A Bayesian Approach to Hyperspectral Leaf Trait Prediction with uncertainty quantification

Dhruva Kathuria , Yoseline Angel, Evan Lang, Alexey N Shiklomanov

Leaf functional traits are leaf features that determine ecosystem functioning, plant growth, and resource allocation. Most traits can be derived from leaf reflectance measurements across the visible to shortwave infrared range using empirical and physical methods. Partial Least Squares Regression (PLSR) is widely used but has limitations in uncertainty quantification and model flexibility. In this study, we present a Bayesian approach for predicting leaf traits from reflectance data (400–2400 nm at 1 nm resolution) that addresses these limitations. The method eliminates spectral transformation while enabling rigorous uncertainty quantification. We apply it to predict carotenoid content (Car A ), nitrogen percentage mass (N M ), and Leaf Mass per Area (LMA). On an independent validation dataset, the Bayesian approach performs comparably to PLSR with added flexibility and robust uncertainty quantification. To enhance computational efficiency, we project the full model to a reduced model using selected wavelengths (14 for Car A , 28 for N M , and 30 for LMA), maintaining predictive performance while enabling faster predictions and trait-specific wavelength insights. The Bayesian method is highly adaptable, supporting future development of nonlinear, hierarchical, and multivariate models with rigorous uncertainty quantification.