The development and transfer of undried grass silage calibrations between near infrared reflectance spectroscopy instruments

Abstract

The first objective of this paper was to compare the accuracy of calibration statistics for seven chemical and biological parameters, of undried grass silage, when developed on two differing types of scanning monochromators commonly used in the agricultural industry. One instrument (Foss NIRSystems 6500) uses a vertical transport mechanism while the second (Bran & Luebbe 500) uses a spinning cup attachment. Calibrations were produced using the modified partial least squares (MPLS) regression technique, in conjunction with either first or second derivatization and four scatter corrections. Optimum equation selection was based on the lowest cross validation error (SECV). The results showed few differences in calibration statistics between the two NIR instruments. The second objective of this paper was to clone the Bran & Luebbe (slave) instrument to the Foss (master) instrument and transfer the calibrations developed on the master to the slave. A range of undried grass silage samples ( n = 20), typical of the calibration set of silages, was selected to clone the two scanning monochromators using the ISI (Infrasoft International) cloning software. The software produces standardization files which when applied to spectra from the slave instrument makes them `look like' spectra from the master instrument and so allows accurate prediction by equations developed on the master instrument. Using an independent set of silages ( n = 33) to compare the predictions of the unstandardized and standardized slave spectra to the master spectra predictions, using the same equations, showed that the standard error of prediction (SEP) was greatly reduced and the R 2 increased after standardization of the spectra. The standardized spectra predictions were highly correlated to the master predictions, thus proving that this backward method of calibration transfer worked very successfully. In comparison the common method of sloping and biasing equations for use on other instruments was examined. Again the SEP values were reduced in line with the cloning method but the resultant slope and bias values were not improved over the cloning method. The average ` H' values were not reduced by sloping and biasing as this technique matches the equations and not the spectra. Calibrations for biological parameters transferred more successfully when the cloning technique was employed. An alternative approach was to use the forward method of calibration transfer, where the calibration set of spectra were standardized to `look like' they were scanned on the slave instrument and then the equations were rerun. These equations were then applied directly to the slave spectra. This method also proved successful but slightly less accurate than the backward calibration transfer. This study has demonstrated that this method of cloning monochromators of two different types and without the use of sealed sample sets or generic standards, has proven very successful even with forages of high moisture content.