基于贝叶斯克里金的山东省小麦产量时空相依模型

Spatio-temporal model of wheat yield in Shandong Province based on Bayesian Kriging

为解决农作物产量预测中的相依性问题,在贝叶斯层次模型的基础上,使用GB2分布替代正态、逻辑斯特等分布形式,并对其尺度、形状、峰度、偏度参数同时引入时间效应t以及采用克里金方法体现的空间效应,与此同时在模型中增加多种协变量以探究影响产量的因素。基于山东省小麦产量数据的实证结果表明:相较于对时间趋势和产量分布分别拟合的两步法,嵌入式模型更好的模拟农产量的时空相依特征、减少预测误差的叠加;GB2分布的引入为拟合农产量提供更大的灵活性;克里金方法的引入改进了模型的预测效果。基于克里金方法的嵌入式时空模型可有效提高小麦产量预测的准确性、降低区域产量保险费率。

In order to solve the issue of dependence in crop yield prediction, on the basis of Bayesian hierarchical model, GB2 distribution is used instead of normal, logistic and other distribution forms, time effect t and Kriging method for spatial effect are also introduced into the scale, shape, kurtosis and skewness parameters of GB2 distribution, at the same time, a variety of covariates are added to the model to explore the factors that affect the yield. The empirical results based on wheat yield data in Shandong Province show that, compared with the two-step method which fits time trend and yield distribution separately, the embedded Bayesian model better simulates the characteristic of spatio-temporal dependence in crop yield and reduces the superimposition of prediction errors;The introduction of GB2 distribution provides greater flexibility for fitting crop yield;The introduction of the Kriging method improves the prediction of the model;The embedded spatio-temporal model based on the Kriging method can effectively improve the accuracy of wheat yield prediction and reduce the area yield insurance premium.