Quantifying and partitioning uncertainties in retrievals of plant traits from imaging spectroscopy data

Abstract:

Plant functional traits are measurable plant characteristics that are related to plant function, condition, and competitive ability and often serve as important parameters in vegetation models. Observations of traits through space and time are therefore highly relevant to plant biogeography and vegetation modeling. Several decades of research have demonstrated that plant traits can be successfully estimated from leaf and canopy reflectance spectra, and mapping plant traits is one of the main objectives of the upcoming NASA Surface Biology and Geology designated observable.

Estimating traits using optical remote sensing is a multi-step process. First, the raw sensor measurements (digital numbers) have to be converted to top-of-atmosphere radiance, which involves accounting for the spectral response function and signal-to-noise ratio of the instrument. Second, estimating surface reflectance from this radiance requires atmospheric correction. Finally, estimating plant traits from this reflectance requires empirical approximations or semi-mechanistic radiative transfer models of the relationships between the target traits and reflectance. Each of these steps introduces uncertainties into the final traits estimates.

Our objectives are (1) to develop a system for quantifying uncertainties associated with estimating plant traits from imaging spectroscopy data, and (2) to use this system to investigate how these uncertainties are influenced by different surface characteristics, atmospheric conditions, and instrument properties. To meet these objectives, we created a workflow that first simulates satellite measurements from known surface and atmospheric characteristics and subsequently tries to retrieve the known characteristics from the simulated measurements.