摘要
在农作物区域产量保险研究中,农作物产量模型具有十分重要的基础性地位。农作物单产数据呈现出复杂的时空相依关系,产量的均值、方差、偏度受技术进步、气候变化等因素影响呈现趋势变化,不同地区产量受气候、生产聚集性等因素影响呈现非对称偏相依特征,并且随着外部环境的变化而变化。因此,对农作物产量建模,需要准确刻画其趋势性、异方差性和空间相依的时变性特征。本文基于空间因子模型理论,用高斯过程刻画变量中间部分的空间相依关系;用厚尾分布对高斯过程的尾部相关性结构进行调整,得到非对称左尾强相依结构,并通过对因子分布的参数建模,使得空间左尾相依强度随着外部环境而变化;此外,在对边际分布的方差建模时,考虑了时间和灾害的影响,得到了非对称的异方差模型。本文所构造的模型能够反映农作物区域产量的时空相依性、时空异质性及其时变性特征,具有较强的可解释性。基于一组区域产量数据的实证研究结果表明,该模型在拟合农作物空间产量数据的时空相依关系时,显著优于目前广泛使用的建模方法。
The crop yield model has very important basic position in crop area yield insurance.Crop yield data shows complex spatial and temporal dependence.The mean,variance and skew of yield tend to have substantial trends due to technological advance and climate change.Affected by climate and production aggregation,crop yield shows time-varying asymmetric skew dependencies across space and time.Therefore,modeling crop yields requires an accurate description of the trending,heteroscedasticity and time-varying spatially dependence.Based on the theories of spatial factor model,we describe the spatial dependence of the middle part of yield by Gaussian process.Fat tail distribution is used to adjust the tail correlation structure to obtain a model with stronger dependence in the lower tail.Modeling of the parameters of factor distribution makes the spatial left-tail dependence changes with the external environment;in addition,when modeling the variance of the marginal distribution,we consider the influence of time and disaster,and obtain the heteroscedastic model with jumps.The model we proposed can reflect the spatial and temporal dependence,spatial and temporal heterogeneity and time-varying characteristics of crop yields.An analysis based on a set of county level yield data shows that the model we proposed performs much better than the currently widely used models in fitting the spatial-temporal dependence of crop yield data.
作者
张译元
孟生旺
ZHANG Yi-yuan;MENG Sheng-wang(School of Mathematics,Harbin Normal University,Harbin 150025,China;School of Statistics,Renmin University of China,Beijing 100872,China)
出处
《数理统计与管理》
CSSCI
北大核心
2022年第5期786-802,共17页
Journal of Applied Statistics and Management
基金
2020年度黑龙江省省属高等学校基本科研业务费科研项目(2020-KYYWF-0357)
2021年度黑龙江省属高等学校基本科研业务费科研项目(2021-KYYWF-0189)。
关键词
空间相依
因子模型
异方差
时变性
spatial dependence
factor model
heteroscedasticity
time-varying
