, vertical spatial distribution and quality and the information on soil chemical, physical and biological attributes are very limited. To establish datasets to cover these gaps and to develop the best combination of the methods of soil qualitative and quantitative assessments and classification of their spatial variability dedicated studies has been conducting. This work is aimed to investigate and reveal the main characteristics of the spectral reflectance of the chemical elements in soils in Armenia using a combination of proximal VIS-NIR and X-ray spectroscopy methods. 30 soil samples were collected from different regions of Armenia characterized with different landscape properties. The content of a number of chemical elements were determined by X-ray spectroscopy and spectral signals of the samples via field spectroradiometer were collected. The comparison of the data of X-ray and VIS-NIR spectroscopy defines three clusters, led by the soil macro components: I cluster - SiO₂, Al₂O₃, Fe₂O₃; II cluster – CaO and III cluster – K₂O. Correlation analysis of the content of these macro components and their spectral reflectance values shows that they vary in spectral patterns. In conclusion it can be stated that the main macro components in the studied soil types have specific spectral patterns. The responsible spectral bands are mainly visible and NIR (400–750nm) and two SWIR bands (800–1750nm and 1950–2400nm). Among the studied elements, the exception was CaO as the most well distinguished macro component over the full spectral range 350nm to 2500nm, except (1750–1950nm).