Characterization of variability and uncertainty in leaf biophysical traits using spectral data

Alexey N Shiklomanov, Boston University, ashiklom@bu.edu (Presenter)
Michael Dietze, Boston University, dietze@bu.edu
Shawn Paul Serbin, Brookhaven National Laboratory, sserbin@bnl.gov

Leaves are the primary pathway for exchanges of energy, water, and carbon between plants and their environment. Constraining leaf traits in models is necessary to improve representation of many ecosystem processes, including primary production, interspecies competition, and water and energy fluxes between the soil, vegetation, and atmosphere. Understanding the scales that dominate the variability in leaf traits is the first step to identifying the processes that drive this variability and for directing future research in plant physiology and forest ecology. In this project, we investigate the variability in leaf structure and biochemistry using a novel Bayesian inversion of the PROSPECT-4 leaf optical properties model. We focus on spectral data because it is easy and fast to collect in-situ, and in the long term, provides an important link to remote sensing. The Bayesian approach has two distinct advantages: First, Bayesian parameter estimates are not point estimates but joint probability distributions, wherein important information about uncertainty, covariance, and skewness is implicitly embedded. Second, this approach accommodates prior information independent of the data at hand. Prior constraints on our inversion model parameters were obtained from literature review of past inversion studies. Structuring the model to include random effects allowed explicit partitioning of uncertainty in model estimates between different sources, including individual leaves, leaf canopy positions, species, plant functional types, plots, and years. We obtained leaf reflectance and transmittance spectra and biochemistry data (e.g. d15N, LMA, and water content) from 14 forested ecosystems in the Eastern and Central USA. This data was used to validate our inversion model as well as to establish statistical relationships between model outputs and directly measured physiological properties. We then applied our inversion model to leaf reflectance spectra from the NASA HyspIRI Spectral Library, which encompasses a far larger range of biomes including temperate, boreal, and alpine forests, arid shrubland, and various agricultural crops. Although inter-species differences contributed the greatest amount to overall variability in leaf structure and biochemistry, we also observed substantial variability between leaves of the same species. Furthermore, we found that PROSPECT inversion performance is highly dependent on both species and the parameter of interest. These results have important implications for simulation of energy fluxes and physiological processes in ecosystem models.

Presentation: 2015_Poster_Shiklomanov_239_110.pdf (2481k)

Presentation Type:  Poster

Session:  General Contributions   (Tue 4:35 PM)

Associated Project(s): 

• Serbin, Shawn: Assimilation of imaging spectroscopy data to improve the representation of vegetation dynamics in ecosystem models ...details

Poster Location ID: 239