Near infrared spectroscopy prediction of mineral content in botanical fractions from semi-arid grasslands

Near infrared reflectance spectroscopy (NIRS) was assessed for its capacity to estimate the mineral content of semi-arid grassland samples. NIRS calibrations were derived for P, K, Ca and Mg contents. Four populations of samples were used: total herbage, with a heterogeneous and complex botanical composition, and its botanical components (grasses, legumes and forbs). One set of samples from each population was selected to develop the specific calibration equations using three mathematical data treatments (log 1/R, first derivative, and second derivative). Reference values from the calibration sample set were regressed on the corresponding spectral data using stepwise multiple regression analysis. The equations were validated with samples from the same four populations that had not been included in the calibration. The NIRS method afforded acceptable accuracy in the prediction of P, K, Ca and Mg contents in the total herbage population and its botanical fractions. Botanical composition and mathematical treatment affected both the accuracy and precision of NIRS analyses; in this sense, the best fits were usually obtained using samples of simpler botanical composition (legumes and grasses), while the first derivative usually led to better results in the estimation of most parameters.