地理研究 ›› 2001, Vol. 20 ›› Issue (6): 739-751.doi: 10.11821/yj2001060012

• 论文 • 上一篇    下一篇

黄土丘陵小流域土壤水分空间预测的统计模型

邱扬1,2, 傅伯杰1, 王军3, 陈利顶1   

  1. 1. 中国科学院生态环境研究中心系统生态开放实验室, 北京 100085;
    2. 北京师范大学资源与环境科学系, 北京 100875;
    3. 国土资源部土地整理中心, 北京 100035
  • 收稿日期:2001-04-19 修回日期:2001-09-17 出版日期:2001-12-15 发布日期:2001-12-15
  • 作者简介:邱扬(1969-),男,四川仁寿人,博士。研究方向为景观生态与土地质量、植物生态与自然保护、 土壤侵蚀与水土保持、干扰生态与环境生态。在国内外核心刊物发表中英文论文30余篇,参编专著1 部。E-mail:qiuyang69@hotmail.com
  • 基金资助:

    中国科学院知识创新工程项目(KZCX2-405);国家杰出青年科学基金资助项目(49725101)

Spatial prediction of soil moisture content using multiple-linear regressions in a gully catchment of the Loess Plateau, China

QIU Yang1,2, FU Bo-jie1, WANG Jun3, CHEN Li-ding1   

  1. 1. Department of Systems Ecology, Research Center for Eco-Environmental Sciences, CAS, Beijing 100085, China;
    2. Department of Resource and Environmental Sciences, Beijing Normal University, Beijing 100875, China;
    3. Land Consolidation and Rehabilitation Center, Ministry of Land and Resources, Beijing 100035, China
  • Received:2001-04-19 Revised:2001-09-17 Online:2001-12-15 Published:2001-12-15

摘要:

在6个土层和10次土壤含水量测定的基础上,利用土地利用与地形等6类20个环境因子变量,建立了黄土丘陵区小流域土壤水分空间预测的6种多元线性回归模型,并提出了5类13个指标对模型进行了评价与比较。研究表明,各模型组之间的差异较大,以直接回归模型组为最优,PCA线性转换回归模型组次之,DCA非线性转换回归模型组最差。在每一组内,模型之间的差异相对较小,以变量全部入选模型稍优于变量逐步筛选模型。6种模型中,通用多元线性回归模型的拟合性最好、预测精度最高,但模型结构最为复杂、需要的环境因子最多;多元线性逐步回归模型不仅拟合性和无偏性方面很好,而且结构最为简单、需要的环境变量最少,因而为最优模型

关键词: 黄土丘陵区, 土壤水分, 空间预测, 多元线性回归模型, 模型评价指标

Abstract:

The multiple-linear regression models with more readily observed environmental variables (land use and topography) were developed to spatially predict soil moisture content using six methods and their performances and cost-benefit were evaluated using 13 indices in Danangou catchment (3.5 km2) in the loess area of China. Soil moisture measurements were performed biweekly at five depths in soil profile (0~5 cm, 10~15 cm, 20~25 cm, 40~45 cm and 70~75 cm) on 81 plots from May to September 1999 using time domain reflectometry (TDR). It is indicated that the 13 measured indices almost exhibit the similar conclusions. In terms of fitness, optimum, precision, outlier and cost-benefit, the with-attributes group models, including generalized multiple-linear regression models with environmental attributes (GMLRMs) and stepwise multiple-linear regression models with environmental variables (SMLRMs), were shown to be superior to those multiple-linear regressions models with linear transformation on environmental attributes by principal component analysis (PCA-based group models) and those regression models with nonlinear transformation by detrended correspondence analysis (DCA-based group models). Within each group models, the models using generalized-method or enter-method are better than those using stepwise-method are. However, such within-group differences are not so evident as that of inter-group. Among the six methods, the GMLRMs are the best in terms of fitness, optimum, precision and outlier based on the 11 performance indices, while the SMLRMs are most effective and economical according to the Akaike information criterion (AIC) and Schwarz or Bayesian information criterion (SIC) that can evaluate the cost-benefit of models.

Key words: hilly loess, soil moisture content, spatial prediction, multiple-linear regression models, model-evaluation index

  • F301.24