296 The impact of multiple calibrations on the accuracy of near-infrared analyses in grain markets.

C. R. HURBURGH, Jr. (1), G. R. Rippke (1), and T. J. Brumm (2). (1) Agricultural and Biosystems Engineering, Iowa State University, Ames, IA 50011; (2) MBS, Inc., Story City, IA 50248.

The use of near-infrared (NIRS) analyzers in grain trading is increasing. NIRS requires a multivariate calibration to convert spectral data to predicted calibration values for constituents such as protein, oil, starch, and hardness. Different calibrations for the same factor, developed from separate databases targeted at specific genetic situations, are in use, which leads to the potential for discrepancies. A study of four corn calibrations and four soybean calibrations for Infratec analyzers showed 20 to 200 percent increases in standard deviation across individual units, even when all calibrations statistically matched chemical references. The accuracy of calibrations developed for particular situations was 20 to 80 percent worse than broad-based calibrations when diverse material was analyzed. These results demonstrate the need to select market use calibrations carefully, to have the maximum uniformity and applicability.