Cluster-based spectral models for a robust assessment of soil properties

Abstract

Soil spectroscopy has proven efficient for providing quantitative properties of soil chromophores and soil classification. However, building generic/global models might bring to light some problems which stem from the false premise that each chromophore affects all soil types equally. The objective of this paper is to offer a new cluster-based approach for the assessment of soil properties and compensating issues that might rise from using heterogeneous soil dataset. For that purpose, 1480 soil samples from varying climate zones in Israel were used. All samples were air-dried, ground, and sieved to 2 mm, then CaCO3, organic matter (OM), clay and sand contents together with the spectral signature were measured according to laboratory protocols. The spectral analysis was performed using the Spectral Angle Mapper (SAM) algorithm, spectral gradient and k-means clusters to split the dataset into distinct clusters. Subsequently, cluster-based models were performed and compared with generic models. Later, we focused on the OM parameter and applied the OM detection limit approach for a more robust and accurate assessment. Our results show that soil property prediction improved significantly when using the cluster-based models compared to the generic models and therefore should be considered when dealing with large and heterogeneous databases.