Near-infrared spectroscopy for moisture content prediction in soil-mixed woody biomass

Accurate measurement of moisture content (MC) in woody biomass is essential for preventing biological degradation during storage and transport, particularly because mineral soil contamination commonly occurs during harvesting. This study evaluated the feasibility of Near-Infrared (NIR) spectroscopy (870 to 2500 nm) as a rapid, non-destructive alternative to oven-drying for quantifying the MC of soil-mixed woody biomass. To reflect field variability, 24 oven-dried samples were reconditioned to obtain a broad MC range (3 to 16%, dry basis). Two preprocessing techniques, Standard Normal Variate (SNV) transformation and Savitzky-Golay (SG) second-derivative filtering, were applied to improve spectral quality and model robustness. The combined SNV + SG preprocessing, integrated with Partial Least Squares Regression (PLSR), yielded the highest accuracy. For logging residues, the optimized model achieved R2 of 0.81 (RMSE = 1.50) in calibration and R2 of 0.79 (RMSE = 1.57) in cross-validation, demonstrating feasible MC estimation despite substantial heterogeneity and soil interference. These results suggest that NIR spectroscopy, when paired with appropriate chemometrics, provides reliable MC prediction across diverse contamination levels and biomass types. Although soil content could not be accurately estimated in this study, future work should develop improved methods for quantifying contamination and extend the approach to portable NIR systems for real-time field applications.