基于支持向量机的典型冻土区土壤制图研究  

SUPPORT VECTOR MACHINE BASED SOIL MAPPING OF A TYPICAL PERMAFROST AREA IN THE QINGHAI-TIBET PLATEAU

作　　者：石伟[1] 南卓铜[1] 李韧[1] 赵林[1] 张秀敏[1] 赵拥华[1] 

机构地区：[1]中国科学院寒区旱区环境与工程研究所,兰州730000

出　　处：《土壤学报》2011年第3期461-469,共9页Acta Pedologica Sinica

基　　金：科技部基础性工作专项(2008FY110200);科技部863项目(No.2008AA12Z205);国家自然科学基金项目(40871037)资助通

摘　　要：基于青藏高原大片连续多年冻土分布的东部边缘,青海省兴海县温泉地区的野外调查数据,通过对研究区遥感数据的分析,开展了土壤制图方法的探讨。以成土因素学说和土壤-景观模型理论为基础,筛选土壤分类潜在变量,在不同的变量组合下运用支持向量机(SVM)的方法建立土壤-景观模型,对整个研究区进行预测性分类。为了更好地检验该方法的有效性,采用五折交叉方式进行结果的验证。并通过对比不同变量组合的交叉验证结果和分布模拟结果图,确定了适合典型冻土区土壤分类的环境变量组合,以较少的样本知识较好地预测该区土壤类型的空间分布。Based on field investigations conducted in the 1980s of Wenquan District,Xinghai County,Qinhai Province,a large tract of permafrost on the east edge of the vast Qinghai-Tibetan Plateau（QTP）,new approaches to soil mapping were explored.Based on the theory of soil forming factors and the theory for soil-landscape modeling and the screening of potential variables in soil classification,a soil-landscape model was built up using SVM coupled with various combinations of the variables,and applied to predictive soil classification of the studied area.To better verify effectiveness of the new approach,a 5-fold cross validation method was used.By comparing the outcomes of the cross validation of the various variable combinations as well as the derived distribution maps,a set of combination of environmental variables was defined suitable for soil classification of typical permafrost area.Thus fewer samples are needed to better predict spatial distribution of the types of soils in the area.

关 键 词：土壤制图 支持向量机 多年冻土区 土壤分类 

分 类 号：S159.9[农业科学—土壤学] O159[农业科学—农业基础科学]