COMPARISON OF NARROW BAND VEGETATION INDICES AND EMPIRICAL MODELS FROM HYPERSPECTRAL REMOTE SENSING DATA FOR THE ASSESSMENT OF WHEAT NITROGEN CONCENTRATION B. Siegmann a, *, T. Jarmer a , H. Lilienthal b , N. Richter b , T. Selige c , B. Höfle d a Institute for Geoinformatics and Remote Sensing, University of Osnabrueck, Barbarastraße 22b, D-49076 Osnabrueck, Germany - (bsiegmann, tjarmer)@igf.uni-osnabrueck.de b Julius Kühn-Institut, Institute for Crop and Soil Science, Bundesallee 50, D-38116 Braunschweig, Germany - holger.lilienthal, nicole.richter)@jki.bund.de c Institute of Soil Ecology, TU Munich, Ingolstädter Landstrasse 1, D-85764 Neuherberg, Germany - selige@wzw.tum.de d Institute of Geography, University of Heidelberg, Berliner Straße 48, D-69120 Heidelberg, Germany - bernhard.hoefle@geog.uni-heidelberg.de KEY WORDS: hyperspectral, AISA-DUAL, wheat nitrogen concentration, empirical regression models, narrow band vegetation indices ABSTRACT: The precise assessment of canopy nitrogen status is one of the key parameters in agriculture for high accuracy yield estimations. The increasing availability of airborne imaging hyperspectral sensors (e.g. HyMap, HySpex, CASI, AISA) provides the required data to derive canopy nitrogen status for large agricultural areas with a high spatial resolution. In this study the potential of vegetation indices – red edge inflection point, normalized difference red edge index and normalized difference nitrogen index – and empirical regression models – support vector regression, partial least squares regression – have been compared for the prediction of biomass nitrogen concentration of wheat from AISA-DUAL data. For empirical regression models the best result was found for support vector regression (r 2 cv =0.86, RMSE cv =0.25, RPD=2.52) while the best result for vegetation indices was found for red edge inflection point (r 2 cv =0.69, RMSE cv =0.35, RPD=1.83). The comparison proves a higher potential of empirical regression models to deliver predictions for biomass nitrogen concentration of wheat. The transfer of the SVR model to the AISA-DUAL data allowed to map the spatial distribution of N concentration with reasonable accuracy and reflected the spatial pattern of N of the investigated fields very well. 1. INTRODUCTION Nitrogen is one of the most important crop limiting factors and a key parameter for crop monitoring and yield estimation in precision farming (Vigneau et al., 2011). Therefore, the assessment and mapping of total canopy nitrogen (N) content of agricultural crops is very important to optimize nitrogen fertilizer management in agronomy. An efficient and precise use of N-fertilizer is helpful to improve yield, reduce costs and lower environmental pollution at the same time (Ju et al., 2009). Spectral reflectance of plants in the visible (VIS) and near infrared (NIR) region of the electromagnetic spectrum is primarily affected by plant pigments (e.g. chlorophyll) and cellular structure of the leaves. Plants with limited N-uptake will have a lower chlorophyll concentration which is an indicator for non-optimal photosynthesis (Clevers & Kooistra, 2012). In this context, hyperspectral remote sensing data showed already a high potential for the spatial and non destructive estimation of chlorophyll- and N-concentration. The availability of airborne hyperspectral imaging systems (e.g. HyMap, HySpex, AISA and CASI) in the last years allows acquiring data with high spatial and spectral resolution, supporting the fast assessment of N-status from agricultural fields (Jarmer & Vohland, 2011; Dorigo et al., 2007; Kokaly, 2001). In this study the performance of narrow band vegetation indices and empirical regression methods derived from hyperspectral AISA- DUAL data is comparative investigated to retrieve detailed information about the spatial distribution of N-concentration from wheat field in Germany. * Corresponding author