Prediction of soil properties using laboratory VIS–NIR spectroscopy and Hyperion imagery

Abstract

Conventional analyses of soil characteristic are expensive and time-consuming. Visible and near infrared reflectance spectroscopy (VIS–NIR) have been useful tools for quantitative analysis of numerous soil attributes. In this study, the fidelity of spatial structure of soil attributes was evaluated by geostatistical methods and elemental concentrations were mapped using Hyperion hyperspectral reflectance data (400–2500nm). Forty-nine soil samples were used to analyze soil organic carbon (SOC), total phosphorus (TP), pH, and cation exchange capacity (CEC). The performance of three different instrumental settings (laboratory, Hyperion and simulated Hyperion spectroscopy) was assessed using either partial least squares regression (PLSR) or stepwise multiple linear regression (SMLR). Models for SOC, TP, and pH showed moderate accuracy (R2>0.6, RPD>1.5), whereas that for CEC exhibited low efficiency (R2<0.5, RPD<1.4). To further evaluate the effect of predicted models, we conducted the fidelity of spatial structure by geostatistical methods and extrapolated VIS–NIR spectral techniques. In addition, areas with low SOC and high P concentration were identified and mapped, thus facilitating the development of remediation strategies in terms of application of fertilization and environmental protection. Finally, Hyperion image offers an alternative method for modeling soil properties and has the great potential for diagnosing the nutrient deficiency, assessing the risk of soil parameters on the environment.