Abstract
The joint interpretation of hyperspectral imaging (HSI) and laser-based 3D observations is challenged by differences in surface sampling, acquisition geometry, and spectral response. This proof-of-concept study evaluated internal spectral consistency and cross-sensor correspondence in independently generated 3D plant representations. We developed an accessible CPU-based workflow for dense, spectrally attributed 3D reconstructions of healthy and Sclerotinia sclerotiorum-infected lettuce plants (Lactuca sativa var. longifolia typ. Romana and var. capitata typ. Canasta) using multi-view VNIR push-broom HSI (204 bands, 400–1000 nm). The resulting models were compared with independently acquired PlantEye (PE) multispectral 3D laser scans using an external reference for registration. A view-selection policy for spectral attribution reduced raw reflectance dispersion by ~64%, achieving 98.7–99.5% canopy spectral assignment coverage and high spatial specificity (median contributing views SAM: 1.68–2.37°). While differences in acquisition geometry precluded point-wise sensor equivalence, both modalities effectively discriminated inoculated from non-inoculated leaves within each plant using four common spectral bands via spatially grouped cross-validation (balanced accuracy: 0.923–0.930 for ‘Canasta’; 0.824–0.866 for ‘Romana’). Furthermore, full 204-band HSI models yielded improved tissue discrimination. The results highlight the need to distinguish internal spectral consistency, local cross-sensor correspondence, and task-level information when evaluating heterogeneous spectral–3D observations for future joint use.