Vertical gradients of photochemicals and nutrient allocation within soybean canopies are well accounted for, but the patterns of hyperspectral reflectance and transmittance intensity collected throughout plant canopies are not well documented in the literature even as hyperspectral remote sensing in agriculture approaches ubiquity. This research quantifies hyperspectral data variability within Node, Branch, and Leaflet canopy levels between plants subjected to nitrogen and drought treatments through exhaustive destructive sampling. The dataset consists of the mean transmittance and reflectance hyperspectral signatures from all leaflet images from 36 greenhouse-grown soybean plants under control, low nitrogen, and a dry-down treatments. Samples were collected over three intervals once they reached stage V4. Pairwise comparisons of positions at each canopy level result in spectral angles and percent differences of NDVI and SPAD as representative vegetative indices (VIs). The mean spectral angles found through Node-level comparisons from both reflectance and transmittance data were less than that at the Branch-level between adjacent leaves (Transmittance: 0.058457 versus 0.088385 radians) while the comparisons were overall more variable (Transmittance IQR = 0.056957 versus 0.045362 radians). The angles between leaflets were centered close to zero. These magnitude and variability trends extended to VI data. Best fit Ordinary Least-Squares Polynomial Regressions successfully modeled each treatment/maturity subgroups’ vertical distributions of VI intensity. Kruskal-Wallis with Post Hoc Conover’s tests determined significance of adjacent branch discrepancies with respect to Node position and maturity. These magnitudes were directly proportional to node height and inversely proportional to maturity, though treatment group influenced specific cases. Treatment groups were significantly segregated based on VI values sampled from 1, 2, or 3 node locations. Effective sampling strategies are highlighted with respect to research logistics, statistical significance, and maximized Euclidean distance between treatment groups. These findings inform future remote sensing stress-detection models that leverage intra-canopy variability.
