Construction of soil moisture three-band indices with Vis-NIR spectroscopy based on the Kubelka-Munk and Hapke model
Jing Yuan, Yuteng Liu, Changxiang Yan, Chunhui Hu, Jiawei Xu
Abstract
• Soil moisture three-band indices based on the Kubelka-Munk and Hapke model were constructed, which possess clear physical interpretability. • The r KM , r Hapke , and ω were used to replace the reflectance R in constructing spectral indices, which are more correlated with soil moisture. • The spectral indices demonstrate robust performance across soils with varying organic matter contents and diverse soil types. Monitoring soil moisture (SM) helps optimize irrigation and increase crop yields. The SM indices with visible near-infrared (Vis-NIR) spectroscopy can provide real-time and non-destructive information. However, the current construction of SM spectral indices is predominantly based on empirical parameterization methods, lacking a solid physical foundation. Additionally, the existing spectral indices are constrained to two-band forms and are all based on several specific forms. In this study, SM three-band indices (TBIs) based on the Kubelka-Munk (KM) and Hapke model were constructed. The converted reflectance ( r ) and the single scattering albedo ( ω ) were used to replace the reflectance ( R ) in constructing spectral indices. The selection of spectral indices forms, sensitive bands and their corresponding optimal spectral bandwidths was carried out based on correlation coefficients and cross-validated coefficient of determination ( R 2 CV ). Based on the field measurement data, the result of the comparative strategy indicates that the modeling performance of these spectral indices constructed based on the KM and Hapke model ( R 2 CV : 82.13%-87.48%) outperforms those based on R ( R 2 CV :52.39%-84.44%). In addition, these spectral indices developed in this study also demonstrate robust performance across soils with varying organic matter contents and diverse soil types. These SM spectral indices, derived from the soil radiation transfer model, possess clear physical interpretability and significantly reduce the complexity of model calibration in the SM prediction process. They enable the efficient development of soil property maps with both rapid processing and high prediction accuracy.