林木调查数据的随机、空间、时间特征的模型处理  

Solutions for Random Effect,Spatial and Temporal Correlation and Heterogeneity in the Observed Forest Data

作　　者：章皖秋[1] 岳彩荣[1] 袁华[1] 

机构地区：[1]西南林业大学林学院,云南昆明650224

出　　处：《西北林学院学报》2016年第5期230-237,共8页Journal of Northwest Forestry University

基　　金：国家自然科学基金(31260156);云南省教育厅基金一般项目(2014Y333);西南林业大学林学一级学科博士研究生创新基金项目

摘　　要：林木调查数据常具有随机性、空间性和时间性3种特征,这些特征造成了传统林木模型精度不高、应用不稳定。尽管已有多种建模方法可处理这些特征,但很少有研究界定这些方法在林木数据的适用范围。介绍混合效应模型、空间回归、地理加权回归、回归克里金4种建模方法在林木因子上的应用,分析它们对林木调查数据的随机效应、空间相关性与异质性、时间相关性与异质性上的适用度。结果表明,混合效应模型能有效处理林木数据的随机效应、空间相关与异质性、以及时间相关性;地理加权回归主要解决数据的随机效应、空间相关与异质性;空间回归与回归克里金只能处理数据的空间相关性;但这4种方法均无法有效处理数据的时间异质性。在实际研究中,可以依据这4种建模法对林木数据随机、空间和时间特征的适用度,分析数据特征来选择合适的建模方法,从而提高研究效率与精度。Randomness, spatial correlation and heterogeneity, and temporal trend are common characteristics of various observed forest data. These characters badly affect the precise and accuracy of traditional forest models. Although there are already many modelling ways which can handle these characters of forest data, few studies so far have addressed the application scopes of these methods. Here,in order to make clear their abilities in dealing with the randomness, spatial and temporal correlation and heterogeneity within the ob- served forest data, four modeling methods, mixed effect model, spatial regression model（SEM）, geographi- cally weighted regression（GWR） and regression kriging（RK）,were analyzed based on their theories and various research cases. It was found that mixed effect model had higher flexibility to deal with the random effect,spatial characteristics, and temporal correlation of forest data; and GWR could handle the random and spatial characteristics within forest data,while SEM and RK only provided solutions for spatial correla- tion. Additionally,these four methods could not effectively explain the temporal heterogeneity in the data. In conclusion, the application scopes of these four methods on solving the randomness, spatial and temporal correlation and heterogeneity in observed forest data had been defined, and it was supposed to provide clues for future forest studies on the selection of the suitable modelling method.

关 键 词：随机效应 空间相关性与异质性 时间相关性与异质性 混合效应模型 空间回归模型 地理加权回归 回归克里金 

分 类 号：S711[农业科学—林学]