Based on the hourly data of the regional automatic weather stations in Ningbo, an analysis is made of the variability coefficients of sunshine duration, temperature, and humidity, as well as the relation between these meteorological elements and flowering periods. The results show that hour by hour accumulated temperatures (℃〖DK〗·h) indicate better relation compared with day by day accumulated temperature, and the predictors are obtained based on the extreme variability and correlation coefficients. The BP neural network method is applied to set up mid term flowering forecast models, and ECMWF fine grid model products are used as well. Trial forecasts perform quite well in forecasting blossom time and duration. The method can help improve meteorological service for agricultural activities.