Potential of NIRS to assess nitrogen, phosphorus and potassium nutrient status of grasslands in La Réunion island

Controlled mineral fertilisation practices are an important component of a sustainable management of grasslands. Assessment of available nutrients and general recommendations on the level of phosphorus and potassium to apply are classically based on soil analysis. Nitrogen, being rapidly leaching, mid or long term recommendation cannot easily be derived from soil composition. Recent approaches tended to show that herbage plant N (Lemaire and Gastal, 1997), P, K (Duru and Huché, 1997) mineral analyses associated to actual biomass measurement, could be useful for the calculation of nutrient combined indexes (IN, IP, IK). Expressing along references curves and a standard optimum value of 100 the limiting factors or eventual excess in the mineral feeding of the plants. It provides a diagnosis on the actual main nutrient status in a specific plot situation. The step has been successfully implemented to provide local advise in the management of the Reunion Island grasslands (Blanfort, 1998). NIR high ability to predict direct nitrogen content is classical, while its use is generally less relevant for the other elements. However as the concern is more related to the appreciation of a combined index, it appeared interesting to test the potential of NIRs to predict these or classify the samples according to class grasslands nutrition levels. A large set of 900 herbage N-P-K content referenced powder samples, has been scanned on a Nirsystem 6500 (400-2500 nm). IN, IPand IK indexes, calculated according to mineral content and measured standing grass biomass were put into MPLS calibration. Along ranges 19 -106 (IN), 29 -199 (IK), 38-143 (IP), SEcv and R_cv values were: 7.5 - .81; 9.2 - .90 and 11.9, .62 for IN, IK and IP, respectively. Even if the narrowness of the relation appears lesser for nitrogen and phosphorus, when values are to be compared to a standard optimum of 100, NIRs prediction could be a useful tool for rapid and low cost diagnosis, all the more reason that it can be associated in a same scanning to classically well demonstrated prediction of the feeding value of the forage. (Texte intégral