| Grain temperature and moisture content (MC) are considered to be the principle factors for the safe storage of grain. Continuous monitoring of temperatures within a grain mass is relatively easy using thermocouples, but monitoring of MC is limited by the availability of sensors. However, temperature and relative humidity (RH), can be used to predict grain MC based on equilibrium moisture content (EMC) equations such as the Modified Henderson, Chung-Pfost or Oswin equations. Current limitations for EMC prediction are sensor accuracy and the applicability of a single EMC regression equation for all grain types.; Temperature and RH sensor error was used to determine the total error in grain MC prediction using established EMC relationships. Error inherent in published EMC regression model standards (+/-2.15% MC to +/-3.8% MC) was greater than the contribution of sensor error (approximately +/-0.5% MC to +/-1% MC) between storage conditions of 20% to 80% RE. Outside these RH ranges, sensor error can contribute substantially (+/-2% MC to +/-8% MC at 95% RH) to the total error. Development of EMC models that exclude ranges of RH above 80% and below 20% may be better in order to develop EMC prediction equations with smaller standard errors.; EMC model parameters for the Modified Henderson and the Modified Chung-Pfost equations were determined from 47 samples of hard red winter, hard white winter, soft red winter, and Durum wheat. A model representing all the above classes of wheat resulted in a standard error of estimate (SEE) of 0.38% MCwb between 8% to 16% moisture content on a wet basis.; Grain physical and chemical properties such as bulk density, hardness, and protein content, which are highly related to the genetic characteristics of the grain, are believed to strongly influence the EMC properties. Relationships between EMC model parameters and grain physical and chemical properties were examined to determine if grain properties could help predict differences between individual models. Resulting correlations were poor and thus grain properties were not considered to be beneficial for model incorporation. Similarly, near-infrared (NIR) spectra was used to predict the EMC regression model parameters (A, B, C) but was found to have no useful predictive ability. Partial least squares (PLS1) was used in this case to develop prediction models. |
