Pimstein, A., Departamento de Fruticultura y Enología, Facultad de Agronomía e Ingeniería Forestal, Pontificia Universidad Católica de Chile, Av. Vicuña Mackenna, 4860 Santiago, Chile, The Remote Sensing Laboratory, Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Negev 84990, Israel
Karnieli, A., The Remote Sensing Laboratory, Jacob Blaustein Institutes for Desert Research, Ben-Gurion University of the Negev, Negev 84990, Israel
Bansal, S.K., Potash Research Institute of India, Dundahera Delhi-Gurgaon Road, Gurgaon 122001, Haryana, India
Bonfil, D.J., Field Crops and Natural Resources Department, The Institute of Plant Sciences, Agricultural Research Organization, Gilat Research Center, 85280 MP Negev 2, Israel

Given the importance of potassium (K) and phosphorus (P) contents to wheat yield and grain quality, and the very little experience that has been gained on nutritional monitoring of other than nitrogen using remotely sensed technologies, a study was undertaken to explore the possibility of identifying these mineral stresses using spectral data. Canopy spectra and biophysical data were collected from commercial and experimental fields in India and Israel. Traditional and newly developed vegetation indices, together with Partial Least Squares (PLS) regression models, were calculated in order to predict potassium and phosphorus contents from the wheat canopy spectral data. Results show that the application of PLS and specific narrow bands vegetation indices reached significant levels of accuracy in the retrieval of K and P levels, in comparison to traditional broad band indices. Additionally, it was observed that a significant improvement is obtained when the mineral total content is considered instead of the relative content. Therefore it was suggested that the biomass should also be retrieved from the spectral data. Finally, as very different crop conditions were included in this study, it was possible to confirm that the level of accuracy in the retrieval of K and P levels is related to the quality and variability of the data used for calibrating the models. © 2010 Elsevier B.V.