Abstract: Based on disease index for winter wheat powdery mildew and meteorological data in Zhengding, Xinji, Guantao and Cixian in Hebei Province during 2001-2010, the key period and key factors were selected using correlation analysis, path analysis, Bayes criterion, Fuzzy math and generalized regression neural network. Bayes grade index model, Fuzzy model and GRNN model were constructed for dynamic early warning of occurrence degree of wheat powdery mildew. The results indicated that the key factors affecting occurrence degree of wheat powdery mildew were mean temperature from the previous three pentads to the current pentad, precipitation from the previous three pentads to the current pentad, rain coefficient from the previous three pentads to the current pentad, and the previous pentad occurrence degree of wheat powdery mildew. The three early warning models had progressively transformed relationships, and the prediction accuracy rates were all above 85%, which could be well applied in early warning of wheat powdery mildew at the pentad scale. Compared with Fuzzy model and Bayes grade index model, GRNN model had the highest prediction accuracy rate. The results can provide useful information that contributes to a better understanding of occurrence of wheat powdery mildew in Hebei and help make policy for disease risk management.

Key words: wheat-maize planting system, carbon footprint, wheat-summer direct-seeding peanut planting system, greenhouse gas emissions