Acquisition And Verification Of Soil-environment Relationships Based On Conventional Soil Map

Conventional soil maps are products of experts’ soil survey and aerial interpretation. They are major data sources for soil spatial distribution, which is essential in watershed management and eco-hydrology. With the development of geographic information technique, traditional soil survey methods are unable to meet the requirements of soil information services. The reasons are listed as follows. First, traditional soil survey methods are often difficult to express, exchange and storage the information since they are derived from experts’ personal knowledge. Second, soil qualitative characteristics, described in those methods, are not corresponding to soil’s spatial distribution, which obviously leads to low accuracy. At last, the traditional methods ask high cost and are limited to one specific region, which toughens the updating process. Therefore, it is essential to retrieve the high-accurate soil map from historical resources and data in Digital Soil Mapping(DSM). This study extracted data of soil type and environment factors from the conventional soil map and terrain data, and then established the soil environment relationship by spatial data mining method, finally, verified it’s reliability and accuracy by field sampling. The proposed method was applied in Nieshui river basin in Huajiahe town, Hongan County, Huanggang city, Hubei province. Based on conventional soil maps obtained from Second National Soil Survey in the case study area, the proposed method was carried out. Major steps of the proposed method were represented as follows:1) Established the geographic information system(GIS) database, which contained seven environment attributions. These seven factors were closely related to the process of soil pedogenesis, including parent material and terrain factors(elevation, slope, aspect, plan curvature, profile curvature and topographic wetness index) extracted from 10 m resolution Digital Elevation Model(DEM).2) Extracted 1410 typical sample data of soil type and environment factors by using the principle of frequency distribution. This process was named as data pre-processing. By doing so, noises and abnormal data in traditional natural resource maps, caused by manual surveys and no reflection of experts’ knowledge, could be significantly reduced.3) Retrieved experts’ knowledge embedded in the map product by using spatial data mining techniques to deal with the typical sample set. Compared with other classification algorithms, the decision tree algorithm was suitable for extracting and representing soil-environment knowledge. Hence, a decision tree algorithm called See5.0 decision tree method was selected to obtain the knowledge of soil environment as a spatial data mining method.4) Predicted soil spatial distribution in Soil-Land Inference Model(So LIM) according to extracted knowledge in step 3) and environment data layers in step 1). Numerous studies had demonstrated that So LIM was a more accurate one than the traditional manual and subjective methods in soil mapping. So LIM used similarity representation to address parameter generalization. The similarity representation for each soil pixel in the parameter domain was calculated by fuzzy logic, which called fuzzy membership degrees. Then, the soil class with the largest fuzzy membership degree was assigned to the soil pixel as its unique soil class. The final spatial distribution map of soil type was Get.5) After contrasting and analyzing the differences between the conventional soil map and updated soil map, the author verified the proposed method’s accuracy through 270 field validation points. These 270 points were collected by three sampling strategies: regular sampling, subjective sampling, and transect sampling. Take the index in the confusion matrix as evaluation standard, which used to verify the accuracy of soil map.The results showed that the soil map retrieved by the fuzzy inference provided more detailed and accurate information in soil spatial distribution than that of the original soil map. The verified accuracies of updated soil map from three sampling mode were 73%, 76% and 71% respectively, they were all higher than the conventional soil map, and the overall classification accuracy of updated soil map was 75%, increased by 11% compared with the conventional soil map, and at the same time the number of pattern spot increased significantly. It is therefore concluded that the proposed method which obtained soil environment relationships from traditional soil map is more accurate in judgment of soil types and determination of the boundary and it facilitates the process of updating the soil map.