Answers are grounded in this paper and cite their sources.
Leaf nitrogen content (LNC) is a key indicator for nutrient management in olive. This study compared spectral preprocessing and predictive models for estimating olive LNC from laboratory hyperspectral reflectance. A total of 120 composite leaf samples collected from a semi-arid Moroccan orchard were analyzed for LNC and measured with an ASD FieldSpec spectroradiometer over 350–2500 nm. Six preprocessing pipelines were evaluated across four modeling tiers: fixed and optimized vegetation indices, partial least squares regression (PLSR), support vector regression (SVR), and a one-dimensional convolutional neural network (1D-CNN). Derivative-preprocessed spectra yielded lower RMSEP than raw or scatter-correction-only spectra. PLSR with standard normal variate (SNV) + detrend + first-derivative preprocessing gave the best cross-validated results, with R2CV = 0.79, RMSEP = 0.127% DW, and RPD = 2.20. Its performance was numerically higher than that of the 1D-CNN (R2CV = 0.75), the best optimized vegetation index (R2CV = 0.73), and SVR (R2CV = 0.70). Variable importance in projection analysis indicated that predictive information was concentrated in the visible and red-edge regions, while a secondary shortwave-infrared contribution at 1134–1146 nm became apparent only after SNV and first-derivative preprocessing. These internal cross-validation estimates indicate potential, not transferability to other orchards, cultivars, sampling campaigns, or field sensing.
No takes yet. Share an insight, caveat, or question.
Eljanyani et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: