Hyperspectral VNIR-spectroscopy and imagery as a tool for monitoring herbicide damage in wilding conifers

International audience

English. Wilding conifers, the naturally regenerated scions of exotic conifer species, are a major ecological and economic issue in New Zealand. Hyperspectral imagery could provide a useful means of monitoring the efficacy of wilding control measures but little research has examined this possibility. Data were collected from an experiment, that included thirty-nine young Pinus contorta (Douglas) trees allocated to a control and two herbicide treatments (triclopyr butoxyethyl ester and diquat dibromide monohydrate). The objectives of this research were to examine the temporal impacts of herbicide on needle discolouration, spectral properties and tree physiology and to develop models to predict photosynthetic parameters from hyperspectral data. In accordance with obvious discolouration on the first day after treatment, values of net photosynthesis rate (A), electron transport rate (ETR), stomatal conductance (G) and optimal quantum yield of photosystem II (Fv/Fm) declined rapidly for plants treated with diquat. Discolouration following application of triclopyr was not evident before the second week, and values of A, ETR, G and Fv/Fm declined slower than those of diquat. Of the tested narrow-band spectral indices calculated from needle spectra photochemical reflectance index (PRI) exhibited the strongest correlation with the four photosynthetic parameters and R2 values between PRI and A, Fv/Fm, ETR and G were, respectively, 0.53, 0.46, 0.68 and 0.29. Models that used raw spectra as inputs showed that sparse partial least squares (SPLS) outperformed the three other approaches tested (PLS, Support Vector Machines, Elastic Net) and R2 values using this approach for A, Fv/Fm, ETR and G were, respectively, 0.64, 0.62, 0.65 and 0.42.