数字土壤制图的推理方法对比研究

Inference Methods of Digital Soil Mapping

数字土壤制图是基于计算机技术利用现代空间分析方法获取详细空间属性的土壤制图技术,是当下信息时代对土壤资源类型、数量以及空间分布进行详尽认识的新方法,而推理方法决定了制图效率和结果的可靠性。不同的研究区域必须与推理方法相互匹配,为研究适合小尺度空间范围上的土壤制图方法,本研究选取母质类型、地形因子及遥感光谱指数,在原始土壤图基础上采用面积加权法布设采样点,选用决策树、支持向量机、随机森林三种推理方法获取土壤—环境知识,从而获得研究区的土壤类型空间分布图,并通过实地采样点数据验证比较三种方法的制图精度。研究结果表明:(1)总体上,每种算法得到的土壤类型空间分布结果与原始土壤图高度相似,但相较于原始土壤图,推理得出的土壤图细节信息更丰富;(2)实地采样点验证结果显示,随机森林分类模型的总体分类精度与Kappa系数均优于决策树与支持向量机分类模型,分类结果最佳;且对比三种分类算法推理得到的各土壤类型的用户精度和生产精度,随机森林算法也较决策树与支持向量机两种算法更优。本研究结果可为数字土壤制图中推理方法的选取提供参考。

Digital soil mapping is a soil mapping technology that uses modern spatial analysis methods and computer technology to obtain detailed spatial attributes.It is a new recognition of the quality,quantity,and spatial distribution of soil resources in the information age.And inference methods determine the efficiency of mapping and the reliability of results.Different research areas have to match their reasoning methods.To find out suitable soil mapping method at a small spatial scale,parent material type,terrain factor and remote sensing spectral index were selected,and sampling points were arranged with the area weighted method based on the original soil map.Three data mining algorithms,including support vector machines,decision tree,the random forest,were used to obtain the information of soil and environment and the spatial distribution of soil types in the study area.And these data were verified with field sampling points and the accuracies of soil mapping were compared among the three methods.The results showed that:(1)The spatial distribution of soil types under various algorithms was highly similar with the original soil map on the whole,and the inferred soil maps showed more detailed information than the original soil map.(2)The random forest model showed the superior overall classification accuracy,Kappa coefficient and the best classification results compared with support vector machine and decision tree models.The user accuracy and production accuracy of different soil types by using the random forest algorithm were better than those by using the support vector machine and decision tree algorithm.The results of this study can provide a reference for the selection of digital soil mapping methods.