Spectrometric estimation of leaf pigments in Norway spruce needles using band - depth analysis, partial least - square regression and inversion of a conifer leaf model

Empirical and physical approaches to estimate leaf pigments in Norway spruce needles are compared. Foliar samples from 13 stands of Norway spruce, that are heterogeneous in terms of soil nutrient availability were collected (n=78). Foliage was separated by age class and subjected to routine biochemical analysis for chlorophyll a and b. Needle reflectance of stacked layers was measured using a high spectral resolution spectroradiometer. Three sets of reflectance were used for further analysis: i) 1 nm spectral resolution, ii) degraded to HyMap spectral bands, and iii) HyMap spectral bands with a normally distributed noise component added (σ=0.002). From reflectance first-derivative of reflectance, continuum removed reflectance, and normalized band-depths were calculated. Relations between spectra and pigments were developed using stepwise multiple linear regression (SMLR) and partial least square regression (PLSR). The conifer leaf model LIBERTY was inverted using an artificial neural network (ANN). LIBERTY was used in the forward mode to simulate needle stack layer reflectance based on typical leaf parameters. First-derivative of modelled reflectance was used to train the ANN. The trained ANN was then applied to the first-derivative of measured reflectance. For validation purpose the empirical relations and the trained ANN were applied to an independent data set obtained from a different field site. Estimated were compared to measured pigment concentrations. PLSR performed best on the calibration data set (in terms of r 2 and rmse). On the validation data set, inversion of the LIBERTY model achieved smallest rmse values for the 1 st year needles and SMLR for the 3 rd year needles.