Advancing Hyperspectral Remote Sensing of Foliar Plant Functional Traits: A Bayesian Hierarchical Regression Approach
Dhruva Kathuria, NASA Goddard Space Flight Center - Morgan State University, dhruva.kathuria@morgan.edu (Presenter)
Yoseline Angel, NASA Goddard Space Flight Center - UMD College Park, yoseline.b.angellopez@nasa.gov
Alexey N Shiklomanov, NASA Goddard Space Flight Center, alexey.shiklomanov@nasa.gov
Plant functional traits are essential for understanding plant performance, ecosystem dynamics, and informing biodiversity conservation and climate change adaptation strategies. Hyperspectral remote sensing, with upcoming satellite missions like Surface Biology and Geology (SBG), provides a promising avenue for assessing plant traits at various spatial and temporal scales. Partial Least Squares Regression (PLSR) is widely used for predicting foliar traits using hyperspectral data. However, using PLSR map traits over wide spatial and phylogenetic extents is challenged by the variability of the relationship between traits and spectra across species, functional types, and biomes.
In this study, we propose a Bayesian Hierarchical Regression (BHR) approach that explicitly accommodates variability in the relationship between traits and spectra across different groups (species, genus, family, etc.), effectively sharing information and improving parameter estimation (which is especially important for undersampled groups). The Bayesian framework allows for robust uncertainty quantification and efficiently handles missing data. We applied the proposed BHR approach to a large collection of datasets on spectra and traits from a wide geographic and bioclimatic range. Our BHR approach performed better than the conventional PLSR algorithm while providing parameter/prediction uncertainties. Furthermore, we identified the most significant sources of variation for each trait, which will inform the development of future retrieval algorithms. Our findings highlight the potential of the BHR approach for advancing hyperspectral remote sensing of plant traits and contribute to a better understanding of plant performance and ecosystem dynamics in the context of global environmental change.
Associated Project(s):
Poster Location ID: 3-4
Presentation Type: Poster
Session: Poster Session 3
Session Date: Thu (May 11) 3:00-5:00 PM
CCE Program: TE