Importance of connectivity grains for AusScan NIR prediction accuracy

The AusScan near-infrared (NIR) calibrations for predicting available energy content of cereal grains for pigs and broiler chickens, are based on results from many experiments that commenced in the mid-1990s (Black et al. 2014). Robust NIR calibrations require information from hundreds of measurements. Limitations in infrastructure capacity, concurrent availability of grains varying widely in characteristics that affect energy availability, and research funds meant that many small experiments were conducted and the results aggregated to develop the NIR calibrations. Results from the first three years of the Premium Grains for Livestock Program could not be used for development of NIR calibrations because only one grain (∼3% of grains in an experiment) was constant across experiments and this was insufficient to satisfactorily adjust for differences between experiments. Consequently, ~30% of grains (known as connectivity grains) used in each experiment have been included in previous experiments to account for variations in environmental conditions across experiments. Inclusion of connectivity grains reduces the number of new grains included in each experiment and increases cost. The impact of including connectivity grains on variance of available energy values and therefore accuracy of NIR calibrations was assessed. The value of connectivity grains was assessed for pig faecal digestible energy (DE), when grains were fed without enzymes, and for broiler apparent metabolisable energy (AME) for combined gender, males and females, when grains were fed with and without enzymes. For each assessment, the unadjusted (raw) measured values with standard errors (SE) were compared with statistically adjusted values using connectivity grains (Table 1). Inclusion of connectivity grains reduced SE of the estimated energy content of grains across all comparisons by 48%, with the decrease in SE ranging from 0.079 MJ/kg as fed (25%) for male broilers with enzymes to 0.231 MJ/kg as fed (82%) for combined gender broilers without enzymes.

Table 1.  Effect of statistically correcting across experiments for connectivity grains on mean grain energy content (MJ/kg as fed) and SE for all grains for which NIR calibrations are available

The impact of reducing SE of measurement on the accuracy of NIR calibrations can be estimated because the accuracy is approximately twice the mean SE of the values used to develop the calibration. Thus, including connectivity grains improved the accuracy of NIR predictions from ± 0.16 (0.079*2) MJ/kg as fed for male birds fed diets with enzymes to ± 0.46 MJ/kg as fed for combined gender broilers fed diets without enzymes. The corresponding value for pig faecal DE was ± 0.22 MJ/kg as fed. An increase in error of prediction from the NIR calibrations of these magnitudes, if connectivity grains were not used, would substantially reduce the practical value of NIR calibrations for use by the pig and broiler industries. These analyses indicate that inclusion of connectivity grains should be continued for future experiments.