A method for soil chromium (Cr) hyperspectral estimation based on coupling Cr-response bands with active constituent...

Heavy metals in soils typically exhibit weak or indirect spectral responses in hyperspectral data, limiting their direct detectability. This limitation arises because their spectral behavior is strongly mediated by spectrally active soil constituents, particularly iron oxides, which dominate absorption features and control spectral response pathways. To address this limitation, a coupled spectral feature framework was developed for the physically interpretable hyperspectral prediction of soil chromium (Cr), integrating Cr-feature wavelengths with those associated with dominant soil constituents to enhance indirect spectral coupling and mechanistic interpretability. A total of 85 soil samples were collected from Wuhan, China, characterized by low soil organic matter and slightly acidic to near-neutral pH conditions. Key spectral variables were extracted using a hybrid feature selection strategy combining the successive projections algorithm (SPA) and recursive feature elimination (RFE), and subsequently used to construct a partial least squares regression (PLS) model. The proposed SPA-RFE-PLS framework effectively reduced hyperspectral dimensionality and identified stable and informative spectral features associated with iron-bearing soil constituents across repeated model runs, indicating that Cr prediction is primarily driven by indirect coupling with spectrally active soil constituents rather than direct spectral absorption features of Cr. The model achieved strong predictive performance for Cr estimation ( $R_{\text{p}}^2=0.80$ ; RPIQ $_{\text{p}}^{}=3.12$ ). The proposed framework outperforms conventional approaches by explicitly leveraging this coupling mechanism, thereby improving both predictive accuracy and model interpretability. Overall, this study provides a mechanism-driven and physically interpretable framework for hyperspectral estimation of soil heavy metals, offering enhanced reliability.