Browsing by Author "Sanhueza Novoa, Pamela Alejandra"

Now showing 1 - 1 of 1
  • Thumbnail Image
    Item
    Development of predictive methods based on VIS–NIR spectroscopy and artificial vision techniques for the detection of water stress in Eucalyptus spp. and a study of their recovery.
    (Universidad de Concepción, 2026) Sanhueza Novoa, Pamela Alejandra; Castillo Felices, Rosario del Pilar
    Water deficit is a major constraint on the productivity, establishment, and adaptability of Eucalyptus plantations, creating a need for rapid and non-destructive analytical tools capable of detecting plant responses before severe visible damage occurs. The main objective of this Ph.D. thesis was to develop and evaluate spectral methodologies based on visible–near infrared hyperspectral imaging (VIS–NIR HSI) and portable near-infrared spectroscopy, combined with chemometrics, machine learning, physiological measurements, recovery assessment, and candidate gene-expression analysis, for monitoring water-deficit responses in juvenile Eucalyptus plants. Two sequential controlled-environment assays were conducted. An initial assay involving four genotypes was used to evaluate the feasibility of VIS–NIR HSI for differentiating mild, moderate, and severe water-deficit conditions and to compare full-spectrum information with conventional vegetation indices. A second, expanded assay involving six genotypes integrated VIS–NIR HSI, portable NIR spectroscopy, physiological measurements, gene-expression analysis, and post-stress monitoring after rewatering. Principal component analysis was used to explore spectral structure, whereas ANOVA–simultaneous component analysis (ASCA) was applied to partition spectral variability associated with genotype, sampling day, treatment condition, biological replicate, and their interactions. Supervised classification models were evaluated for binary discrimination between control and pooled water-deficit samples and for multiclass discrimination among the control condition and three progressive water-deficit stages. Partial least squares-discriminant analysis, support vector machines, k-nearest neighbors, Random Forest, Extreme Gradient Boosting, and multivariate regression methods were evaluated within the analytical workflow. VIS–NIR HSI detected spectral changes associated with progressive water deficit more effectively than the evaluated vegetation indices. In the initial assay, supervised models based on mean spectra achieved external-validation errors between 0 and 2%, while support vector machine classification generated the most consistent spatial reconstructions of stress progression. In the expanded assay, binary discrimination between control and water-deficit samples was effective for both spectral platforms, whereas direct classification of the control condition and the three progressive water-deficit stages was less accurate because adjacent stress stages showed substantial spectral overlap. ASCA demonstrated that VIS–NIR HSI variability was dominated by genotype, which explained 30.2% of the spectral variance, whereas portable NIR variability was primarily associated with sampling day and treatment condition, explaining 40.2% and 10.6%, respectively. An ASCA-guided hierarchical HSI workflow achieved complete correct classification in the evaluated external-validation set, while portable NIR provided a more moderate but interpretable classification of water-deficit progression. Physiological measurements confirmed coordinated reductions in net CO₂ assimilation, stomatal conductance, photosystem II performance, and plant water status. Recovery after rewatering was incomplete and strongly genotype dependent, with the six evaluated genotypes showing contrasting degrees of restoration of water status and physiological function during the monitored recovery period. Water deficit induced DHN1, LEA4, and TRDX expression in genotype-dependent patterns, although high transcript accumulation did not necessarily indicate greater physiological tolerance. VIS–NIR HSI showed its strongest association with non-photochemical quenching, whereas portable NIR spectroscopy showed the clearest relationship with plant water potential. Overall, the results demonstrate that VIS–NIR HSI and portable NIR spectroscopy provide complementary, biologically meaningful information for non-destructive water-deficit monitoring in Eucalyptus. The proposed design-aware analytical framework supports controlled-environment phenotyping and genotype screening, although independent validation across new genotypes, experimental campaigns, nurseries, and field-like environments is required before operational implementation.
Síguenos en...