Plant Trait Estimation from Hyperspectral Imagery in a Heterogeneous Herbaceous Environment
Roelofsen, Hans1; Bodegom, van, Peter2; Kooistra, Lammert3; Witte, Flip1
1KWR Watercycle Research Institute, Nieuwegein, NETHERLANDS; 2Department of Ecological science, subdepartment Systems ecology, VU University, Amsterdam, NETHERLANDS; 3Laboratory for Geo-Information Science and Remote Sensing, Wageningen University, Wageningen, NETHERLANDS

By hosting photosynthetic processes and storing carbon, the world's biosphere plays an important role in the global carbon cycle and climate regulation. An objective of earth system sciences is to quantify the exchange of carbon, water and nutrients between the biosphere and atmosphere. This is frequently done using airborne and space borne imaging spectroscopy products, by quantitatively estimating leaf chemical constituents (sensu: plant traits). However, examples so far appear to be limited to particular biomes with dense homogeneous canopies where tree crowns span one or more pixels. To achieve global comprehension of vegetation - atmosphere interactions, plant trait estimations urgently need to be expanded to other ecosystems.

We explored possibilities for this by estimating plant traits in herbaceous vegetation in a temperate biome. Such canopies are a complex 3D constellation of many small (< 1m height) plant species. As such, the spectral signature of the canopy is a weighted mixture of individual plant spectra, intertwined with contributions from soil background, woody components, dead organic matter and shadow. In addition to the complex spectral properties, high variation in plant trait values over short distances was also anticipated.

Airborne hyperspectral data (APEX airborne hyperspectral scanner, 400-2500nm, 300 spectral bands, developed under the ESA PRODEX program) were related to plant traits measured in 40 vegetation plots using partial least squares regression. Community mean traits identified included leaf N, P and C content (LNC, LPC, LCC), Chlorphyll a and b (Chl-a, Chl-b), lignin, tannin and phenols, as well as specific leaf area (SLA) and leaf dry matter content (LDMC) and spanned a wide range of abiotic conditions (wet-dry, nutrient poor-rich, acidic-alkaline). Special care was taken to acquire a properly mixed vegetation sample that was representative for the entire plot.

Preliminary analysis suggests that reflectance is strongly related to canopy concentration of LCC (NIR), SLA (VIS + NIR), lignin (NIR) and tannin (NIR). Perhaps surprisingly, Chl-a and Chl-b content of the canopy appear to show only weak correlation with spectral data. LNC is correlated to leaf chlorophyll and suffers from poor correlation as well. This could be due by the low variation in reflection in the VIS regions, where influence of chlorophyll is known to be pronounced. LDMC appears equally poor related to APEX spectra, possibly due to APEX lacking spectral bands in spectral regions where absorbance by canopy water is expected (around 1400 and 1940 nm).

Band selection techniques will be applied to identify the relevant spectral bands and improve the trait estimations. By comparing with studies in other biomes, we aim to identify spectral regions that are consistently relevant for trait estimation. Finally, as APEX is designed as simulator for future hyperspectral spaceborne imagers, this dataset can evaluate spectral configurations in their ability to map traits in herbaceous environments. So far, the results suggest that expanding global trait estimation to herbaceous vegetation is feasible, but the estimation accuracy can vary between those traits.