Abstract:

In order to explore and find a suitable flowering prediction model for rape in Jiujiang area, the observational data of rape and meteorological data from 1995 to 2023, which were obtained from the Hukou County Meteorological Bureau, were used to construct the flowering prediction model for rape based on four methods: the grey correlation analysis method in pre−flowering phenology, the effective accumulated temperature method, the BP neural network and random forest regression. The simulation results of regression (1995−2018) and the prediction verification (2019−2023) for four models were compared. The results indicated that: (1) the grey correlation analysis method in the pre−flowering phenology of flowering prediction model for rape could be established based on the number of days from sowing to flowering, transplanting, budding and bolting, the model using BP neural network and random forest regression could be constructed based on seven meteorological factors, such as the average temperature in February, the sunlight situation in November, etc. (2)Among the effective accumulated temperature models established with different base temperatures and starting times, the model with a base temperature of 5℃ and reporting from the actual bolting had the best fitting effect. (3)Summarized the simulation results of the four models, the effective accumulated temperature method had the best applicability and high stability, which could be applied to the actual meteorological services for the rapeseed flowering period in JiuJiang area. 

Key words: Rape, Flowering prediction, Effective accumulated temperature method, Grey correlation analysis method, Pre?flowering phenology, BP neural network, Random forest regression