• Hyperspectral features of SOC-related factors improve SOC estimation accuracy. • SOC sequestration is regulated by complex interactions, not single factors. • Indirect spectral models outperform direct models in SOC prediction accuracy. • Soil enzyme activity shows strongest hyperspectral response, aiding SOC estimation. The accurate estimation of soil organic carbon (SOC) is crucial for effective soil resource management, improvement agricultural productivity, and mitigating climate change. To investigate the hyperspectral response mechanisms of SOC and its influencing factors, this study was conducted based on a long-term conservation tillage experiment in a winter wheat field. Soil samples were collected from a depth of 0–60 cm during 2020–2023. Hyperspectral data were analyzed using multiple preprocessing and feature selection techniques, and prediction models for SOC and its influencing factors were constructed. The quantitative relationships between different transformed spectra and SOC, as well as its influencing factors, were determined. The optimal spectral features for SOC and its influencing factors were identified. Based on the response of SOC content to its stabilization mechanisms, the optimal spectral features of influencing factors were reorganized to form new characteristic wavelength combinations. These combinations were further used to develop an indirect spectral model for SOC prediction. This approach enabled the identification of the spectral response mechanisms of SOC stabilization processes. The results indicated that spectral reflectance was inversely correlated with SOC, GSSI (general soil structure index), Dm (fractal dimension of mechanical aggregates), MI-OC (microaggregate organic carbon), and SSA (soil sucrase activity). The characteristic regions for SOC were primarily distributed at 400–430 nm, 675–691 nm, and 713–754 nm. These regions were primarily located within the visible spectrum and did not overlap with the optimal characteristic wavelengths of its influencing factors. In the direct modeling of SOC, the optimal preprocessing and wavelength selection methods were the 1st-order differentiation and SBS (sequential backward selection), respectively. RF (random forest) was identified as the most effective modeling method (R v 2 = 0.69, RMSE v = 0.32, MAE v = 0.25). In the spectral estimation methods for SOC influencing factors, the 1st-order differentiation was the optimal preprocessing method, and RF demonstrated superior performance as the modeling approach. The spectral response effects of SSA, GSSI, mechanical aggregate parameters, and FSP-OC (total Organic Carbon of Free Silt and clay Particles) were higher than those of SOC. The indirect model for SOC constructed from the optimal combination of characteristic wavelengths of its influencing factors exhibited higher accuracy (R v 2 = 0.74, RMSE v = 0.27, MAE v = 0.21) than the direct model. SOC sequestration was influenced by the combined effects of multiple soil factors rather than a single indicator. The results confirm the feasibility of hyperspectral SOC monitoring and support its application in rapid soil fertility assessment.
Yang et al. (Thu,) studied this question.