[Objective]Digital soil mapping is based on Jenny's classic theory,of which the core is that soil is the yield of the interactions among numerous soil forming factors. To establish relationships between environmental variables and soil attributes,a soil-landscape model is built up and used to predict soil types or attributes. Although this model has been extensively used in digital soil mapping,investigations are still on the way on what the relationships between environmental variables varying with the region and soil attributes are. Therefore,how to extract rapidly and accurately knowledge of soil environment has become the key to the current researches in this field. Knowledge acquisition based on sampling points is often affected by the number of sampling points,errors in sampling processes and representativeness of sampling points. Traditionally,soil mapping is based on manual soil surveys and tends to have errors in the following two fields,i.e. enclosure and displacements of boundaries. The knowledge acquired from soil maps cannot be used to predict local soil conditions,and especially lacks details specific to soil grade and issues the users are interested in. To solve these problems and attain soil environment information high in accuracy,a method based on knowledge intercomplementation and fusion is set forth.[Method]This paper uses decision tree coupled with a uncertainty model to extract soil environment information. Through limiting threshold values,the decision tree model can be used to predict soil type rapidly and efficiently. With the soil lowering in type level,the prediction lowers steadily in accuracy. Therefore,ignored uncertainty and exaggerated uncertainty can be used to judge truly and efficiently accuracy of the inferred map to a certain extend and hence to realize accurate quantitative evaluation of the inferred map. So coupling of the two models can not only save money and time,but also raise efficiency and realize scientific re-extraction and fusion of soil information. The NieshuiRiverBasin in Huajiahe Town,Hongan County,Huanggang City of Hubei Province was selected as a case for study. The proposed method proceeded in three steps. 1)By means of the standard See 5 algorithm,decision trees were constructed and used to extract soil-environment information and hence spatial distribution for soil mapping;2)With the aid of the SoLIM software,a spatial distribution map of exaggerated and ignored uncertainties was plotted. The two kinds of uncertainties appeared in the processes of classifying a geographic entity,i.e. ignored uncertainty,which is attributed to the similarity of the studied soil to all the soil types,and exaggerated uncertainty,which is associated with the deviation of the studied soil from the prototype specified in the processes of soil hardening. A similarity model can be used to estimate the two uncertainties;and 3)Soil samples were collected again based on the spatial distribution map of the uncertainties. The higher the values of exaggerated and ignored uncertainties,the higher the probability of a soil beingmis-classified.So resampling should be performed in location low in uncertainty value. The soil environment information acquired from the resampled soil sample set combined and updated or optimized the original knowledge. In the end,soil mapping by inferring was performed with the aid of the SoLIM software and based on the eventually obtained soil-environment knowledge,and validated with the validation set of 253 sampling sites in the field for accuracy.[Result]Results show that the soil map plotted through inferring contains more specific spatial distribution information and reaches up to 86.9% in accuracy as validated with the field validation sites. Obviously it is 13% higher than the soil maps so far available. Moreover,its Kappa coefficient is 0.842,higher than 0.8,indicating its high degree of consistency,and conformity with the attributes and spatial distribution of the soils in the study area. [Conclusion]It is,therefore,concluded that the proposed method which acquire the knowledge of soil-environment relationship using the uncertainty models feasible and effective.This method not only increases spatial detailedness of the soil map,but also improves its accuracy.%土壤与环境关系知识的获取是精细数字土壤制图的关键,如何快速准确地获取该知识成为现阶段研究的重点.以湖北省黄冈市红安县华家河镇为例,利用土壤—环境推理模型(Soil-Land Inference Model,SoLIM)得到土壤类型的夸大和忽略不确定性分布图,依据不确定性分布图在可信度高的位置重新采集样点,对样点进行数据挖掘,获取环境因子组合,建立其与土壤类型的对应关系.结合原始规则,生成新的土壤—环境关系知识,并将其用于土壤推理制图,获得新的土壤类型分布,利用253个野外独立样点进行精度验证.结果表明:推理土壤图显示了更加详细的空间分布信息,经野外验证点验证,总体精度为86.9%,高于原土壤图精度约13%.因此,利用不确定性模型重新获取土壤—环境关系知识的方法是可行且有效的,该方法不仅增加了土壤图的空间详细度,而且提高了土壤图的精确度.
展开▼
