Spectral Predictability of Soil Organic Matter Depends on Its Humin Fraction Rather than Spectral Fusion

Soil organic matter (SOM) governs critical soil functions, including carbon storage, nutrient cycling, and microbial activity; yet the specific fractions responsible for its spectral predictability remain poorly understood. This study addresses a fundamental research gap by comparing visible–near-infrared (vis–NIR), mid-infrared (MIR), and fused spectroscopy for predicting SOM and its components: humic acid (HA), fulvic acid (FA), and Humin. Using 93 soil samples from subtropical croplands in southeastern China, we employed partial least squares regression with full spectra and LASSO-selected wavelengths to build predictive models. Results demonstrated that both vis–NIR and MIR individually provided moderately strong predictive performance for SOM and Humin (R2 = 0.79–0.90, CCC = 0.85–0.93), while FA remained unpredictable (R2 < 0.24) due to weak, overlapping spectral features. The strong predictability of SOM was primarily attributed to the Humin fraction, which comprises approximately 50 percent of total SOM and exhibits abundant spectrally active functional groups. Contrary to expectations, spectral fusion did not improve predictions because both spectral regions already contained complementary information, and fusion introduced redundancy and scale imbalance rather than increasing effective dimensionality. This study establishes that accurate SOM estimation depends fundamentally on the predictability and abundance of the Humin fraction, providing new mechanistic insights for spectroscopic soil carbon monitoring and highlighting the need for component-specific modeling approaches in soil organic matter research.