应用Fisher判别分析和案例推理两种方法的土壤类型预测及制图比较  

Comparison between Fisher discriminant analysis and case-based reasoning for soil type prediction and mapping

作　　者：邱琳[1] 张富[2,3,4] 李安波[4] 

机构地区：[1]江苏省农业科学院农业经济与信息研究所,江苏南京210014 [2]华北水利水电学院资源与环境学院,河南郑州450011 [3]苏州工业园区博士后科研工作站数字城市站,江苏苏州215021 [4]南京师范大学虚拟地理环境教育部重点实验室,江苏南京210046

出　　处：《江苏农业学报》2012年第6期1459-1465,共7页Jiangsu Journal of Agricultural Sciences

基　　金：江苏省农业科技自主创新基金项目[SCX(11)4030];江苏省"三项三新"工程项目[SXGC(2012)410];国家自然科学基金项目(41101372);华北水利水电学院高层次人才科研启动项目(001321);虚拟地理环境教育部重点实验室开放基金项目;科技部中小企业技术创新基金项目(11C26213201271);中国博士后科学基金项目(201150M1511)

摘　　要：土壤类型预测目前没有公认、成熟的模型和方法,原因是缺乏在同一地区不同模型之间的比较研究。该文利用已知类型的土壤样点及其所处位置的高程、坡度、平面曲率、剖面曲率、复合地形指数等数据,分别采用Fisher判别分析和案例推理两种方法对安徽宣城样区进行土壤类型预测和制图表达。结果表明,在土纲级别两种方法均能够较好地预测土壤类型,但由于土壤样点的剖面数量一定,随着从土类到土族级别的降低,两种方法预测土壤类型的准确率也逐步降低。但各个级别的土壤类型预测中,案例推理的预测能力要优于Fisher判别分析方法。Nowadays soil map has been the data bottleneck for ecological modeling,watershed simulation,precision agriculture,environmental monitoring,etc,so soil data of high spatial and temporal resolution is needed urgently.Digital soil mapping（DSM） can be a way to solve the soil data crisis.DSM focuses on soil prediction model（property or type）.Scientists have presented many prediction models of soil type,but there has no uniform and acknowledged one because of lacking of comparison among different prediction models in the same region.So this study aimed to compare the predicting and mapping effectiveness of Fisher discriminant analysis（FDA） and case-based reasoning（CBR） on soil type based on some environmental data in the same area,Xuancheng city,Anhui province,with scope being 118°37′31″E—118°40′15″ E,30°50′55″N—30°52′30″ N,and area being 11.3 km2.There were 79 soil profiles,containing 48 calibration profiles and 31 validation profiles,respectively,being dug and described.Soil samples from each profile were also collected for analyzing soil property,such as pH value,organic matter,cation exchange capacity,mechanical composition,exchangeable H＋,exchangeable Al3＋,free iron oxide,etc.The soils were classified into 3 soil orders,3 soil suborders,7 soil groups,9 soil subgroups,etc.Since the variations of climate,vegetation and parent material were not obvious in the study area,terrain can be considered as determining factor of soil type identification,and environmental data which contains elevation,slope gradient,profile curvature,plain curvature and compound terrain index,were utilized to predict soil types.The key point of FDA was the establishment of discriminant functions,and that of CBR was the quality of case base.The statistics of validation profiles showed that the prediction accuracy of soil order was 84.2% by FDA and 92.7% by CBR respectively.The same results were seen on soil suborder.The prediction accuracy of soil group was 23.7% by FDA and 61.3% by CBR respectively,a

关 键 词：数字土壤制图 土壤类型预测 FISHER判别分析 案例推理 

分 类 号：TP79[自动化与计算机技术—检测技术与自动化装置]