Building relationships between plant traits and leaf spectra to reduce uncertainty in terrestrial ecosystem models

Abstract:

Despite climate projections, there is uncertainty in how terrestrial ecosystems will respond to warming temperatures and increased atmospheric carbon dioxide concentrations. Earth system models are used to determine how ecosystems will respond in the future, but there is considerable variation in how plant traits are represented within these models. A potential approach to reducing uncertainty is the establishment of spectra-trait linkages among plant species. These relationships allow the accurate estimation of biochemical characteristics of plants from their shortwave spectral profiles. Remote sensing approaches can then be implemented to acquire spectral data and estimate plant traits over large spatial and temporal scales. This paper describes a greenhouse experiment conducted at Brookhaven National Laboratory in which spectra-trait relationships were investigated for 8 different plant species. This research was designed to generate a broad gradient in plant traits, using a range of species grown in different sized pots with different soil type. Fertilizer was also applied in different amounts to generate variation in plant C and N status that will be reflected in the traits measured, as well as the spectra observed. Leaves were sampled at different developmental stages to increase variation. Spectra and plant traits were then measured and a partial least-squares regression (PLSR) modeling approach was used to establish spectra-trait relationships. Despite the variability in growing conditions and plant species, our PLSR models could be used to accurately estimate plant traits from spectral signatures, yielding model calibration R² and root mean square error (RMSE) values, respectively, of 0.85 and 0.30 for percent nitrogen by mass (N_mass%), R² 0.78 and 0.75 for carbon to nitrogen (C:N) ratio, 0.87 and 2.39 for leaf mass area (LMA), and 0.76 R² and 15.16 for water (H₂O) content. This research forms the basis for establishing new and more comprehensive spectra-trait relationships that could be used to reduce uncertainty in the parameterization of models representing the structure of functioning of terrestrial ecosystems. In the future, remote sensing from unmanned aerial systems and satellites could provide greater temporal and spatial scaling to take spectral signatures on global vegetation.