西湖龙井茶开采期影响因子及预报模型

基于西湖龙井茶主栽品种（龙井43和龙井群体种）开采期及气象资料，应用积温和逐步回归方法，分别构建西湖龙井茶的积温预报模型和逐步回归预报模型，并利用多元线性回归方法，对两个模型的预报结果进行集成，构建集合预报模型。结果表明：积温预报模型、逐步回归预报模型、集成预报模型均通过0.01水平的显著性检验；龙井43和龙井群体种的积温预报模型回代检验平均绝对误差（MAE）分别为3.6d和2.8d，2a试预报MAE分别为2.5d和1.0d；逐步回归预报模型的回代检验MAE分别为0.9d和1.4d，2a试预报MAE分别为1.6d和0.8d；集成预报模型的回代检验MAE分别为0.7d和1.1d，2a试预报MAE分别为1.3d和0.8d。3种预报模型对西湖龙井茶开采期预报均具有应用价值，集成预报模型较积温预报模型和逐步回归预报模型的预报效果更理想，具有实际生产指导作用。

Research on the Factors of Xihulongjing Tea Picking Date and Its Prediction Model

Based on the picking date of the main species of Xihulongjing Tea (Longjing43 and Longjingqunti) and its meteorological data, the accumulated temperature prediction model and the stepwise regression prediction model of Xihulongjing tea were constructed by using accumulated temperature and stepwise regression method, also the prediction of multiple regression ensemble method was integrated by these two prediction results, using the multiple linear regression method. The results showed that the accumulated temperature prediction model, stepwise regression prediction model and integrated prediction model all passed the significance test of P<0.01. The simulated mean absolute error（MAE） of accumulated temperature prediction model were 3.6d and 2.8d, while the prediction MAE of 2-year test prediction were 2.5d and 1.0d for Longjing43 and Longjingqunti respectively. In addition, the simulated MAE of stepwise regression analysis were 0.9d and 1.4d, the prediction MAE of 2-year test prediction were 1.6d and 0.8d for Longjing43 and Longjingqunti separately. The prediction of multiple regression ensemble method was more accurate than single method with the simulated MAE value were 0.7d and 1.1d ,while the prediction MAE of 2-year test prediction were 1.3d and 0.8d, the prediction of multiple regression ensemble method would provide more scientific support for guiding tea production. These three forecasting models are of practical value for the prediction of the picking up period of Xihulongjing tea. The prediction of multiple regression ensemble method is more ideal and with more practical application value than accumulated temperature forecasting model and stepwise regression analysis model.