摘要
土壤质地类型是陆面过程模型、水文模型和大气模型的重要输入参数。在黑河流域中游地区选取一研究区,利用200个实测土壤剖面样点资料及13类环境因子组合,分别根据基于支持向量机、决策树和模糊逻辑的理论方法进行土壤质地预测制图。结果表明:支持向量机方法对训练点和验证点的测试精度都偏低,分别为90%和55%;基于决策树方法对训练点的验证精度最高,达到98%,但其对验证点的预测准确率为57%;基于模糊逻辑方法的综合评价精度最好,对验证点的预测精度达到64%,对训练样点的验证精度为74%。从制图结构来说,支持向量机方法弱化了类型环境因子的预测性能,减小了连续环境因子分布的空间差异性,生成的土壤质地图结构相对简单,精度偏低;基于决策树方法对未知环境因子组合预测具有不确定性,生成土壤图斑块相对破碎;模糊逻辑方法可弥补决策树方法的一些不足,生成的土壤质地图既较好地保留了斑块结构的完整性,也较好描述了土壤类型与典型环境因子之间的相关性。因此针对研究区数据特点,采用基于决策树和模糊逻辑相结合的土壤制图方法更有优势。
Soil texture is a key input parameter for the land surface process models,hydrological models and atmospheric models.Many digital soil mapping methods based on the soil-landscape model concept have been widely studied and applied.The general methods include the decision tree algorithm,the support vector machine approach and the fuzzy logic theory.In this paper,a study was conducted to compare the above three different soil mapping methods in an area of Zhangye of Heihe River Basin by integrated 200 ground measured soil samples and 13 types of environmental factors.Meanwhile,different soil texture maps based on the three methods were predicted in the study area respectively.Results show that：the support vector machine model gets relatively low accuracies both for test soil samples and training samples,which are 90% and 55% respectively.The decision tree model gets the highest accuracy of 98% for training datasets among the three methods,but its accuracy for testing data decreases into 57%.The fuzzy logic model gets the highest accuracy of 64% for testing data and a compromise accuracy of 74% for training data.As for the structure characteristics of the texture soil maps,the study finds that the support vector machine model predicts a much simplified soil texture map as it may weak the prediction ability of thematic and continuous environmental factors.The decision tree model often gets unstable predictions of unexpected combinations of environmental factors,resulting in a relatively fragile structure of the soil texture map.The fuzzy logic model predicts the most reasonable soil texture map among them,because it can not only keep the soil texture structure much holistic,but also can illustrate an appropriate relationship between different soil types and different environmental factors.It is suggested that combining the decision tree algorithm with the fuzzy logic theory might be an appropriate method to map the soil texture distribution of Heihe River Basin.
出处
《遥感技术与应用》
CSCD
北大核心
2011年第2期177-185,共9页
Remote Sensing Technology and Application
基金
国家自然科学面上基金项目(40871190)
国家自然科学青年基金项目(40901160)资助
关键词
土壤质地制图
决策树
支持向量机
模糊逻辑
黑河流域
Soil texture mapping
Decision tree
Support vector machine
Fuzzy logic
Heihe River Basin
