Autumn canopy phenological transitions are increasing in length as a consequence of climate change. Here, we assess how well hyperspectral indices in the visible and near-infrared (NIR) wavelengths predict nitrogen (N) concentrations in lower-canopy leaves in the autumn phenological transition as they are generally understudied in leaf trait research. Using a Bayesian framework, we tested how well published indices are able to predict N concentrations in Fagus grandifolia Ehrh., Liriodendron tulipifera L., and Betula lenta L. from mid-summer through senescence, and how related the indices are to autumn phenological change. No indices were able to determine a trend in differences in N in mid-summer leaves. Indices that included wavelengths in the green and NIR ranges were the first indices able to detect a trend and had among the highest correlations with N concentration in both the last green collection and the senescing collection. Models were unique when indices were fit to data from different phenophases. Indices that focused on only the red edge (i.e., the sharp increase in reflectance between the red and NIR wavelengths) had the strongest explanatory power across the autumn phenological transition, but had less explanatory power for individual collections. These indices, as well as those that have been correlated with chlorophyll (CCI) and carotenoids (PRI), were the strongest descriptors of autumn progression. This study provides insights on challenges and capabilities to monitor a leaf's N concentration throughout and across canopy senescence.
Keywords: Autumn; Hyperspectral; Leaf traits; Phenology; Resorption.