High resolution reflectance (hyperspectral) measurements have potential for rapidly measuring plant function at spatiotemporal scales far greater than are possible with traditional methods (e.g., CO2 gas exchange). However, to date, most ground-based hyperspectral data come from highly productive agricultural or tropical settings and it is unclear if physiological traits in less productive ecosystems, such as drylands, can be assessed using hyperspectral data. Advances in this arena would support forecasts of productivity relationships with climate, would increase our understanding of species responses to global change, and would improve measurement of ecological functioning in drylands, which cover ~40% of the Earth’s terrestrial surface. We tested whether we could accurately predict in-situ rates of photosynthesis and total leaf nitrogen (N) in three native shrubs: Artemisia tridentata (sagebrush), Larrea tridentata (creosote bush), and Prosopis glandulosa (honey mesquite). These common species frequently comprise a majority of surface cover and are dominant shrubs within the vegetation communities of North America’s Great Basin, Mojave, and Chihuahuan Deserts, respectively. For each species, we paired leaf-level measurements of hyperspectral reflectance (350–2500 nm @ 3-8 nm resolution recorded with an ASD FieldSpec4 Hi-Res portable spectroradiometer) with photosynthetic CO2 response curves (collected with a Licor 6400) from 40 shrubs at each desert study site (N= 120). We harvested the leaves used in gas-flux and hyperspectral chambers to determine photosynthetic leaf area/biomass and to quantify leaf N concentrations. We used partial least squares regression (PLSR) models, validated via a ‘leave-one-out’ approach, to estimate rates of CO2 gas exchange and N concentration from full spectrum profiles (350-2500 nm) for each shrub.
Results/Conclusions
In spite of substantial variation in rates of photosynthesis and N concentration within and among the species, PLSR models were able to predict rates of photosynthesis with a high degree of accuracy (r2 ≥ 0.72 across shrub species). PLSR models for N concentration were slightly less accurate but suggest great potential for hyperspectral assessment of plant nutrient status in drylands. Our ability to predict rates of photosynthesis with hyperspectral data in this study indicates strong opportunities for remote sensing to significantly advance our understanding and capacity to predict dryland functioning.