Abstract: he prediction model to drought risk for maize in different growing stages was established, by using multiscale SPI, discriminant analysis and linear sliding average method, based on the data of precipitation during each ten days and maize yield in Chaoyang city, Liaoning province. The results showed that the accuracy of the model increased with the growing stages promoted and got to 83.8%. Accuracy of the model for highrisk group, lowrisk group, and normal group were different, which was highest and could be 90.9% to highrisk group, and was lowest and could be 78.9% to normal group. It had highest accuracy at milkyripening stage, which increased to 83.8%from 67.6% at former stage. It indicated that the milkyripening stage was the key growing stage for maize, had more prominent influence on the final yields. The results indicated that the prediction model had high accuracy, could meet the demand of business service, especially for the regions where drought was the main disaster.

Key words: Risk prediction, Drought disaster, Different growing stages of maize, Multiscale SPI, Discriminant analysis