OOS 81-1
Relationships among leaf traits and leaf spectra: Prediction, clustering and functional types

Thursday, August 13, 2015: 1:30 PM
341, Baltimore Convention Center
Keely L. Roth, Land, Air & Water Resources, University of California Davis, Davis, CA, USA
Angeles Casas, Land, Air & Water Resources, University of California Davis, Davis, CA, USA
Margarita Huesca, Land, Air & Water Resources, University of California Davis, Davis, CA, USA
Michael L. Whiting, Land, Air & Water Resources, University of California Davis, Davis, CA, USA
Susan L. Ustin, Land, Air & Water Resources, University of California Davis, Davis, CA, USA
Background/Question/Methods

Ecology has been moving from conventional plant functional types (cPFTs) towards traits-based functional types. Spectroscopy data, at leaf and canopy scales, are widely successful for estimating functional traits (e.g., pigments, water, dry matter). Growing availability of these data, including the potential launch of global hyperspectral sensors (e.g., NASA’s Hyperspectral Infrared Imager (HyspIRI)), emphasizes the importance of understanding how accurately they can be used to estimate leaf functional traits. Determining how leaf spectral signatures relate to both cPFTs and traits-based functional types may enable us to leverage this knowledge to improve remotely-sensed trait estimates.

Using leaf spectra and traits measured from a range of species and cPFTs over multiple sites and seasons in California, we evaluated three approaches for estimating traits: Partial Least Squares Regression (PLSR), PROSPECT radiative transfer model inversion, and Multiple Gaussian (MG) spectral fitting. Using hierarchical clustering analysis, we created sets of optical leaf functional types (oLFTs; based on spectra) and traits-based leaf functional types (tLFTs). Lastly, we tested whether incorporation of either cPFT or oLFT information improved spectral estimates of traits using PLSR and PROSPECT inversion.

Results/Conclusions

Results using PLSR and PROSPECT inversion support previous findings (PLSR: R2 values from ~0.6-0.8 and PROSPECT: RMSE values ~0.011 g/cm2 (water) to 24 μg/cm2 (chlorophyll)). MG showed promise for estimating pigment and water concentrations simultaneously. PLSR and MG were sensitive to the set of values used to calibrate estimates, which varied here by species, site and season, and PROSPECT inversions relied on representative trait ranges. Clustering analysis revealed greater variation within the spectra than in the traits, likely because only a subset of traits was included in this analysis. oLFTs corresponded more closely with tLFTs than with cPFTs. While some clustering by cPFTs, site and species was observed, not all oLFTs were clearly distinct in these attributes, indicating the importance of inter-site and intra-specific variation. Preliminary results suggest that the incorporation of oLFTs can improve the accuracy of trait estimates using spectral data.

The oPFT concept offers a great opportunity for expanding the utility and applications of spectroscopy data within ecological research. We show spectral data are good representatives of combinations of key functional traits at leaf-level. Understanding how oPFTs relate to suites of functional traits is valuable when using spectral data across large, diverse regions not only to estimate traits, but also for addressing questions about diversity and its impacts on ecosystem services, as well as community composition and assembly.