Abstract. Milling yields of long-grain rice were successfully predicted using multi-parameter linear modeling, although the data was limited to lots of only two cultivars (Grigg and Siebenmorgen, 2015). Thus, modeling was expanded to include six currently-produced long-grain cultivars. For unfractioned rice, milling yields [milled rice yield (MRY) and head rice yield (HRY, adjusted to account for degree of milling variations)] were regressed to combinations of physical parameters comprising bulk density (Ib) of rough rice, and fissured-kernel percentage (FKP), chalkiness, and average kernel thickness of brown rice. The resulting models for prediction of milling yields were cross-validated with equivalent data of thickness-fractioned rough rice of the same cultivar lots. Thickness fractions comprised ( 2.05 mm). Milling yields were first predicted using Ib, FKP, and chalkiness parameters (Model I). The adjusted coefficient of determination (R 2 ) value indicates that Model I effectively predicted HRY (R 2 = 0.90); however prediction of MRY was poor (R 2 = 0.22), and related model cross-validations (Valid. R 2 = 0.67 and 0.59, MRY and HRY, respectively) were marginal. Adding the kernel thickness parameter in Model II improved milling yield predictions and cross-validations (R 2 = 0.72 and 0.94, and Valid. R 2 = 0.82 and 0.74, for MRY and HRY, respectively). As the chalkiness parameter estimates in Models I and II were statistically insignificant, Model III considered only the Ib, FKP, and thickness parameters. Model III maintained effectiveness in predictions and cross-validations of milling yields (R 2 = 0.71 and 0.95, and Valid. R 2 = 0.83 and 0.72, for MRY and HRY respectively). The Ib, FKP, and thickness parameters of Model III effectively predicted MRYs and HRYs of both unfractioned and thickness-fractioned long-grain rice of this set of cultivar lots.
Read full abstract