Remote sensing is used to understand the composition, diversity, and structure of forested landscapes. A next step is to link these measurements to forest dynamics and scale-up our knowledge of tree demography from field plots to landscapes and regions. Information derived from hyperspectral canopy reflectance, such as foliar chemistry, structure, and species identity, may provide a link between remote sensing and forest productivity. We previously developed an empirical relationship between plot-level tree growth and canopy hyperspectral reflectance for single-species, even-aged reforestation plots in a tropical dry forest and identified a suite of indices that capture growth across multiple species. Our objective is to understand which leaf traits may be underlying the relationship between tree growth and canopy reflectance. We ask; 1) Which leaf traits are measurable with leaf reflectance, 2) of these traits, how well do they predict growth, and 3) how strongly can growth be predicted directly from leaf reflectance? To address these questions we measured stem growth via tree inventories, and measured foliar traits and leaf reflectance of the same trees. We used partial least squares regression (PLSR) to predict traits from reflectance and a linear mixed effects model to predict growth from measured and spectrally-derived leaf traits.
Results/Conclusions
A principle component analysis (PCA) of all leaf traits showed the primary variation was due to traits associated with photosynthesis (35%), and traits associated with defense (18%). Leaf reflectance explained 4-97% of the variation for each foliar trait, with the best-predicted traits including carbon to nitrogen ratio (C:N), leaf mass per area (LMA), percent nitrogen, and phenolic compounds. Macro and micronutrients such as magnesium, iron, and sodium were also detectable with R-squared values of at least 0.7. A multivariate linear mixed effects regression model using foliar concentration of LMA and phenolic compounds with a random intercept for species explained 48% of the variation in diameter growth rate (using leave one out cross validation). These two traits support the PCA analysis and highlight two functional axes of variation in tropical species, one that reflects the light acquisition and the other defense. Leaf reflectance alone explained 53% of the variation in growth rate using PLSR. Both interspecific and intraspecific variation in growth rates, traits, and reflectance among the 5 species were important in predicting growth. Overall our results demonstrate the potential to measure forest dynamics from remote sensing by predicting tree stem growth rate.