Clustering of spatially-resolved scattering and spectroscopy data for characterizing local variations in spruce wood

Wood is a hierarchical material composed of structural elements organized across multiple length scales, from the molecular arrangement of cellulose chains to the tissue-level organization of growth rings. However, the extent and type of local variations in wood nanostructure in the sub-millimeter scale remain insufficiently understood, despite their importance for the material’s mechanical behavior and moisture dynamics. Here the structural and chemical differentiation of two tissue types in Norway spruce, earlywood and latewood, was investigated using spatially resolved small-angle X-ray scattering (SAXS), wide-angle X-ray scattering (WAXS), and near-infrared (NIR) spectroscopy. Given the complexity and scale of the resulting datasets, machine learning techniques were integrated to enhance the analysis and classification of structural variations across different tissue types and moisture conditions. Principal component analysis was employed to reduce dimensionality, while clustering algorithms—such as Gaussian mixture model and spectral clustering—enabled classification of wood structures based on SAXS and WAXS fitting results and NIR spectral features. All the clustering methods were able to separate earlywood and latewood, although not always in exactly the same way. The study provides qualitative and quantitative insights into the hierarchical organization of wood tissues, revealing distinct nano-structural differences in cellulose microfibril arrangement and moisture response between earlywood and latewood. Especially, consistent differences related to microfibril angle were observed. The results demonstrate that machine learning-driven analysis effectively enhances interpretability and consistency in multimodal datasets, enhancing structural characterization in materials science.