NIRS Prediction Equations

Development Of NIRS Prediction Equations

The NIRS prediction equations were developed through compositional evaluations of a large number of corn hybrids by the UW corn breeding project, the UW Corn Silage Research Consortium, and the UW Extension corn program in the Department of Agronomy during 1992, 1993, 1995 -2003. Replicated silage trials were conducted at numerous locations throughout Wisconsin. Silage samples (~1kg) from each plot were collected at approximately 65% whole-plant moisture. Samples were oven dried at 60oC for approximately seven days, then they were ground with a hammer mill to pass a 1-mm screen. Each year all samples were scanned on a NIRSystems 6500 near-infrared reflectance spectrophotometer. Standard NIRS procedures were used to select calibration sets for wet laboratory analyses (Martens and Naes, 1989; Shenk and Westerhaus, 1991, 1994).

Samples from each calibration set were analyzed for NDF, crude protein, starch, and IVTD. A modification to the NDF procedure was the treatment of samples with 0.1 ml (per sample) of alpha-amylase during refluxing and again during sample filtration as described by Mertens (1991). Total nitrogen was determined using a Leco Model FP528 nitrogen analyzer. Crude protein was calculated by multiplying total nitrogen by 6.25. Starch concentration was estimated as in Bal et al. (2000). All compositional data were calculated on a dry matter basis. Duplicate 0.25-g samples were used to determine IVTD by a modification of the method of Goering and Van Soest (1970). The 48-hour fermentation was performed using using procedures equivalent to Tilley and Terry (1963) except that buffer and mineral solutions were as described by Goering and Van Soest (1975). After removal from the incubation tubes samples were placed in a freezer. Undigested residue was subjected to the NDF procedure as described previously.

The calibration sets used in 1992, 1993, 1995 – 2003 were combined in order to provide a single global calibration set for NDF, crude protein, and IVTD. The calibration set for starch concentration was obtained from 293 samples collected from 1995 to 2003. For all components, laboratory data were used to develop prediction equations relating NIR wavelengths to each quality parameters following the guidelines of Shenk and Westerhaus (1991, 1994). Criteria used to select equations were high coefficients of multiple determination and low standard errors of calibration and cross validation. Modified partial least square (PLS) analyses were used to determine what wavelengths to include in calibrations (Martens and Naes, 1989).

Data from the UW Corn Silage Breeding and the UW Extension programs have also incorporated relevant information from 2004 to 2011. The results of this most recent update are presented in the following table and graphs summarizing NIRS prediction statistics used for predicting silage quality in 2006.


The prediction equations will continue to be periodically upgraded by including spectral and compositional data from new samples that appear to have unique spectral properties. The UW NIRS prediction equations are available under license from WARF (#P01304US).