Abstract: Based on daily meteorological data from 1961 to 2020 and maize developmental stages data from 1980 to 2019, water deficit index was selected to construct drought grade sequence during maize growing season and to analyze characteristics of maize drought in Jilin province. Dynamic early warning model of maize drought grade was built and evaluated based on rank correlation analysis and Bayes discriminant analysis. The results showed that the west and middle of Jilin province and Yanbian city were typical maize drought areas. Maize drought had high frequencies at two stages, which were sowing to seedling and jointing to tasseling. The basic accuracy rate (BAr) of dynamic early warning model on spring drought of maize was 60%−90%, while that was 80%−100% on strangle hold drought in typical maize drought areas. Besides, average BAr of forecast test of early warning of drought grade in different subdivisions were more than 90%. The BAr of forecast test of early warning model was better in the east part of Jilin province than that in the west and middle. In 2020, the accuracy rate (Ar) of forecast test on spring drought was 55.7%−78.7%, and the Ar of forecast test on strangle hold drought was 60.7%−80.3%. The BAr between early warning grade and actual grade was over 91%. The early warning model of maize drought grade based on Bayes discriminant analysis was suitable for Jilin province.

Key words: Maize, Early warning of drought grade, Bayes discriminant analysis, Water deficit index, Jilin province