Development of empirical forecasting models for rice blast based on weather factors

Regression equations used as empirical models to predict rice blast caused by Pyricularia grisea on cv. Jinheung at Icheon, South Korea, and on cvs. IR50 and C22 at Cavinti, Philippines, were generated, using weather factors identified by the WINDOW PANE program to be highly correlated with disease. Consecutive days with RH≥80% (CDRH80), number of days with RH≥80% (NDRH80), consecutive days with precipitation, and number of days with precipitation ≥ 84 mm day^?1 were important variables predicting blast at Icheon. Total precipitation, precipitation frequency, mean maximum and minimum temperatures, number of days with wind speed above 3.5 m s^?1, CDRH80, and NDRH80 were important predictors of blast at Cavinti. The Allen's predicted error sum of squares (PRESS) criterion and a cross-validation procedure were used to evaluate the models using data that were not included in model development. Validation test showed that all models developed for the two sites, except the models predicting maximum lesion number and panicle blast incidence at Icheon, and panicle blast severity on IR50 at Cavinti, predicted blast reasonably well based on low PRESS values and close to zero average prediction errors. These models can be applied in actual rice production systems, but future validation is needed to further improve their predictive ability.